WSPRadar.org

HAM RADIO STATION & ANTENNA BENCHMARKING

0. Why WSPRadar?

Radio amateurs continually modify and improve their stations. A new antenna goes up, its height or orientation changes, a feedline is replaced, a common-mode choke is reworked, or a receiver, filter or preamplifier is added. The same question follows almost automatically: did the change actually improve the station — and if so, where, when and by how much?

On the air, this can initially seem easy to answer. More contacts are completed, a remote operator gives a better signal report, a WebSDR shows a stronger signal or WSPR produces more spots. Such observations are valuable, but they do not measure the changed component alone. The result always arises from the complete station interacting with the radio path: antenna, feedline, radio, transmit power, receiver, local noise, interference, terrain, ionosphere, remote station and time all contribute at once.

This is the fundamental measurement problem. A better report may reflect a favorable moment of propagation. An additional contact may involve a different remote station. A higher spot count may come from changing network activity or better conditions. Even completely accurate observations therefore do not automatically reveal the cause.

Experienced radio amateurs address this problem with increasingly controlled methods: repeated comparisons, beacon transmissions, WebSDRs, Reverse Beacon Network data, WSPR and especially rapid live A/B switching. A fast A/B test is far more informative than two contacts made hours apart because the transmitter, power, frequency, remote station and much of the radio path remain similar. Established WSPR comparison experiments likewise show that common conditions and the shortest practical — or simultaneous — comparisons are more robust than long separated measurement blocks [Ref-1] [Ref-2] [Ref-3] [Ref-4] [Ref-5].

Yet even a careful rapid A/B comparison normally observes one radio path during one short interval. QSB, multipath propagation, QRM and local noise can change during the switch. AGC, S-meter resolution, unequal signal chains and subjective reports add further uncertainty. An observed advantage may be real, but at first it applies only to that station, direction, time and propagation state.

The real challenge is therefore not merely to observe a difference. It is to determine whether the difference repeats under many comparable conditions, how large it typically is, on which radio paths it appears, when it recurs and how much evidence supports it.

This is where WSPR provides an unusually powerful foundation. Its repeated, time-stamped and machine-decoded low-power transmissions create observations across many stations, distances, directions and propagation states in a worldwide volunteer network [Ref-6] [Ref-7] [Ref-8]. Depending on band activity and the observation period, hundreds to thousands of reports can accumulate over hours or days.

WSPRadar turns that stream of reports into an experimental evidence system. It brings comparable observations together, checks whether relevant stations were demonstrably active, accounts for reported transmit power where appropriate, prevents a few prolific stations from silently dominating station-balanced summaries, and keeps every result traceable to the contributing stations and observations. The activity check follows an important observational principle: silence should not become counter-evidence until operation is independently observable [Ref-9].

The result is more than a spot count and more than a single winner-versus-loser number. WSPRadar can show whether a pattern is broad or path-specific, whether it is associated with distance or direction, whether it appears once or recurs by time of day, whether many stations agree, and whether the paired evidence represents the wider result. It helps move station experimentation from “this looked better once” toward “this difference repeatedly appeared here, under these conditions, with this much support.”

WSPRadar is not a calibrated antenna range, and it does not turn public WSPR reports into laboratory measurements. It provides a practical bridge between everyday station experimentation and amateur science: semi-quantitative, geographically rich, time-aware and auditable evidence about complete stations and controlled signal paths under real operating conditions.

Used this way, WSPR becomes more valuable to the wider amateur community as well. Accurate callsigns, locators and power reports, stable operation and documented changes turn routine beaconing into evidence that can be revisited, compared and learned from rather than merely watched on a map.

0.0 WSPR in 2 Minutes

WSPR stands for Weak Signal Propagation Reporter. Joe Taylor, K1JT, and Bruce Walker, W1BW, described it as a worldwide network of low-power stations exchanging beacon-like transmissions to explore possible propagation paths. A WSPR-2 transmission lasts just under two minutes and occupies only about 6 Hz. Its message normally contains a callsign, a four-character Maidenhead locator and reported transmit power in dBm. Decoder-reported signal-to-noise ratio (SNR) is referenced to a 2500 Hz bandwidth, and successful decodes are possible at approximately -28 dB; a less negative SNR means a stronger signal relative to the receiver noise [Ref-6] [Ref-8].

When reporting is enabled, a receiver uploads each successful decode as a spot. A spot records the transmitter and receiver identities, their reported locations, time, band, transmit power and decoder-reported SNR. Public archives consequently contain a large, continuously growing record of successful radio observations contributed by independently operated stations around the world. Services such as wspr.live and WSPRDaemon preserve and expose this observational record for analysis [Ref-10] [Ref-11].

Data sources. WSPRadar uses wspr.live as its primary data source. The WSPRadar project is grateful to the people behind wspr.live and WSPRDaemon who make this public database infrastructure available and keep it running.

Under concurrent load, WSPRadar can route a complete new run to WSPRDaemon WD2 and then WD1, as capacity permits. This ordered capacity spillover is distinct from provider failover: if a selected source fails, WSPRadar discards the unpublished attempt and restarts the complete run on the next source. Every completed run remains pinned to one archive; records from different sources are never combined.

One limitation is central to every analysis: the archive records successful decodes, not a complete log of every attempted transmission or every active receiver. WSPRadar therefore constructs an opportunity only when independent evidence shows that the relevant remote transmitter or receiver was active. For RX, another eligible receiver must have decoded the same transmitter. For TX, the remote receiver must have decoded another signal on the same band. Without such supporting activity, a missing Target spot is not automatically treated as a radio failure.

This distinction turns WSPR from a collection of successful spots into evidence that can support questions about practical reach, consistency and relative performance without pretending that every missing report represents a failed radio path.

0.1 What WSPRadar can show

WSPRadar is a WSPR-based antenna and station performance analysis and benchmarking system. It evaluates one Target: the station under test, normally your station, represented either as a complete installed station or as a controlled transmit or receive path. A peer is a remote station whose radio path contributes to the analysis. Decode Rate is the percentage of independently confirmed opportunities in which the Target-side decode succeeded: the Target decoded the peer in RX, or the peer decoded the Target in TX. WSPRadar then answers one of two broad questions.

Performance asks how the Target behaved across independently confirmed WSPR opportunities. It can reveal practical footprint, at-least-once reach, Decode Rate, successful signal levels, distance and direction structure, temporal behavior, and the breadth and depth of the supporting evidence.
Benchmark asks how the Target behaved relative to a meaningful Reference under matched conditions. It can reveal paired Target-minus-Reference Delta SNR, joint and one-sided Decode Outcomes, how much evidence was pairable, and where and when the relative difference appeared.

The question determines the appropriate evidence design:

Analysis	Question	Practical examples
RX Performance	How broadly and consistently does my receiver decode signals that were independently confirmed elsewhere?	Establish the receiving footprint of a newly commissioned antenna or station; see whether reception is broad but intermittent or narrower and consistent; identify recurring direction, distance or UTC-hour patterns, including periods that may warrant a separate check for local noise or intermittent hardware.
TX Performance	Where, when and how consistently is my transmitter decoded by receivers independently shown to be active?	Map where a QRP beacon or newly installed antenna is heard; see when and in which directions independently confirmed active receivers decode the station most consistently; establish a station baseline after commissioning, repair or relocation, and use comparable repeat runs to determine whether its observed behavior later changes.
RX Benchmark Hardware A/B	Did two local receive paths differ while observing the same remote transmissions?	Compare two antennas, each feeding its own simultaneous receiver and decoder chain, as complete receive paths; attribute a difference specifically to the antennas only when the remaining chains are matched, characterized or confirmed by crossover; feed one antenna through a characterized splitter into two receivers to compare receiver or decoder paths; place a preamplifier, filter, feedline or common-mode choke in only one otherwise controlled path and benchmark the two documented complete receive paths.
TX Benchmark Hardware A/B	Did two local transmit paths differ under simultaneous or tightly scheduled operation?	Feed two antennas from separate calibrated transmit chains and transmit simultaneously with synchronized cycles, distinguishable signals and adequate isolation; use one transmitter and a controlled RF switch to alternate between two antennas on a fixed UTC schedule; compare two feedlines, matching networks, filters or complete transmit paths while controlling actual power, timing and the remaining chain.
RX/TX Benchmark Reference Station / Buddy Test	How does my complete station compare with one known station?	RX: compare your receiver with a known Buddy receiver while both observe the same remote transmitters in the same cycles; TX: compare your transmitter with a Buddy transmitter at the same remote receivers in the same cycles; repeat a stable, well-understood Buddy design as a relative whole-station baseline before and after documented station work, without treating the Buddy as an absolute calibrated standard.
RX/TX Benchmark Local Median Neighborhood	How does my station compare with the typical active WSPR group nearby?	See whether your receive or transmit station is broadly above, near or below the cycle- and path-specific median of active local peers inside the selected radius; commission a station when no single suitable Buddy Reference is available; identify directions, distances or UTC periods where the station departs from that contextual local baseline, while checking neighborhood membership and radius sensitivity.
RX/TX Benchmark Local Best Station	How does my station compare with the strongest active nearby peer available on each path and cycle?	Compare your station with the strongest qualifying nearby station available on each path and cycle; find directions or distance ranges where your station approaches or trails the changing local-best envelope; in comparable repeat runs, track whether the observed gap narrows or widens while checking radius and pool membership, without treating the result as a ranking against one permanent competitor or a stable calibrated baseline.

The Reference is part of the scientific question, not just a display choice. A controlled Hardware A/B Test provides the strongest basis for attributing an observed difference to local paths or components, but only to the extent that the remaining chains are controlled. A Reference Station / Buddy Test compares two complete installed stations, including their QTHs, equipment, terrain and local noise environments. Neighborhood benchmarks provide changing contextual baselines rather than fixed or calibrated standards.

These perspectives make WSPRadar useful for more than formal antenna comparisons. Performance can establish a station baseline, show where a station is dependably heard, reveal directional or distance-dependent behavior, identify recurring daily patterns, and help locate when an intermittent change appeared. Benchmark can compare antennas, feedlines, filters, preamplifiers, receivers or complete paths; contrast two complete stations; or place one station in the context of its active local neighborhood.

WSPRadar can identify the shape, scope and timing of an observation. It can show that a difference is broad, concentrated, intermittent, recurring or supported by only a narrow subset of stations. It cannot by itself determine that the cause was antenna gain, radiation efficiency, take-off angle, calibrated receiver sensitivity, local noise or one particular hardware component. No later statistic can remove a variable that the physical experiment did not control.

0.2 What one run produces

A WSPRadar run produces a connected evidence package for one clearly bounded station question — not a universal score and not a leaderboard.

A Performance run brings together practical reach, two complementary Decode Rate weightings, successful Target SNR, distance and direction structure, chronological change, recurring UTC-hour behavior, contributing stations and the underlying opportunities. A Benchmark run combines paired Delta SNR with Decode Outcomes and evidence coverage so that a favorable paired median cannot hide extensive one-sided evidence or a narrow pairable subset.

Every result follows the same evidence path:

Map → Segment Inspector → Performance/Benchmark Evidence → Temporal Evidence → Station Insights → Selected Station Evidence → Drill-Down

The map provides the geographic overview. Segment-level evidence shows how the observation changes with distance and direction and how much support lies behind it. Performance or Benchmark Evidence separates the main result from its complementary evidence. Temporal Evidence shows whether the pattern changed during the run or recurred at particular UTC hours. Station Insights reveals which station identities contribute. Selected Station Evidence follows one exact radio path, and Drill-Down exposes the observations, same-cycle comparisons or scheduled A/B pairs behind the summaries.

This layered structure is one of WSPRadar's central strengths: the high-level pattern remains connected to its evidence. An operator can move from where the effect appears, through how consistently it appears and how well it is supported, down to the individual observations from which the conclusion was built.

A credible result is therefore not simply the largest value on the screen. It is one in which the experiment design, geographic pattern, temporal behavior, station breadth, evidence depth and row-level audit support the same bounded interpretation. Repeating the design in another suitable operating window can then test whether the observation is experimentally repeatable rather than only internally consistent within one run.

The complete run can also be preserved as a reproducibility package containing its analysis definition, processed evidence, tables, figures and metadata, ready to be reviewed later or shared with another operator alongside the physical station notes that WSPRadar cannot infer.

0.3 Your first useful run

The quickest way to understand WSPRadar is to begin with a maintained demo. A demo presents a complete historical Performance or Benchmark analysis with a prepared experimental context, allowing the evidence path to be explored before your own station is involved.

The value of the demo becomes clear in the connection between its layers: the geographic overview, distance and direction, isolated versus recurring temporal behavior, the number and diversity of supporting stations, the pairability of Benchmark evidence, and the selected-path and row-level observations behind the summary.

A demo is a worked example of WSPRadar's method, not evidence about your own station. Once the evidence path is familiar, the most useful first analysis of your own station begins with one clear question: establish its RX or TX Performance baseline, compare two controlled local paths, benchmark against a known station, or place it in its local WSPR context.

The aim is not to produce a flattering number. It is to obtain a result you can understand, question, repeat and use to make a better-informed decision about the station.

Part I: Operator Guide

This part takes you from an operating question to an evidence-matched conclusion. Chapter 1 establishes the common experiment, selects RX or TX and Performance or Benchmark, and introduces the shared evidence path. Chapter 2 then follows that path within the exact analysis family and Reference design and closes with an optional expert diagnostic for temporary Delta SNR departures. Chapter 3 explains how to strengthen, report and preserve the result. Exact controls remain in Part II; exact calculations and scientific edge cases remain in Part III.

In this guide, the experiment is the physical on-air operation and station configuration. A run or analysis is WSPRadar's configured processing of the resulting observations. A result is the Performance or Benchmark evidence produced by that run.

1. Choose and Prepare the Analysis

Start with the station question and the physical experiment. The interface choice follows from that question; it does not define it.

1.1 Build a strong experiment foundation

A useful WSPRadar result begins with one sentence stating what is being tested and what observation would count as support. Decide whether the run is exploratory — intended to find a possible pattern — or confirmatory — intended to test a pattern already identified.

Use one exact band and a UTC window in which the Target was operating. Enter callsigns exactly as uploaded and verify the Target QTH. Record the antenna, feedline, radio, tuner, gain or power settings, decoder, software version, schedule and any deliberate change. Keep every variable outside the question as stable as practical.

For TX, keep actual and reported power accurate and stable unless power is the tested variable. For RX, keep gain, filtering, audio routing, decoder settings and upload behavior stable unless one of them is under test. Keep clocks synchronized. In Benchmark, verify that the Reference was operating as intended: the Target-Active Gate establishes observable Target participation but does not prove Reference uptime.

Before a confirmatory repetition, fix the direction, band, Reference design, filters, thresholds, schedule and primary geographic or temporal scope. Treat alternative radii, time windows or scopes as separate sensitivity analyses rather than choosing only the version that looks most favorable.

1.2 Choose the analysis that matches the question

Operating question	Analysis
Which independently confirmed signals does my receiver decode, where, when and how consistently?	RX Performance
Where and how consistently is my transmitter decoded by receivers independently shown to be active?	TX Performance
How do two local receive paths, two complete receiving stations, or my receiver and a local neighborhood Reference differ?	RX Benchmark
How do two local transmit paths, two complete transmitting stations, or my transmitter and a local neighborhood Reference differ?	TX Benchmark

Choose Performance when the Target itself is the question and no Reference is required. Performance combines at-least-once reach, Decode Rate, successful Target SNR, geography, time and evidence support. It describes the complete Target station under the selected real-world conditions.

Choose Benchmark when the question is explicitly relative to a Reference. The Reference determines the meaning of the result:

Hardware A/B is the strongest design for a local component or path question, but it isolates that component only to the extent that the remaining paths are controlled.
Reference Station / Buddy Test compares complete installed stations and their operating environments.

Local Median Neighborhood compares the Target with a changing typical local baseline inside the selected radius.

Local Best Station compares the Target with a changing strongest local peer on each qualifying path and cycle.

Use the narrowest Reference design that supports the intended claim. A complete-station or neighborhood benchmark cannot be turned into isolated antenna gain by later filtering or averaging.

1.3 Follow the evidence path

Every completed result follows the same operator path:

Map → Segment Inspector → Performance/Benchmark Evidence → Temporal Evidence → Station Insights → Selected Station Evidence → Drill-Down

Map. Locate the broad distance and direction pattern. Read sector color together with the station and opportunity, spot or pair support. A colored sector is a prompt to inspect, not the conclusion.

Segment Inspector. Select the geographic scope relevant to the question. Every following evidence view uses that active scope, allowing a broad map pattern to be separated into distance- and direction-dependent behavior.

Performance or Benchmark Evidence. In Performance, combine reach, both Decode Rate weightings and successful Target SNR. In Benchmark, combine station-balanced and observation-level Delta SNR, Decode Outcomes and Joint Evidence Share. These quantities answer different questions and should not be collapsed into one score.

Temporal Evidence. Use the chronological view to see when behavior changed during the run and the UTC-hour view to see whether a time-of-day pattern recurred across dates. Read signal-level evidence together with its station, opportunity or pair support.

Station Insights. Check whether the pattern is supported across many callsign + locator identities or is concentrated in a few paths. Read every station-level value together with its evidence counts.

Selected Station Evidence. Inspect one representative, surprising or high-impact path. This shows whether the segment summary also describes that path, whether its behavior is intermittent, and whether its time pattern differs from the wider scope.

Drill-Down. Verify the retained opportunities, same-cycle pairs or scheduled pairs behind a result. Use it to check identities, locator changes, timing, one-sided evidence and isolated outliers.

The rest of Part I applies this same path to each analysis question without inventorying every title, axis or layout detail.

2. Run and Interpret Your Analysis

Use the section matching the selected Direction and result type. Exact control labels, defaults and ranges are in Chapter 4; exact eligibility, matching, weighting and aggregation are in Chapter 7.

2.1 RX Performance

Question answered. Which remote transmitter-cycles, independently confirmed by another eligible receiver, did the Target receiver also decode; how consistently did it do so; what successful SNR did it observe; and where and when did that behavior occur?

Minimum valid setup. Use the exact Target reporting callsign and QTH, one band and a window with observable Target receiver activity. Keep the receive chain stable. Performance does not introduce a Reference and does not isolate one component of the receive system.

What WSPRadar evaluates. A confirmed RX opportunity exists when another eligible receiver decoded the same remote transmitter in the same Target-active cycle. Heard by Target means the Target also decoded it; Heard by others only means the independent receiver did but the Target did not. Evidence without independent confirmation remains auditable but does not enter Decode Rate. The exact classification is in Section 7.4.

Read the evidence path. On the Map, sector color shows the Station-balanced Decode Rate of qualifying remote transmitters in each distance-and-direction segment. Station markers and the footer distinguish paths heard by the Target at least once from paths heard only elsewhere. Use this first to locate broad RX footprint and directional structure, not to judge receiver sensitivity from color alone.

In Segment Inspector, first compare station breadth with confirmed-opportunity depth. Many opportunities from only a few transmitters are deep but narrow evidence; agreement across many transmitters is broader. Then read the three Performance views together:

At-least-once reach asks which qualifying transmitters were heard at least once during the window. It measures breadth and normally increases with a longer run.
Decode Rate asks how consistently the Target decoded confirmed opportunities. The Station-balanced rate gives each transmitter one vote; the Opportunity-level rate gives each confirmed cycle one vote. A difference between them shows that high-volume transmitters behave differently from the wider station population.
Successful Target SNR describes only successful decodes. It helps show whether the successful signals themselves differ with distance, but misses have no Target SNR and cannot appear there.

In Temporal Evidence, successful-SNR deviation compares each transmitter path with its own usual successful level during the run. Values above 0 dB mean successful decodes were stronger than usual for their respective paths; values below 0 dB mean weaker. The accompanying station and opportunity evidence shows whether a signal-level change coincided with changed Decode Rate and whether the pattern was broadly supported. The chronological view identifies changes during the run; the folded UTC-hour view identifies recurring daily behavior.

In Station Insights, read each transmitter's Decode Rate together with Heard by Target and Heard by others only counts. Select a typical path, an outlier and any path contributing unusually large evidence. Selected Station Evidence then shows the actual successful SNR and opportunity history of one transmitter path rather than the station-relative summary across the segment. Drill-Down verifies the contributing cycles and distinguishes confirmed opportunities from Target evidence that lacks independent confirmation.

Common interpretation patterns. Broad reach with high Decode Rate means many paths opened and were decoded consistently. Broad reach with lower Decode Rate means many paths opened at least once but were intermittent. Limited reach with high Decode Rate means fewer qualifying paths opened, but those that did were comparatively consistent. If successful SNR remains steady or rises while Decode Rate falls, weaker signals may have disappeared below the decoder threshold, leaving only stronger successful decodes. A pattern confined to one azimuth, distance range or UTC period can be operationally useful, but it describes the installed receiver under those paths and conditions rather than a context-free sensitivity number.

Boundary and confirmation. RX Performance combines antenna, feedline, receiver, gain, filtering, decoder, local noise, interference and propagation. It does not directly measure receiver sensitivity, antenna gain, absolute noise or propagation mode. Repeat a suspected pattern in another suitable window. When the intended conclusion is about one hardware change, use a controlled RX Benchmark or a crossover rather than relying only on separated before-and-after Performance runs.

Evidence-matched conclusion.

For this Target receiver, band, UTC window and selected transmitter population, RX Performance describes at-least-once reach, Decode Rate within independently confirmed transmitter-cycles, successful-decode SNR and the geographic and temporal scope in which those observations appeared. State the weighting used, the station and opportunity support, and whether the pattern was broad, intermittent, directional, distance-dependent or recurring.

2.2 TX Performance

Question answered. Which remote receivers independently shown to be active decoded the Target transmitter; how consistently did they do so; what successful SNR did they report; and where and when did that behavior occur?

Minimum valid setup. Use the exact Target callsign and QTH, one band and a window in which the Target transmitter was operating. Keep the RF path, schedule and actual power stable, and report power accurately. Performance evaluates the complete transmitted station rather than one isolated component.

What WSPRadar evaluates. A confirmed TX opportunity exists when the remote receiver was active in a Target transmit cycle, demonstrated by another qualifying same-band decode. Target heard means that receiver also decoded the Target; Other signals heard only means it decoded qualifying same-band activity but not the Target. A Target report without independent receiver-activity confirmation remains auditable but does not enter Decode Rate. The exact denominator is in Section 7.4.

Read the evidence path. On the Map, sector color shows the Station-balanced Decode Rate of qualifying active receivers in each distance-and-direction segment. Markers and footer counts distinguish receivers that heard the Target at least once from receivers that heard only other qualifying signals. Use the map to locate practical transmitted footprint and directional structure.

In Segment Inspector, compare receiver breadth with confirmed-opportunity depth, then read the three Performance views together:

At-least-once reach asks which qualifying active receivers heard the Target at least once during the window.
Decode Rate asks how consistently the Target was reported within confirmed receiver opportunities. The Station-balanced and Opportunity-level rates reveal whether frequently reporting receivers behave differently from the wider receiver population.
Successful Target SNR shows normalized SNR for successful Target reports. It is conditional on a decode and depends on the accuracy of reported transmit power.

In Temporal Evidence, successful-SNR deviation shows when successful reports were stronger or weaker than each receiver path's own usual successful level. The accompanying station and opportunity stacks show whether a change in successful SNR was accompanied by a change in practical detectability and how much support each time bin contains. Chronological change and recurring UTC-hour behavior should be distinguished.

In Station Insights, read each receiver's rate with its Target heard and Other signals heard only counts. Selected Station Evidence exposes one receiver path's actual successful SNR and opportunity history, which helps determine whether the segment summary reflects many receivers or masks a path-specific effect. Drill-Down verifies Target reports, independent receiver activity and any Target reports that lack independent receiver-activity confirmation.

Common interpretation patterns. Broad reach and high Decode Rate indicate that many qualifying active receivers heard the Target consistently. Broad reach and lower Decode Rate indicate a large but intermittent footprint. A persistent advantage in one azimuth or distance range can be consistent with installed antenna and terrain behavior; a short isolated improvement can instead reflect propagation or receiver availability. Stable successful SNR with falling Decode Rate can indicate that only stronger surviving reports remain. Differences between station-balanced and Opportunity-level rates reveal whether a few high-volume receivers are driving the pooled view.

Boundary and confirmation. TX Performance combines transmitter, actual power, feedline, matching, antenna, terrain, remote receiver systems, noise and propagation. Reported-power normalization cannot correct an incorrect power report or an unmeasured feedline loss. The result does not directly measure EIRP, efficiency, antenna gain or take-off angle. Repeat the pattern across another suitable window; use TX Benchmark when the question is specifically whether one transmit path differs from another.

Evidence-matched conclusion.

For this Target transmitter, band, UTC window and selected active-receiver population, TX Performance describes at-least-once reach, Decode Rate within independently confirmed receiver-cycles, successful reported SNR and the geographic and temporal scope in which those observations appeared. State the weighting, receiver and opportunity support, reported-power basis and whether the pattern was broad, intermittent, directional, distance-dependent or recurring.

2.3 RX Benchmark

Question answered. How did the Target receive side differ from the selected Reference while both observed the same remote transmitter identities in the same WSPR cycles?

Shared RX Benchmark evidence. Paired Delta SNR is formed only where Target and Reference produced comparable same-transmitter, same-cycle evidence. Positive Delta SNR favors the Target; negative favors the Reference. Decode Outcomes preserve Joint, Only Target, Only Reference and asynchronous evidence around that paired subset. Joint Evidence Share describes how much retained evidence was pairable; it is not a win rate. Reference uptime must be understood independently, and the Target-Active Gate makes one-sided categories intentionally asymmetric.

Read the evidence path. On the Map, sector color summarizes the station-balanced median Delta SNR of the remote transmitters in each distance-and-direction segment. Marker categories show whether each transmitter identity contributed Joint or one-sided evidence. Read color with the station and spot counts: a strong-looking sector supported by only a few transmitters is narrower evidence than a similar result repeated across many paths.

In Segment Inspector, Decode Outcomes show two complementary compositions: station breadth and observation volume. Station Medians give each remote transmitter one Delta-SNR value and therefore equal weight; the Joint-Spot distribution shows every paired observation and can be influenced by high-volume transmitters. Agreement between them supports a broad shift, while disagreement shows that observation volume and station breadth tell different stories.

Temporal Evidence shows whether paired Delta SNR changed during the run or recurred by UTC hour. Read it with Benchmark Evidence Coverage: a Delta-SNR pattern supported by broad Joint coverage is different from a pattern visible only in a thin paired subset. One-sided evidence can reveal practical near-threshold differences that paired Delta SNR alone cannot describe, but it does not supply a missing-side SNR.

In Station Insights, inspect each transmitter's Joint and one-sided counts together with its median Delta SNR. Selected Station Evidence shows one transmitter path's paired Delta SNR and evidence coverage over time, helping distinguish a representative path from an outlier or intermittent path. Drill-Down verifies that Target and Reference rows match the intended transmitter, cycle, callsign and locator identities and that the configured correction has the expected sign.

Common interpretation patterns. A station-median distribution mostly on one side of zero, broad Joint coverage and recurrence across time or adjacent segments support a consistent complete-path difference. A pooled Joint-Spot shift without a similar station-median shift can be driven by a few prolific transmitters. A clear paired median alongside many Only Target or Only Reference outcomes means that the paired strength difference describes only part of the practical decode evidence. A difference confined to one direction or UTC period can be real and useful while remaining path- or condition-dependent.

Boundary and confirmation. Paired analysis is conditional on both sides producing comparable evidence and therefore cannot describe all missed signals. Same-cycle matching controls the remote transmitter and timing, but it does not remove local receiver-chain, antenna, noise or QTH differences. Strengthen the result with enough Joint stations, independent Reference-uptime knowledge, repetition and the design-specific control described below.

2.3.1 Hardware A/B: simultaneous receive paths

Use this design for two local antennas, feedlines, filters, preamplifiers, receivers or complete receive chains operated simultaneously at the same physical test QTH. Target and Reference need distinct exact reporting callsigns and the same Target grid-4. Components intended to be common must be physically common; shared grid-4 matching does not prove co-location or path equality.

This is the strongest RX design for attributing a difference to a local path. The result still compares the complete documented receive paths unless receiver, audio, gain, decoder and routing differences have been characterized. A broad recurring Delta-SNR shift plus compatible one-sided evidence supports one path outperforming the other under the tested conditions. A common-input calibration, splitter-output swap or hardware crossover is the most useful confirmation because it can separate the tested component from a persistent chain offset. Appendix C describes Reference SNR calibration.

Under the documented simultaneous RX Hardware A/B setup, paired Delta SNR and Decode Outcomes described the observed difference between the Target and Reference receive paths for the shared transmitters, cycles and selected geographic scope.

2.3.2 Reference Station / Buddy Test

Use a known external receiver whose QTH, identity, equipment, operating schedule and local environment are understood. RX pairs share the same remote transmitter and cycle, but the two receiving stations remain at different QTHs with different antennas, terrain, hardware and noise.

Interpret this as a benchmark of complete installed receiving stations. It can show where one station was relatively stronger, how that relationship changed with direction, distance or time, and whether one-sided reach differed. It cannot isolate receiver sensitivity, antenna gain or local noise as the cause. Repetition with the same well-understood Buddy and stable operating conditions is the most useful confirmation.

For the shared transmitter paths and cycles in this run, paired Delta SNR and Decode Outcomes described how the two complete receiving stations compared under their respective environments.

2.3.3 Local Median Neighborhood

The Reference is the cycle- and path-specific median of one contribution from each active local receiver identity inside the selected radius. Membership can change from cycle to cycle, so the result is a contextual local baseline rather than a fixed station comparison.

Inspect the contributing local identities, Joint Evidence Share and radius sensitivity. A shift can reflect the Target, a changed neighborhood composition or both. Choose the primary radius from local geography and station density before interpreting the result; use other defensible radii as sensitivity analyses.

Relative to the active median receiver neighborhood inside the selected radius, the Target showed the reported paired Delta SNR and Decode Outcomes for the observed transmitter paths and cycles.

2.3.4 Local Best Station

The Reference is the strongest qualifying local receiver evidence available for each remote transmitter path and cycle. The winning local identity can change continuously, creating a demanding best-peer envelope rather than a local average or one permanent competitor.

Inspect which local station supplies the Reference and whether the observed gap is broad or concentrated in a small set of winners. Radius and pool membership remain part of the result. Report it as the Target's gap to the changing strongest local receiver, not as a ranking against one fixed station.

Relative to the strongest qualifying local receiver selected for each path and cycle inside the stated radius, the Target showed the reported paired Delta SNR and Decode Outcomes.

2.4 TX Benchmark

Question answered. How did the Target transmitter or scheduled path differ from the selected Reference at shared remote receivers?

Shared TX Benchmark evidence. Same-cycle TX Benchmark compares Target and Reference at the same remote receiver in the same WSPR cycle. Sequential Hardware A/B instead uses deterministic scheduled pairs at the same receiver. Successful TX SNR is normalized to reported power before Delta SNR is formed; the result therefore depends directly on accurate power reporting. Decode Outcomes preserve Joint and one-sided evidence, but an exclusive observation has no missing-side SNR and is not power-normalized.

Read the evidence path. On the Map, sector color summarizes station-balanced median Delta SNR across remote receivers. Marker and footer categories show Joint and one-sided receiver evidence. Read each sector with receiver breadth and spot or scheduled-pair depth.

In Segment Inspector, compare the station-level Decode Outcomes with the observation- or pair-level composition. Station Medians give each remote receiver one equal vote, while Joint-Spot or Scheduled-Pair Delta SNR shows the full paired observation distribution. A shift shared across many receivers is different from one dominated by a few high-volume receivers.

Temporal Evidence shows whether Delta SNR changed through the run or recurred by UTC hour. Benchmark Evidence Coverage shows whether paired evidence remained broad through those times. For sequential TX, inspect whether the result is tied to one schedule phase or switching period; for simultaneous TX, inspect whether it is tied to one receiver, audio-frequency assignment or short interval.

In Station Insights, read each receiver's median Delta SNR with its Joint and one-sided counts. Selected Station Evidence reveals the paired result and evidence coverage at one receiver path. Drill-Down verifies receiver identity, reported powers, same-cycle pairing or scheduled-pair assignment, and correction sign.

Common interpretation patterns. A consistent station-median shift across many receivers, directions and times supports a broad complete-transmit-path difference. A shift limited to one azimuth or distance range can indicate useful installed directional behavior without becoming a context-free gain value. Strong paired Delta SNR with substantial one-sided evidence means the strength difference and practical near-threshold reach must both be reported. A raw-pair median that differs from the station-median view indicates that high-volume receivers weight the observation-level evidence differently.

Boundary and confirmation. TX Benchmark remains conditional on pairable evidence and accurate reported power. Simultaneous designs retain transmitter-chain, frequency-response, isolation and coupling differences. Sequential designs remain time-separated. Same-cycle one-sided evidence is also affected by the Target-Active Gate. Strengthen the result with broad receiver support, accurate power measurement, repeated runs and the method-specific controls below.

2.4.1 Hardware A/B: simultaneous transmit paths

Use two distinguishable transmitter chains and callsigns at the same physical test QTH, deliberately synchronized in the same WSPR cycles and placed on clear, non-overlapping frequencies within the WSPR passband. Measure or otherwise establish actual power at the comparison point relevant to the question, and ensure adequate isolation between the active transmitters and antennas.

Same-receiver, same-cycle Delta SNR removes the sequential time gap and is the strongest TX design when the two transmitter chains can be controlled. It still compares the complete documented transmit paths. Frequency-selective QRM, chain response, coupling and power error can remain. Swap audio-frequency assignments and, where practical, cross the tested antennas or components between chains. Appendix A covers parallel WSJT-X setup.

Under the documented simultaneous two-transmitter Hardware A/B setup, same-receiver, same-cycle Delta SNR and Decode Outcomes described the observed difference between the Target and Reference transmit paths for the selected receivers and geographic scope.

2.4.2 Hardware A/B: sequential transmit paths

Use a deterministic schedule that assigns complete WSPR transmissions to Target and Reference phases. One transmitter switched between two RF paths is normally the strongest arrangement because callsign, frequency reference and transmitter remain common. Enter each path's actual recurrence and UTC phase, verify the physical schedule-to-path mapping without RF, and report actual power. Device-specific scheduling and switching guidance is in Appendix B.

WSPRadar forms one-to-one Scheduled A/B Pairs automatically. Pair Delta remains sequential: short balanced alternation reduces but does not eliminate propagation, interference, schedule-position and switching effects. Inspect incomplete pairs and chronological behavior with the paired median. Reverse the Target and Reference schedule assignments in a confirmatory run; persistence of the physical-path advantage after the role reversal is substantially more persuasive than repetition with the same phase assignment.

Under the documented deterministic schedule, Scheduled-Pair Delta SNR and one-sided pair outcomes described the observed difference between the switched Target and Reference paths for the selected receivers, times and geographic scope.

2.4.3 Reference Station / Buddy Test

Use a known external transmitter whose QTH, identity, actual and reported power, equipment and operating schedule are understood. TX pairs share the same remote receiver and cycle, but Target and Reference remain complete stations at different QTHs with different antennas, feedlines, terrain and paths.

Interpret the result as a benchmark of complete installed transmitting stations. Same-receiver pairing controls the receiving endpoint, not the two transmit sites or radio paths. Power-reporting accuracy is especially important. Repeat with the same well-understood Buddy and stable configurations rather than treating the Buddy as an absolute calibrated standard.

For the shared receiving stations and cycles in this run, paired Delta SNR and Decode Outcomes described how the two complete transmitting stations compared under their respective operating environments.

2.4.4 Local Median Neighborhood

The Reference is the cycle- and receiver-path median of one contribution from each active local transmitter identity inside the selected radius. It is a changing local baseline, not one fixed station, and it depends on the active membership and accuracy of their reported powers.

Inspect the local contributors, Joint Evidence Share and radius sensitivity. Report whether the Target tends to sit above, near or below the active local baseline for particular receivers, directions or times. A changed result can reflect the Target, the local pool or both.

Relative to the active median transmitter neighborhood inside the selected radius, the Target showed the reported paired Delta SNR and Decode Outcomes for the observed receiver paths and cycles.

2.4.5 Local Best Station

The Reference is the strongest qualifying local transmitter evidence available at each remote receiver and cycle after the applicable correction. The winning local identity can change continuously, producing a best-peer envelope rather than a local average or permanent competitor.

Inspect which station supplies the Reference, its reported power and whether the Target gap persists across receivers, distance, direction and time. Report the result as comparison with a changing strongest local transmitter inside the stated radius.

Relative to the strongest qualifying local transmitter selected for each receiver path and cycle inside the stated radius, the Target showed the reported paired Delta SNR and Decode Outcomes.

2.5 Find and Review Temporary ΔSNR Departures

This is an optional expert diagnostic tool, not a routine step intended for every operator or every Benchmark analysis. Use it when a controlled Benchmark has enough paired evidence around the period of interest and the question specifically concerns a temporary departure from one radio path's usual Target-versus-Reference behavior. Start with the ordinary Benchmark evidence path first; enable outlier reporting only when its additional event-level detail serves the investigation.

The detector does not search for the largest raw Delta SNR in the run. It asks whether one exact callsign + locator path temporarily moved above or below a stable local expected Delta SNR by enough to pass the configured departure, robust-score and baseline-stability requirements. The result is a set of intervals for expert review, not an automatic explanation of why the observations changed.

2.5.1 When to use this expert diagnostic

Use the detector only with Benchmark evidence. Its native evidence unit is a simultaneous Joint Spot or, for sequential TX Hardware A/B, one complete Scheduled Pair. Only those units contain both Target and corrected Reference SNR and therefore a paired Delta SNR. Only Target and Only Reference outcomes remain useful diagnostic context but cannot themselves qualify an event.

The method is most useful when the selected path has repeated paired observations before, during and after a suspected change. It deliberately abstains when a reliable local baseline cannot be supported on both sides. No reported event can therefore mean either that the retained evidence did not pass the configured gates or that the local evidence was insufficient or unstable; it does not establish that the path was unchanged.

Enable Report ΔSNR outlier candidates and set the three expert controls described in Section 4.6, then run the analysis manually. Changing the toggle or a threshold marks the analysis definition as changed but does not automatically start another run.

2.5.2 How detection works in practice

For each exact path, WSPRadar:

keeps the paired observations at their native WSPR-cycle or Scheduled-Pair times;
estimates the path's expected local Delta SNR from robust summaries before and after a possible departure, excluding the candidate itself;
requires those two baseline sides to have enough populated evidence and to agree within the configured maximum;
groups nearby, same-sign residuals using the path's observed evidence cadence;
tests the grouped event against the same absolute-departure, robust-score and sign-agreement rules regardless of its duration; and
trims weak leading and trailing evidence so that the reported interval begins and ends on observations that individually meet both user thresholds.

Weaker observations can remain inside an interval when they connect stronger observations, but they cannot extend its outer boundaries. A failed broad candidate can be split at a supported return to baseline so that a strong internal section is tested on its own without choosing a new, more favorable baseline.

Detection is completed before the Temporal Evidence display is aggregated. Changing a 1h, 6h or other display bin therefore cannot create, merge, split or remove an event. After path events qualify independently, contemporaneous same-sign events can be grouped for review: path-specific means one path, directionally coherent means several paths in adjacent compass sectors, scope-wide means several separated directions, and multiple paths means several paths with direction unavailable for at least one contributor. That context does not change whether an individual path qualified. The exact construction, support counts, timing rules and formulas are in Section 7.11.

2.5.3 Read and investigate a reported event

Read an event card from the interval down to its source evidence:

Spot impulse, Short burst and Sustained excursion describe the retained temporal shape after boundary trimming. They do not use different qualification thresholds or express different certainty.
Each path line identifies the exact callsign + locator and direction. Expected local Delta SNR is the candidate-excluded two-sided baseline; Observed median Delta SNR summarizes the retained units in the reported interval; Largest single-cycle departure is the most extreme retained residual from that baseline. That report/export metric is independent of the all-path temporal *, which selects the greatest-absolute-residual unit only among individually qualifying native units in each review event and never uses an unsupported or nonqualifying episode peak.
For a multi-unit event, Chronological WSPR-cycle evidence lists the contributing UTC time, path, direction, local baseline, Delta SNR and residual. A one-unit Spot impulse needs no duplicate evidence table.
The green ↓ Show in Station Insights and ↓ Show Drill-Down Details actions sit to the right of their exact path timeframe. The first selects that path, preloads Drill-Down and navigates to Station Insights for the wider run history; the second makes the same selection and preload but navigates directly to Drill-Down. Both open Outlier Focus over the complete supported pre-event baseline flank, guarded provisional episode and post-event baseline flank, clipped only to the completed analysis window; this detector-support interval may exceed 24 hours. The focused Delta SNR plot shows one actual retained Joint Spot or complete Scheduled Pair at its native time instead of a bin median, IQR or density layer. Identical * markers identify every native unit in the current window that individually meets both configured departure and robust-z gates as part of a reported candidate, including all such units in a multi-unit burst or episode. A muted Focused episode band identifies the selected reported episode. It spans that episode's reported retained-evidence interval, padded by half one native evidence-unit width at each end and clipped to the focused window so a one-unit impulse remains visible; it is neither a confidence interval nor a measurement of physical-event duration. Overlays show the expected local Delta SNR, the pre/post flank baselines over their actual support intervals, symmetric robust-z guides at 1, 2 and 3 plus the configured qualifying threshold, and the configured absolute-departure boundary for the focused episode only. Other starred candidate units may have been evaluated against different local baselines and robust spreads. These are detector guides, not confidence intervals, and crossing any one guide cannot qualify a candidate by itself. Use the underlying Target and corrected Reference SNR values and nearby one-sided outcomes to check whether the Delta SNR movement came mainly from one side, whether either signal approached the decode edge, and whether pairability changed nearby.

The displayed first-to-last span is the interval between retained observations. It does not assert uninterrupted behavior between them. Compare the interval with contemporaneous paths, station logs, switching schedules, gain or power changes, interference observations and independent measurements before assigning a cause.

2.5.4 Adjust selectivity and preserve the result

The three controls always retain their literal meaning:

increasing Minimum absolute ΔSNR departure (dB) requires a larger departure;
increasing Minimum robust z-score requires the departure to be larger relative to nearby robust variability; and
decreasing Maximum pre/post baseline difference (dB) requires a more stable two-sided baseline.

The opposite changes make reporting more permissive. The same three values apply to Spot impulses, Short bursts and Sustained excursions; duration provides no hidden discount. For exploratory work, use candidates to identify intervals worth auditing and record any threshold changes. For confirmatory work, fix and preserve the thresholds, Benchmark design, correction, path population, band and UTC scope before inspecting the result, then test whether a comparable departure recurs in a separate suitably controlled run.

Report the observation as a temporary local Delta SNR departure in the retained paired evidence, together with its exact path, interval, sign, supporting units and detector settings. The detector is descriptive: it does not calculate a p-value, correct for the number of searched paths or events, or determine the physical cause.

When outlier reporting is enabled, preserve the active-scope findings with the analysis export. It adds a path-event summary and a chronological paired-evidence CSV linked by package-local Event and Path event IDs. The evidence table deliberately repeats the path, direction and path-event class so it remains readable on its own. Section 8.4 defines the two files and their conditional inclusion.

3. Strengthen and Communicate Your Result

A strong WSPRadar result combines a clear experiment, broad evidence and language that matches the actual observation.

3.1 Judge breadth, consistency and repeatability

Judge the result from the complete evidence picture:

participating station identities;
qualifying confirmed-opportunity, spot or scheduled-pair volume;
agreement across stations;
station-balanced and observation-level summaries;
adjacent geographic segments;
temporal views;
Decode Outcomes;
identity and locator quality;
experiment control and repetition.

Evidence is broader when several identities and adjacent segments agree. It is more internally consistent when station-balanced, observation-level and time views tell a compatible story. It is better controlled when the selected playbook's operating requirements were followed and documented.

Internal consistency and experimental repeatability are different. Agreement among the station-balanced, observation-level, geographic and time views describes the evidence within one run. Repeating the experiment in another suitable window tests whether the observed pattern persists under new operating and propagation conditions.

WSPRadar deliberately does not collapse these dimensions into one proof grade. The visible counts, distributions and underlying rows let the operator judge the result in the context of the actual experiment.

The observed time, distance, direction, Decode Rate, successful-SNR or Delta-SNR pattern is the evidence. An explanation such as antenna directivity, a local-noise change, propagation mode, overload or an intermittent component is an interpretation. Match the wording to the observation first, then test the explanation through a controlled change, crossover, independent measurement or repetition.

3.2 Strengthen a result through repetition and control

Use an initial exploratory run to identify a possible pattern. Before a confirmatory repetition, freeze the direction, band, benchmark, filters, evidence thresholds, schedule and primary geographic or temporal evaluation scope, including Maximum peer distance from Target (km) as defined in Section 4.4. Run alternative maximum distances as separately preserved sensitivity analyses rather than selecting only the most favorable scope after seeing the result.

When the result will support an important station decision:

extend the observation window across the propagation states named in the conclusion;
prefer multi-day evidence for statements spanning complete daily cycles;
repeat the experiment on another day or propagation period;
for sequential TX Hardware A/B, reverse the Target/Reference schedule assignments;
keep non-tested variables stable between repetitions;
compare runs with the same direction, band, benchmark, filters and evidence thresholds;
investigate any identity, locator or short interval that supplies a large fraction of the evidence;
preserve setup notes so a later run can reproduce the station configuration.

Small observed differences become more useful when they recur across stations, time periods, adjacent segments and controlled repetitions. A reversed sequential TX assignment is especially useful because it can expose schedule-, switch-path- or time-of-cycle effects that ordinary repetition leaves in the same role.

TX and RX use different peer populations and opportunity definitions. Compare like-for-like TX and RX runs when investigating station balance or an "alligator" pattern.

3.3 Write an evidence-matched conclusion

A minimum operator statement identifies the Target and, for Benchmark where applicable, the fixed Reference or local benchmark definition. It also identifies the TX or RX direction, band, UTC window, geographic scope, result type, displayed value and supporting station/evidence count.

A full technical report also states:

the applicable weighting levels: Station-balanced and Opportunity-level Decode Rate for Performance, or station-level and observation-level Delta SNR for Benchmark;
qualifying-station and confirmed-opportunity counts for Performance, or joint-station and joint-spot/pair counts for Benchmark;
Decode Outcomes for Benchmark;
experiment conditions and any Reference correction;
filters and evidence thresholds;
whether the pattern repeated across time, stations or runs;
any alternative radius or scope used as a sensitivity analysis.

Performance wording

For this Target, band, UTC window and selected peer population, the displayed Decode Rate describes the fraction of independently confirmed opportunities in which the Target also produced qualifying evidence. State whether the reported value is the Station-balanced Decode Rate or the Opportunity-level Decode Rate. Qualifying stations, confirmed opportunities, geographic scope and temporal views describe the breadth, depth and recurrence supporting that result.

A complete Performance statement can additionally say whether at-least-once reach was broad or limited, whether participation was consistent or intermittent, where distance or directional patterns appeared, whether a UTC-hour pattern recurred and how successful Target SNR behaved. Describe these as observed WSPR behavior of the complete station under the selected conditions, not as isolated gain, sensitivity or efficiency.

Benchmark wording

For this Target, Reference, band, UTC window and selected segment, station-balanced Delta SNR favored the Target/Reference by the displayed amount. The observation-level Delta SNR, joint station and spot/pair counts, Joint Evidence Share and Decode Outcomes describe the supporting paired and one-sided evidence.

For a controlled Hardware A/B result, name the complete paths compared and any crossover or calibration. For a Reference Station / Buddy Test, state that complete installed stations and their environments were benchmarked. For a Local Neighborhood Benchmark, state the radius, method and changing Reference definition.

Match the design name to the quantity being described:

A Hardware A/B Test compares the documented local paths.
A Buddy Test compares complete installed stations and their environments.
Local Median Neighborhood compares the Target with the active median-neighborhood definition inside the selected radius.
Local Best Station compares the Target with a changing best-peer envelope.
A directional result describes the observed WSPR paths and participating stations rather than an absolute radiation pattern.
Benchmark map colors use a run-scaled, symmetric dB color bar: blue favors the Reference, red favors the Target and 0 dB is equality. Use the numerical color-bar values when comparing maps from different runs.

Use terms such as "observed difference," "favored in the selected evidence," "conditional reach" and "complete installed station comparison." Reserve isolated antenna gain, efficiency, receiver sensitivity, causation and statistical significance for experiments that actually measure or test those quantities.

The complete supported/unsupported wording reference is in Chapter 8.

3.4 Preserve the run and its context

Use Prepare All Results for Download to build the current analysis export package. It contains the current configuration, run metadata, processed evidence, tables and high-resolution figures.

Preserve external notes alongside that package:

physical antenna and feedline arrangement;
switch or splitter topology;
transmitter or receiver hardware;
power measurements and reporting basis;
decoder and software versions;
operating schedule, physical schedule-to-path mapping and any reversed assignment;
calibration procedure;
weather, faults or intentional changes relevant to the run.

WSPRadar can preserve the configured analysis and processed evidence, but it cannot infer every physical detail of the station. Combining the export package with concise station notes makes comparison and reproduction substantially stronger. Chapter 8 documents the exact export contents and remaining reproducibility boundaries.

Part II: Controls and Troubleshooting

Use this part as an operating reference while setting up, repeating or diagnosing an analysis. It documents the exact controls, defaults, saved behavior and scientific consequences that affect the operator.

Optional expert use of Benchmark Delta SNR outlier detection is introduced in Section 2.5. Section 4.6 owns its controls; its formal scientific definition is in Section 7.11.

4. Controls and Configuration

WSPRadar distinguishes controls that change the retained scientific evidence from controls that only change how completed evidence is inspected.

Control class	Effect	Saved?	Rerun required?
Scientific controls	Change identity, band, time, Reference design, eligibility, normalization, filters, thresholds or geographic population.	When applicable	Yes; the previous result is cleared
View controls	Change the active inspection scope, selected station, evidence visibility or display aggregation without reclassifying retained evidence.	Supported durable choices only	No
Temporary interface choices	Change only the current on-screen arrangement, temporary table filters, documentation visibility or a prepared download.	No	No

Versioned configurations store the applicable scientific settings and supported durable view choices. Exact calculations are defined in Chapter 7; Section 8.4 summarizes selected public machine-readable configuration, URL and export names. The formal JSON Schema is authoritative for the exhaustive saved-configuration field contract.

4.1 Workflow controls

Control	What it does	Important behavior
`Input view`	Switches between `Guided` and `Classic`.	Both expose the same scientific configuration. The chosen editor is not saved.
`Load Demo`	Loads a maintained historical profile.	Loading does not start an analysis. Filter, evidence-threshold and result-view changes retain the demo context; changing the experiment definition detaches it from the demo.
`Load Config`	Loads a versioned JSON `.config`.	Invalid identities, dates, choices, ranges, duplicate fields and unsupported schema versions are rejected rather than guessed.
`Save Config`	Saves the applicable scientific inputs and supported durable view settings.	The file stores absolute UTC boundaries but not result rows, external experiment notes or transient table filters. In Classic, saving remains unavailable until the Question and, for a Benchmark, the Benchmark design are complete.
`Run RX Analysis` / `Run TX Analysis`	Runs the selected Performance or Benchmark result.	In Classic, running remains unavailable until the Question and, for a Benchmark, the Benchmark design are complete. Changing a scientific control after a run clears the result and requires a new run.
`Prepare All Results for Download`	Builds the current export package.	Uses the completed evidence and current inspector selections.
`Load full documentation` / `Hide full documentation`	Shows or hides the complete web manual.	Presentation state only.
`Prepare PDF`	Builds the selected-language manual as PDF.	The full web manual does not need to be open first.

Configuration compatibility. Saved files preserve the inputs and durable view choices applicable to the selected analysis. Invalid or unsupported files are rejected rather than silently reinterpreted. The formal JSON Schema is the authoritative exhaustive saved-configuration contract; Section 8.4 gives a concise operator-facing summary of selected public identifiers. Loading or saving a configuration does not create an additional result; only the selected Performance or Benchmark analysis is run.

Demo context lifecycle. A loaded demo keeps its visible context when only population filters, evidence thresholds, Inspector scope or other result-view controls are changed, so an adapted view can still be interpreted against the example from which it began. Changing the Question or direction, Target callsign or QTH, band, measurement window, Benchmark design or identity, neighborhood radius, TX schedule, or correction intent/value removes the demo metadata and profile identity from later saves because the setup no longer represents that documented experiment. Any scientific edit also ends exact-demo cache identity, even when the explanatory demo context remains visible. A population- or evidence-changing scientific edit clears any preselected Performance and Benchmark Station Insights identity; the path may no longer exist in the new result. Result-view-only controls do not clear that selection.

4.2 Question, Target and measurement-window controls

Classic presents the scientific setup in a question-led order. The first panel, Question, requires one of four complete analysis choices: RX Performance, TX Performance, RX Benchmark or TX Benchmark. This single choice sets both the RX/TX direction and whether the run produces stand-alone Performance evidence or a Target-versus-Reference Benchmark. The second panel, Target and measurement window, then collects the existing Target identity, QTH, band and absolute UTC interval.

UI label	Default	What it controls
Question	none; required	One of `RX Performance`, `TX Performance`, `RX Benchmark` or `TX Benchmark`; sets direction and result type together.
Target callsign (receiver under test) / Target callsign (transmitter under test)	blank	Exact archive reporting identity. Standard callsigns, valid `/` variants, letter-only reporting identifiers and one optional terminal alphanumeric hyphen suffix are accepted.
Target QTH (4 or 6 characters)	blank	Target grid-4 matching, map center, geometry and local-radius origin.
Operating Band	`20m`	Exactly one of `LF`, `MF`, `160m`, `80m`, `60m`, `40m`, `30m`, `22m`, `20m`, `17m`, `15m`, `12m`, `10m`, `8m`, `6m`, `4m`, `2m`, `70cm` or `23cm`.
UTC measurement window	fixed 24-hour window ending at the current 15-minute UTC boundary	The absolute evidence interval used by the run.
Start Date/Time (UTC) and End Date/Time (UTC)	the effective default window	Dates begin in 2008; one run is limited to 31 elapsed days. Edited values are rounded down to effective 15-minute boundaries and shown back in the controls.

Use the callsign or reporting identifier exactly as uploaded. KFS, KFS/SE, DL1MKS, DL1MKS/P, DL1MKS/1, DL1MKS/QRP and DL1MKS-1 are distinct identities; WSPRadar does not apply hidden prefix or suffix matching.

A four-character Maidenhead locator identifies a broad grid square; six characters identify a smaller subsquare. Performance and Benchmark select Target archive rows from the exact callsign plus the first four characters of Target QTH. The full configured QTH still anchors map, distance, azimuth, solar and local-neighborhood calculations.

4.3 Benchmark-design controls

For RX Benchmark and TX Benchmark, Classic displays a third panel named Benchmark design and requires one of:

Hardware A/B
Known Reference Station
Local Neighborhood

Classic omits the Benchmark design panel entirely for RX Performance and TX Performance, because Performance has no Reference. The direction-specific Run RX Analysis / Run TX Analysis action and Save Config remain unavailable while the Question is incomplete or while a Benchmark question has no complete Benchmark design. Performance and Benchmark are mutually exclusive result types: one run produces only the selected result. Section 8.4 summarizes selected public machine-readable configuration, URL and export names; it is not an exhaustive field or parameter catalog.

UI label	Default / range	Applies to	Scientific effect
Is there an established Target–Reference offset?	`No established offset — use 0.0 dB`	Guided Hardware A/B and Known Reference Station	Distinguishes no established correction, use of one established correction, and a deliberate offset-establishment run.
Reference-side SNR correction (dB)	blank = `0.0`; `-99.9` to `+99.9 dB`	Benchmark	Added to Reference SNR before Target-minus-Reference Delta SNR is calculated. Enter decimal points, for example `1.2`.
Reference callsign	blank	Hardware A/B and Reference Station	Exact Reference reporting identity.
Reference Locator	independent grid-4 for Reference Station; derived Target grid-4 for Hardware A/B	Benchmark	Controls Reference archive matching.
Local Benchmark Method	`Local Median Neighborhood`	Local Neighborhood Benchmark	Selects the median local Reference or the strict changing Local Best Station.
Neighborhood Radius (km)	`100`; 10–250 km in 10 km steps	Local Neighborhood Benchmark	Defines the local Reference pool around Target QTH.
TX A/B Method	`Simultaneous TX`	TX Hardware A/B	Selects same-cycle two-transmitter matching or deterministic sequential pairing.
Repeat Interval	`10 min`; `4, 6, 10, 12, 20, 30, 60 min`	Sequential TX A/B	Actual recurrence of each physical path.
Target Start / Reference Start	`00 UTC` / `02 UTC`; distinct even phases below Repeat Interval	Sequential TX A/B	Assigns transmissions to Target and Reference schedule phases.

For TX Hardware A/B, Repeat Interval is each path's actual recurrence, not necessarily a transmitter's displayed Frame value. Compare the one-hour preview with the observed on-air starts and physical switch mapping. Device examples are in Appendix B; pair construction is in Sections 7.1 and 7.7 [Ref-12].

Switching the Question or Benchmark design hides controls that do not apply. Saved configurations contain only the inputs applicable to the selected analysis. Values whose scientific meaning changes under the new design are cleared rather than reinterpreted.

Reference-side SNR correction sign

A positive correction increases corrected Reference SNR and therefore reduces Target-minus-Reference Delta SNR. Enter a measured target - reference calibration offset with the same sign. For example, a common-input calibration of +1.6 dB is entered as +1.6 dB. Section 7.5 defines the equations.

The correction applies to the Reference receive/transmit path or schedule in Hardware A/B, the known Reference Station, the selected Local Best Station value, or each local contribution before the Local Median Neighborhood is formed.

Guided choice	Meaning	Required value
No established offset	No defensible correction has been established.	`0.0 dB`
Use an established correction	Apply a documented signed additive offset valid for this setup.	Enter the established value
Set up an offset-establishment run	Collect evidence from which an offset can be derived; WSPRadar does not choose or calculate the offset automatically.	`0.0 dB` during the establishment run

A constant correction cannot repair clipping, unstable AGC, intermittent routing, frequency-dependent response or incorrect power reporting. Hardware A/B calibration should use a common input or calibrated reference plane. A geographically separated Reference Station can support only a repeatable baseline for that particular pair, band and setup — not an absolute calibration. Appendix C gives the practical procedure.

4.4 Filters and evidence thresholds

Choose filters and thresholds from the intended population and evidence floor before a confirmatory run. Changing them after inspecting the result creates a different analysis and should be retained separately.

Control	Default	Applies to	Effect and use
Exclude Special Callsigns Q, 0, 1	Performance on; Benchmark off	all results	Excludes peer identities beginning with `Q`, `0` or `1`. Retain beacon/telemetry-like identities when they are part of the question; exclude them when the intended population is ordinary amateur activity.
Exclude Moving Stations	Performance on; Benchmark off	mapped peers	Excludes callsigns reporting more than one grid-4 in the otherwise eligible global population. Use Drill-Down to distinguish movement from bad locator data.
Solar state at Target QTH	`All 24h`	all results	Keeps `Daylight (Elev > +6°)`, `Nighttime (Elev < -6°)`, `Greyline (-6° to +6°)` or all cycles according to Target-QTH solar elevation.
Maximum peer distance from Target (km)	`22000`; choices `2500`, `5000`, `10000`, `15000`, `20000`, `22000`	all results	Removes peers at or beyond the selected distance from analysis, processed artifacts and exports. Target-Active gating may still use out-of-scope evidence solely to establish Target operation.
Minimum joint evidence per station	`1`; range 1–50	simultaneous Benchmark	Requires repeated Joint peer-cycles before a station contributes paired Delta SNR; the same numeric floor also applies to exclusive categories.
Minimum scheduled pairs per station	`1`; range 1–50	sequential TX A/B	Requires repeated complete scheduled pairs before a station contributes Pair Delta; one-sided pair categories use the same numeric floor.
Minimum confirmed opportunities per station	`5`; range 1–100	Performance	Requires enough Target-plus-counter opportunities before a peer contributes. Low values increase coverage but make rates coarse and weakly supported.
Minimum qualifying stations per map segment	`1`; range 1–10	all maps	Requires broader identity support before a segment is drawn.

The two exclusion defaults apply only to untouched interactive setups. A Performance setup starts with both exclusions on; a Benchmark setup starts with both off. After the operator changes either exclusion manually, that explicit value persists across Question changes rather than being replaced by a result-type default. Loaded configurations, demos and analysis URLs likewise retain their explicitly saved choices.

Maximum peer distance from Target (km) limits the analysed population after the archive rows have been retrieved, so reducing it does not avoid the archive row limit. A smaller Local Neighborhood radius and Exclude Special Callsigns Q, 0, 1 can reduce the population retrieved for some analyses; Section 5.6 covers oversized requests.

4.5 Map, inspector and export controls

Control	What it changes	Saved?	Reruns analysis?
Segment distance and direction	Active geographic inspection scope	Separately for Performance and Benchmark	No
`Heard only by other stations.` / `Only other signals heard.`	Visibility of Performance peers with only counter-evidence	Yes	No
`Include Unpaired Evidence`	Visibility of Benchmark identities represented only by exclusive or asynchronous evidence	Yes	No
Selected station row	Selected Station Evidence and selected Drill-Down identity	One exact `callsign + locator` per result type	No
Segment time aggregation	Chronological Segment Inspector temporal view; choices adapt to the run duration	Yes	No
Selected-station time aggregation	Chronological selected-path view; choices adapt to the run duration	Yes	No
`Zoom window`, `Center date (UTC)`, `Center time (UTC)`, `← Earlier`, `Later →`, `Outlier Focus` and `Filter table`	Optional native-time Drill-Down plots and centered table interval for exactly one selected station; table filtering affects displayed rows only	No	No
`Prepare All Results for Download`	Export package and current inspection selections	n/a	No

The chronological choices and the default used when no explicit compatible saved choice exists are:

Complete run duration	Offered bins	Default
Up to and including 6 hours	`2m`, `10m`, `30m`, `1h`, `2h`, `3h`, `6h`	`10m`
More than 6 hours through 24 hours	`2m`, `10m`, `30m`, `1h`, `2h`, `3h`, `6h`	`30m`
More than 24 hours through 7 days	`30m`, `1h`, `2h`, `3h`, `6h`, `12h`, `24h`	`12h`
More than 7 days	`1h`, `2h`, `3h`, `6h`, `12h`, `24h`	`12h`

The 2h choice is therefore available for every run duration. Legacy 5m and 15m values remain accepted for saved-configuration and URL compatibility, but they are not offered for a new choice; an explicitly loaded valid legacy value remains selectable so it is not silently changed. Chronological aggregation never changes opportunity classification, Benchmark pairing or the fixed one-hour UTC-folded profiles. Empty Performance time or distance bins remain missing evidence rather than synthetic zero-rate observations.

Drill-Down zoom is transient and is available only for exactly one selected station. Choose Off, or a complete 1h, 3h, 6h, 12h or 24h interval. Center date (UTC) and Center time (UTC) select the center of that interval; WSPRadar derives its exact start and end, moves the complete interval against a run boundary instead of shortening it, and lets ← Earlier or Later → step by one complete selected window. The resolved bounds appear on one line as Selected window: {start} to {end} UTC. The zoom restricts the focused figures and Drill-Down table; Filter table then changes only the displayed table and never the focused plots or completed analysis. Its metric plot is deliberately not a two-minute aggregate: simultaneous Benchmark shows one actual Delta SNR dot per retained Joint Spot at its canonical cycle time; sequential TX A/B shows one actual Pair Delta SNR dot per retained complete Scheduled Pair at its planned Target-start coordinate; Performance shows the actual normalized Target SNR of each successful confirmed opportunity at its canonical cycle time. These are individual retained scientific evidence units after WSPRadar's consolidation, matching and filters, not untouched provider rows. No bin median, IQR, density background, colorbar, full-run median or UTC-hour-folded metric panel is drawn in the focused view. The companion Performance outcome or Benchmark coverage view may retain its chronological aggregation, while Segment and full-window Selected Station Evidence remain density-based aggregated views. Both Target and Reference component rows of an admitted sequential pair remain together in the table. Focused figure titles use the compact format DG2CAD (JN47mv) - Time Window: {start} to {end} UTC.

An outlier action preloads Outlier Focus over the complete supported pre-event baseline flank, guarded provisional episode and post-event flank, clipped only to the completed analysis window and allowed to exceed 24 hours. Candidate provenance remains attached while the operator moves to a manual fixed window. In a focused Benchmark plot, identical * markers identify every native unit in the current window that individually meets both configured departure and robust-z gates as part of a reported candidate; a muted Focused episode band distinguishes the selected reported episode. The band covers its reported retained-evidence interval with half one native evidence-unit width of padding at each end, clipped to the focused window so an impulse remains visible. It is a selection cue rather than a confidence interval or physical-duration measurement. Expected local Delta SNR, time-limited pre/post flank baselines, robust-z guides at 1, 2 and 3 plus the configured qualifying threshold, and the configured absolute-departure boundary belong only to the focused episode; other starred candidates can have different baselines and robust spreads. Robust-z and departure lines are detector guides rather than confidence intervals; crossing one line alone does not satisfy the detector's separate support, stability, event and agreement requirements. Manual and outlier-linked focus state stay outside the analysis definition, saved configuration and public URL. When focus is active, the export can add its separate figures without replacing the ordinary full-run selected-station figures. Export contents are defined in Section 8.4.

4.6 Benchmark outlier-detection controls

Benchmark Delta SNR outlier detection is an optional expert analysis over retained native paired evidence. A native paired unit is a simultaneous Joint Spot or, for sequential TX Hardware A/B, one complete Scheduled Pair. Detection runs separately for every exact peer callsign + locator path and is independent of the selected Temporal Evidence display bin. One-sided evidence cannot supply a missing Delta SNR. Section 2.5 explains operation and interpretation; Section 7.11 defines the method formally.

Control	Default / range	Method symbol	Scientific effect
`Report ΔSNR outlier candidates`	off	—	Enables the optional detector and Outlier Report for Benchmark results. When it is off, WSPRadar adds no outlier detection, fields, markers, report semantics or outlier export metadata to the result.
`Minimum absolute ΔSNR departure (dB)`	`6.0`; `0.1`–`100.0 dB` inclusive	$D_{\min}$	Requires the event's median residual, and every reported boundary anchor, to depart from the local baseline by at least this magnitude.
`Minimum robust z-score`	`3.0`; `0.1`–`100.0` inclusive	$Z_{\min}$	Requires the event median and every reported boundary anchor to meet this absolute robust local-variability score. The score is descriptive, not a calibrated probability or conventional Gaussian significance level.
`Maximum pre/post baseline difference (dB)`	`3.0`; `0.1`–`100.0 dB` inclusive	$H_{\max}$	Rejects a candidate when the pre-event and post-event flank medians differ by more than this magnitude, so an unstable or shifted baseline is not reported as a temporary excursion.

The default combination prioritizes large absolute movements: the 6 dB gate sets the hard floor, while the robust z-score gate still requires that movement to be large relative to the path's local robust variability.

The toggle and three thresholds are saved when applicable. Changing any of them marks the configuration as changed but does not automatically start an analysis; calculate a new result with the normal direction-specific Run RX Analysis / Run TX Analysis action. All three thresholds apply unchanged to Spot impulses, Short bursts and Sustained excursions. There is no duration bonus or weaker threshold for a longer event. Lowering $D_{\min}$ or $Z_{\min}$, or increasing $H_{\max}$, makes reporting more permissive; the opposite choices make it more selective. For confirmatory use, set and record the thresholds before inspecting the candidate events rather than tuning them until a desired event appears.

5. Troubleshooting and Data Quality

Confirm the run definition before changing filters or thresholds. A wider scope can retain more evidence, but it cannot repair a wrong identity, band, time window or physical schedule.

5.1 Confirm the run definition first

Target identity: exact callsign or reporting identifier, including suffix.
QTH: configured locator and the first four characters actually uploaded.
Band: one exact selected band and actual operating band.
UTC evidence window: exact effective start and end shown in the controls.
Actual operation: Target transmission/reception and spot uploading.
Reference operation: exact Reference identity and overlapping uptime for Benchmark.
Design mechanics: clock synchronization, TX schedule-to-path mapping, switching, signal routing, actual and reported power.

Only after these checks should you change evidence thresholds, exclusions, solar state or geographic scope.

5.2 Diagnose by symptom

Symptom	Next checks
No result or no Target evidence	Check exact identity/QTH/band/window, actual operation, strict `code = 1` or historical-fallback status, and upstream availability.
Benchmark has no Delta SNR	Check shared remote peers in overlapping cycles or scheduled pairs, Reference uptime, clocks, schedule mapping, joint threshold, filters and scope.
Benchmark has Delta SNR but little pairable evidence	Read Joint Evidence Share and Decode Outcomes; check Reference uptime, power, thresholds, scope and whether the paired subset represents the wider station population.
Performance has very few peers	Check independent network activity, minimum confirmed opportunities, exclusions, solar state, time window and maximum peer distance.
Many Performance rows lack independent confirmation	The Target observations remain auditable but do not enter Decode Rate without the required independent activity evidence.
`Only Reference = 0`	Check Target-active conditioning, thresholds and active scope; zero can be correct.
Unexpected Hardware A/B Delta SNR sign	Verify physical A/B mapping, Target/Reference order, correction sign, schedule phases, actual/reported power and calibration. Reconcile one path in Drill-Down.
Local result changes with radius	Inspect local contributors and report radius sensitivity rather than selecting only the most favorable radius.
Run stops because the source result is too large	Shorten the UTC window. `Exclude Special Callsigns Q, 0, 1` or a smaller Local Neighborhood radius can reduce relevant source queries; maximum peer distance cannot because it is applied after retrieval.
Recent spots appear incomplete	Allow about five minutes after the final cycle, then check upload and upstream status.

An upstream-data problem changes what the source supplied. An experiment-design problem changes whether the retained evidence answers the intended question. Diagnose and report them separately.

5.3 Callsign and locator checks

Performance and every Benchmark design match Target archive rows by exact callsign plus Target QTH grid-4. A Target uploading JN37 while configured as JN38 does not match.

Reference Station uses exact Reference callsign plus an independent four-character Reference Locator. RX and simultaneous TX Hardware A/B derive the Reference grid-4 from Target QTH; sequential TX Hardware A/B uses the shared Target identity and distinguishes paths by schedule. Local References are selected geographically.

Callsigns must satisfy the documented 3–15-character reporting-token rule. Locators must contain four or six valid Maidenhead characters. Syntax validation does not prove legal assignment, physical location or actual operation. Peer identity is exact callsign + full reported locator; stale or changing locators can split or move one physical station.

5.4 Historical decode-code fallback

WSPRadar first requests WSPR-2 rows with code = 1. If that strict request returns no Target-side evidence, it retries without the predicate for historical compatibility and reports the fallback in run status. The fallback broadens selection and can differ between Performance and Benchmark.

5.5 How the Target-Active Gate shapes evidence

The Target-Active Gate retains simultaneous cycles only when Target participation is observable. Reference reports from periods when the Target was offline are therefore not counted as automatic Target failures.

The gate is intentionally Target-centric. Reference uptime remains an experimental responsibility, and swapping Target and Reference can change one-sided Decode Outcomes and the eligible population. Sequential TX Hardware A/B uses deterministic scheduled pairs instead. Section 7.3 defines the conditioning formally.

5.6 Working with upstream data

Public WSPR archives can contain duplicates, false spots, incorrect locators or power values, delayed uploads and later corrections. wspr.live describes fresh data as arriving after a delay of a few minutes; waiting about five minutes after the final cycle is a practical estimate, not a completeness guarantee [Ref-10].

WSPRadar reduces sensitivity to isolated bad rows through identity consolidation, medians, eligibility thresholds and Drill-Down, but repeated plausible errors can remain. Correct calculations cannot repair an incorrect reported power, locator or operating identity.

System Audit Status records the provenance needed to interpret the run:

Status element	Meaning
Data source	The single upstream archive used for the completed run. Evidence from different archives is not combined within one run.
Historical fallback	Whether source selection was repeated without the strict WSPR-2 decode-code condition.

These status items document where the evidence came from and whether the historical compatibility fallback was used; they do not define a different scientific method.

An archive retrieval larger than 1,000,000 complete rows is rejected before analysis rather than silently truncated. Shorten the window or use a relevant archive-side population filter as described in Section 5.2.

Part III: Scientific Foundations, Methods and Claims

Part III is the scientific methods reference for technically critical radio amateurs, HamSCI contributors and reviewers. It defines the observational data, analysis targets, constructed evidence units, descriptive summaries, conditioning, missingness, weighting, dependence, transformations and reproducibility boundaries behind WSPRadar. It is intentionally more formal than the operator guide.

6. Literature, Prior Art and Positioning

This chapter is a focused methodological review, not a systematic or exhaustive literature search. Peer-reviewed articles, preprints, amateur technical reports and software documentation support different kinds of claims; each source is used only for the contribution it actually demonstrates. The review does not imply that prior literature validates every WSPRadar metric or methodological choice.

6.1 From reporting network to experimental dataset

Taylor and Walker presented WSPRnet not merely as a live map but as an archive: “The WSPRnet database represents a rich source of experimental data for propagation studies.” Their example groups observations by time of day over several weeks, illustrating both the value of accumulated reports and the need to interpret them as observational rather than controlled laboratory data. [Ref-6]

Frissell et al. place WSPRNet alongside the Reverse Beacon Network and PSKReporter as established amateur-radio observation networks that provide long-term bottomside-ionosphere observations. They distinguish these networks from purpose-built scientific instruments and recommend cross-calibration between instrument networks. The review supports scientific use of amateur observations; it does not make each contributing receiver a calibrated sensor. [Ref-7]

The WSPR archive therefore combines unusual temporal depth and geographic reach with heterogeneous stations, successful-decode selection, user-supplied identities and powers, changing equipment and generally unknown operating schedules. These properties motivate explicit eligibility and conditioning rather than direct interpretation of spot absence.

6.2 Making observational WSPR data interpretable

Lo et al. used 7 MHz WSPR reports to study greyline propagation and warned that no authoritative operating schedules exist for WSPR equipment. Before interpreting a missing path, they checked whether a transmitter was heard elsewhere or whether a receiver heard another station, and they emphasized callsign/location consistency and multiple sites. [Ref-9]

That activity-check principle is direct prior art for WSPRadar's Target-Active Gate and independently confirmed opportunities: silence should not become counter-evidence until relevant operation is observable. Lo et al. do not define WSPRadar's asymmetric Target conditioning, Performance analysis target, station balancing, Decode Outcomes or local References; those remain WSPRadar design choices for different analysis questions.

6.3 Antenna and station-comparison lineage

Toledo (2010): why slow alternation fails. Sivan Toledo tested one antenna for roughly an hour and then another, finding path-SNR changes comparable with the apparent antenna difference. He concluded that this naive design could not isolate the antennas and proposed per-cycle switching or simultaneous transmissions with separate hardware. WSPRadar's deterministic interleaved TX A/B schedule follows the same practical logic: short separation reduces temporal confounding, but does not eliminate it. [Ref-3]

Milazzo (2011): operator-led end-to-end comparison. Carol Milazzo compared two stations 29 km apart through one receiver 1,750 km away, corrected reported SNR for transmit-power differences, compared the trend with VOACAP, noted unequal duty cycles and examined reciprocal RX reports. The case study demonstrates the practical value of common-receiver WSPR comparison while also showing the limits imposed by different QTHs, hardware, local noise, a single selected receiver and no formal uncertainty analysis. [Ref-4]

Griffiths and Squibb (2017): same-signal RX comparison as station diagnosis. For two receivers at separate QTHs, they retained reports of the same transmitter at the same time and related SNR difference to soil moisture, time, distance and station changes. The work shows how paired WSPR observations can diagnose complete receive systems and reveal structure hidden by spot totals. Because antennas, QTHs, noise and equipment differed, it supports comparative station evidence rather than isolated calibrated antenna gain. [Ref-5]

Vanhamel, Machiels and Lamy (2022): conditioned simultaneous RX. Their peer-reviewed experiment conditioned two nominally identical 160 m WSPR receiver stations and compared common remote transmissions simultaneously. This is the strongest direct precedent in this review set for RX Hardware A/B and for characterizing receive-chain offsets before interpreting antenna differences. Their propagation results also show that polarization and ionospheric effects remain coupled to reported SNR. [Ref-2]

Zander (2022): simultaneous same-receiver TX comparison. Zander models two local antennas driven by separate nominally equal-power transmitters with different callsigns in the same WSPR cycle. A remote receiver contributes only when it reports both signals in the same interval. Under the same-time, common-path and equal-power assumptions, common path loss and receiver noise cancel in the SNR difference; frequency-selective interference, failed decodes, quantization and transmitter-chain differences remain. Because each difference is formed within one remote receiver, receiver calibration is not required for that pair, while equality or correction of the two transmitted powers remains essential. [Ref-1]

Zander reports about 1,000 observations per preliminary experiment, of which roughly 150–200 joint reports from 15–35 receivers were retained, with sample standard deviation near 3 dB. The paper's sub-dB statement concerns precision of an arithmetic mean under its model and sample assumptions, not traceable total accuracy. Geographic sampling, antenna directivity and unknown elevation angles remain systematic limitations. The study supports simultaneous same-receiver Delta SNR, but not WSPRadar's sequential one-transmitter design, station-balanced medians, Decode Outcomes or neighborhood References.

6.4 Analysis infrastructure and related tools

Griffiths and Robinett demonstrated a relational time-series self-join for the same transmitter, time and band reported by two receivers, together with SNR-difference plots, medians, quartiles, time heatmaps, distance/azimuth views and export. This is important precedent for inspectable comparison infrastructure, not for WSPRadar's exact eligibility, conditioning or summary statistics. [Ref-13]

WSPR.Rocks provides rapid SQL-based WSPR exploration, maps, tables, SpotQ and heatmaps. WSPRdaemon emphasizes robust multi-receiver acquisition, scheduling and added noise/Doppler metadata. SOTABEAMS WSPRlite/DXplorer, WSPR-Station-Compare, the Antenna Performance Analysis Tool and WATT provide additional comparison, reporting and visualization workflows [Ref-14] [Ref-15] [Ref-16] [Ref-17] [Ref-18].

These systems establish substantial prior art in data acquisition, exploration, ranking, comparison, mapping and reporting. WSPRadar's positioning therefore rests on its integrated experiment definitions, conditional populations, hierarchical weighting, complementary paired/one-sided evidence and audit path — not on being the first WSPR analysis tool.

6.5 What WSPRadar inherits, integrates and adds

WSPRadar inherits accumulated WSPR observations, activity checks, reported-power correction, common-condition pairing, calibrated receive-chain comparison, database joins and geographic/time inspection. It integrates them into one TX/RX workflow with:

Performance based on independently confirmed opportunities;
Hardware A/B, Reference Station and dynamic Local Neighborhood Benchmarks;
same-cycle or deterministic scheduled-pair matching;
reported-power normalization and optional Reference-side correction;
paired Delta SNR separated from one-sided Decode Outcomes;
station-balanced and observation-level summaries;
map-to-segment-to-station-to-row audit; and
versioned configuration, processed evidence and reproducibility export.

Within the reviewed sources, WSPRadar's clearest specific additions are the explicit conditional Performance denominator, the paired-versus-one-sided evidence split, dynamic Local Median/Best References, hierarchical station-balanced geographic aggregation and an integrated audit path across all supported designs.

This is a bounded integration and methods claim, not a global priority claim. Median aggregation itself is not novel. WSPRadar should be described as a structured experimental and audit layer above a spot browser, not as a substitute for the upstream archives, other analysis tools or calibrated RF measurement.

7. Scientific Methods

This chapter defines the scientific contract of a WSPRadar run. WSPRadar starts from reported observations, constructs eligible evidence units, derives quantities such as normalized SNR and paired Delta SNR, and then calculates descriptive summaries. Those summaries are exact for the retained evidence under the selected rules. They become estimates of a broader or future population only if an additional sampling and dependence model is supplied; WSPRadar does not make that inferential step automatically.

It is useful to distinguish five levels:

Reported observations: uploaded WSPR spots with callsigns, locators, power, time and SNR.
Constructed evidence units: qualifying opportunities, peer-cycles, Joint units and scheduled A/B pairs formed by WSPRadar’s eligibility and matching rules.
Derived quantities: normalized SNR, Decode Outcomes and Target-minus-Reference Delta SNR for an individual evidence unit.
Descriptive summaries: rates, medians, reach, evidence shares and temporal or geographic summaries calculated from the retained evidence.
Interpretation beyond the run: statements about future behavior, a wider population or a physical cause. Such generalization requires additional assumptions and experimental control; the calculation alone is not sufficient.

Notation used below

Symbol	Meaning
$i$	one peer identity, defined as exact `callsign + reported locator`
$c$	one eligible WSPR cycle or, for sequential TX A/B, one scheduled pair
$g$	one retained geographic scope or segment
$b$	one distance or time bin
$S_{i,c}$	Target success indicator within an eligible Performance opportunity
$O_{i,c}$	Performance opportunity indicator after activity, identity and population rules
$D_{i,c}$	paired Target-minus-Reference Delta SNR where both sides are observed
$T_{i,b},J_{i,b},R_{i,b}$	Only Target, Joint and Only Reference counts in Benchmark scope $b$

An indicator is 1 when its condition is met and 0 otherwise. The notation makes denominators and weighting explicit; the text after each formula explains the same calculation in ordinary station terms.

This chapter uses summary or descriptive statistic for the rates, medians, shares and distributions calculated from the retained evidence. The distinction matters because a value can be calculated exactly for the retained rows while still describing a narrow or selected population.

Method orientation

Design	Lowest comparison unit	Conditioning / eligibility	Principal summary	Primary boundary
RX Performance	one remote-transmitter peer-cycle	Target receiver active; same transmitter independently decoded elsewhere	peer Decode Rate, then equal-peer mean; pooled opportunity rate retained	conditional observability, not calibrated sensitivity
TX Performance	one remote-receiver peer-cycle	Target transmitter active; peer receiver independently active on band	peer Decode Rate, then equal-peer mean; pooled opportunity rate retained	conditional observability, not all attempted transmissions
RX Hardware A/B / Buddy	one remote-transmitter peer-cycle	Target active; both receivers report the same transmitter-cycle for Delta SNR	station median Delta SNR, then median across stations	complete receive paths unless chains are controlled
Simultaneous TX Hardware A/B / applicable Buddy or Local Benchmark	one remote-receiver peer-cycle	Target active; same receiver-cycle for paired Delta SNR	station median Delta SNR, then median across stations	power, chain and joint-decode selection
Sequential TX Hardware A/B	one remote receiver in one scheduled Target/Reference pair	deterministic disjoint schedule and complete in-window pair	station median Pair Delta, then median across stations	time separation and switching/schedule effects
Local Median Neighborhood	one Target/local-Reference peer-cycle	Target active; one contribution per active local identity	local median Reference, then station/segment Delta medians	changing uncalibrated membership
Local Best Station	one Target/best-local peer-cycle	Target active; strongest qualifying local identity	best-local Reference, then station/segment Delta medians	changing envelope, not fixed competitor

The hierarchy can be read from left to right: WSPRadar first decides which evidence units belong to the analysis, then calculates a peer- or path-level quantity, and only then forms the displayed station-balanced summary. The formulas below make those steps auditable; the text following each formula explains the same operation in ordinary station terms.

7.1 Data source, observation units and time model

WSPRadar reads public WSPR reports from one selected read-only archive for each completed run. Reports are observational records produced by heterogeneous transmitters, receivers, decoders and reporting systems. A completed run does not combine data sources; the selected archive belongs to the run provenance.

A spot is one reported successful decode row. A WSPR cycle is the two-minute interval aligned to an even UTC minute. Same-cycle analyses consolidate qualifying rows by side, peer identity and cycle before classification. The effective UTC boundaries shown in the controls define the analysis window.

The lowest unit differs by design:

Performance and simultaneous Benchmark use one peer identity in one eligible WSPR cycle.
Sequential TX A/B retains exact scheduled starts, assigns them to Target or Reference by the configured modulo schedule and forms deterministic one-to-one Target/Reference pairs for each peer. Both planned starts must lie within the run window.
Local Neighborhood Benchmark additionally constructs a cycle/path Reference from qualifying local identities before forming Target-minus-Reference evidence.

These units are constructed from reported spots; they are not additional radio measurements. Their purpose is to define unambiguously the conditions under which a success, missed decode or paired difference is counted.

Historical code = 1 fallback changes the source-row selection only when the strict request has no Target-side evidence. Run status records which source path was used. Upstream delay and data-quality limitations are described in Section 5.6.

7.2 Identity, matching and row consolidation

WSPRadar treats reported identity as scientific data rather than a cosmetic label.

Analysis	Target matching	Reference / peer identity	Lowest result unit
RX Performance	exact RX callsign + Target QTH grid-4	TX callsign + full reported TX locator	Target-active peer-cycle
TX Performance	exact TX callsign + Target QTH grid-4	RX callsign + full reported RX locator	Target-active peer-cycle
Reference Station / Buddy	exact Target callsign + Target grid-4	exact Reference callsign + independent Reference grid-4; remote peer identity	consolidated peer-cycle
RX Hardware A/B	exact Target callsign + Target grid-4	exact Reference callsign + same derived Target grid-4; remote TX identity	consolidated peer-cycle
Simultaneous TX Hardware A/B	exact Target callsign + Target grid-4	exact Reference callsign + same derived Target grid-4; remote RX identity	consolidated peer-cycle
Sequential TX Hardware A/B	exact shared Target callsign + Target grid-4, split by schedule	same callsign/grid-4 on Reference schedule; remote RX identity	scheduled Target/Reference pair
Local Neighborhood Benchmark	exact Target callsign + Target grid-4	local identity inside radius; remote peer identity	Target/local-Reference peer-cycle

Target archive selection uses grid-4 even when a six-character QTH is configured. The full QTH remains relevant to distance, azimuth, solar elevation and local-radius geometry. Shared Hardware A/B grid-4 matching does not prove physical co-location.

If several qualifying non-identical rows represent one logical side/peer/cycle identity, WSPRadar retains the strongest qualifying normalized SNR as the best observed value for that logical identity. This prevents exact repeats or weaker secondary decodes from lowering the retained side value, but it is not a representative central value for one physical receiver. Different multi-receiver/reporting behavior on the two sides can therefore introduce asymmetry. Local Median Neighborhood instead forms a median within each local identity before aggregating across identities.

The local pool excludes the Target by exact callsign. A base callsign and suffixed callsign are distinct unless the exact Target form matches. Bad, stale or changing locators can split one physical station, move it geographically or trigger the moving-station exclusion.

7.3 Target-active conditioning and eligibility

Let $A_c$ indicate observable Target participation in cycle $c$:

TX: at least one qualifying Target transmission report exists somewhere in the cycle.
RX: the Target receiver uploaded at least one qualifying decode in the cycle.

Performance and simultaneous Benchmark condition on $A_c=1$. This protects known Target downtime from becoming automatic counter-evidence, but it changes the analysis population: the result describes cycles in which Target participation was observable, not all clock time or all planned attempts.

The conditioning is asymmetric. Reference uptime is not a second gate and must be controlled or documented externally. Swapping Target and Reference can therefore change eligible cycles and one-sided Decode Outcomes even when the sign of Joint-only Delta SNR reverses as expected.

Every Joint observation already implies Target participation, so the gate does not change Joint-only Delta SNR values. It changes the population of one-sided/asynchronous outcomes and, in Performance, the opportunity denominator. Sequential TX A/B uses deterministic scheduled eligibility rather than the simultaneous Target-Active Gate.

Target-active evidence may be established globally even when the peer that proves activity lies outside the selected geographic analysis scope. That peer establishes $A_c$ only; it does not enter scoped outcomes, summaries or exports.

7.4 Performance analysis target, classification and summary statistics

For peer $i$ and Target-active cycle $c$, let $O_{i,c}=1$ when independent activity evidence makes that peer-cycle a qualifying opportunity after the selected band, identity, filter and scope rules. Let $S_{i,c}=1$ when the Target also produces the required evidence in that opportunity, with $S_{i,c}\le O_{i,c}$.

RX independent activity: another eligible receiver reports the same transmitter identity in the same cycle.
TX independent activity: the peer receiver reports another same-band transmitter in the same cycle.

Target evidence without the independent activity needed for $O_{i,c}=1$ remains auditable but is excluded from Decode Rate. Within qualifying RX opportunities, WSPRadar distinguishes cycles heard by the Target from cycles heard only by other eligible receivers. Within qualifying TX opportunities, it distinguishes cycles in which the peer receiver heard the Target from cycles in which it heard only other qualifying signals on the same band.

For one qualifying peer:

$$n_i=\sum_c O_{i,c},\qquad h_i=\sum_c S_{i,c}$$

$$r_i=100%\times\frac{h_i}{n_i}$$

Here, $n_i$ is the number of qualifying opportunities retained for peer $i$, $h_i$ is the number of those opportunities in which the Target succeeded, and $r_i$ is that peer's Decode Rate. A peer contributes only when $n_i$ meets the configured minimum.

For geographic scope $g$ with qualifying peer set $I_g$, the Station-balanced Decode Rate is:

$$R_{station}(g)=\frac{1}{|I_g|}\sum_{i\in I_g} r_i$$

The Opportunity-level Decode Rate is:

$$R_{opportunity}(g)=100%\times\frac{\sum_{i\in I_g}h_i}{\sum_{i\in I_g}n_i}$$

In plain terms, the first calculates one rate per peer and then gives every peer one equal vote. The second pools all qualifying opportunities and therefore gives more influence to peers that contributed more opportunities. These are complementary summaries of the retained evidence, not competing estimates of one uniquely defined “true” Decode Rate.

At-least-once Peer Reach is:

$$Reach(g)=100%\times\frac{|{i\in I_g:h_i\ge1}|}{|I_g|}$$

The numerator counts qualifying peer identities that produced at least one Target success; the denominator counts all qualifying peers in the scope. A peer with one success and a peer with many successes both count once for Reach.

Reach is a breadth measure and normally increases with observation duration. It does not describe how consistently those peers were decoded; Decode Rate answers that separate question.

Successful Target SNR is defined only where the Target was decoded/reported. It is therefore a success-conditioned distribution. Missed opportunities have no Target SNR and no synthetic value. Decode Rate and successful SNR must be interpreted jointly because a system that adds marginal decodes can show lower successful-SNR summaries while improving practical reach.

The Performance analysis target is the Target's conditional participation among independently observable opportunities in the retained population. It is not unconditional receiver sensitivity, the success probability of every attempted transmission or absolute station efficiency.

7.5 Power normalization, correction and Benchmark Delta SNR

WSPR reports SNR on the WSJT scale in dB relative to a 2500 Hz reference bandwidth and carries reported transmit power in dBm [Ref-8]. WSPRadar normalizes successful TX-side SNR to reported 30 dBm:

$$SNR_{norm}=SNR_{measured}-P_{TX(dBm)}+30$$

In practical terms, a signal with 10 dB less reported transmit power is raised by 10 dB for this comparison. For example, an SNR reported as -15 dB at 20 dBm is normalized to -5 dB at 30 dBm. In words, the reported transmit-power difference is removed by expressing every successful TX-side SNR as though the reported power had been 30 dBm. This removes only the reported power term. It does not correct antenna gain, radiation efficiency, feedline loss, EIRP, receiver calibration or local noise.

Reference-side correction is additive:

$$SNR_{R,corr}=SNR_R+C_R$$

For a paired observation:

$$D_{i,c}=\Delta SNR_{i,c}=SNR_{T,i,c}-SNR_{R,corr,i,c}$$

Here $C_R$ is the signed additive Reference-side correction. In the first equation, $SNR_R$ and $SNR_{R,corr}$ denote Reference SNR before and after correction. In the paired equation, the indices $i,c$ identify the peer and matched evidence unit; $SNR_{T,i,c}$ is the corresponding normalized Target SNR and $SNR_{R,corr,i,c}$ is the corrected Reference SNR. This is simply corrected Target SNR minus corrected Reference SNR for one matched evidence unit. Positive $D_{i,c}$ favors the Target; negative favors the Reference. A positive correction makes the Reference stronger before subtraction and therefore lowers Delta SNR. The entered calibration offset uses the same target - reference sign.

The value $D_{i,c}$ is an observed paired difference for exactly one retained comparison unit. It is calculated exactly from the two retained SNR values and the configured correction, if any. Its interpretation still depends on what the two sides represent and how well the physical experiment controlled the remaining chains.

In same-transmitter RX pairs, the common reported TX-power term cancels. TX pairs involving different signals depend directly on reported-power accuracy and on any uncorrected transmitter/feedline difference. A Reference correction is scientifically defensible only when the offset is approximately additive and stable over the relevant band, level, hardware state and time.

7.6 Paired evidence, Decode Outcomes and missingness

Benchmark answers two linked evidence questions:

the distribution of Target-minus-Reference Delta SNR among Joint comparison units; and
the composition of retained evidence into Only Target, Joint, Only Reference and, at identity level, Both (Async).

Delta SNR exists only when both sides produce comparable evidence. The Joint subset is therefore selected on successful observation of both sides. This paired selection is not missing at random in the ordinary statistical sense: weak signals, collisions, QRM, decoder behavior, power differences and path conditions can affect whether a pair exists.

One-sided evidence has no missing-side SNR to reconstruct. It cannot be assigned an artificial Delta SNR and is not power-normalized as a pair. In TX Benchmark, unequal actual or reported powers can strongly affect one-sided outcomes even when Joint Delta SNR is normalized.

Both (Async) means that an identity has retained evidence from both sides but lacks a qualifying same-cycle or scheduled pair for the relevant station category. It indicates broader two-sided participation without contributing paired Delta SNR.

Successful-SNR censoring in Performance and Joint-decode selection in Benchmark are distinct selection processes. WSPRadar exposes Decode Outcomes and Joint Evidence Share so the paired Delta-SNR summary can be read against the wider retained evidence rather than treated as the complete station population.

7.7 Aggregation hierarchy and weighting

WSPRadar aggregates the evidence hierarchically so that one high-volume peer does not dominate station-balanced summaries solely by reporting more observations.

Performance

Classify each eligible peer-cycle.
Aggregate qualifying Target successes and counter-evidence by peer identity, while retaining Target observations without independent confirmation separately for audit.
Apply the minimum opportunity count.
Calculate one peer Decode Rate $r_i$.
Calculate the equal-peer mean $R_{station}$.
Retain $R_{opportunity}$ as the complementary volume-weighted summary.

Simultaneous Benchmark

Consolidate Target and Reference evidence by peer and cycle.
Calculate $D_{i,c}$ for Joint cycles.
Apply the minimum Joint count per peer.
Calculate the peer median:

$$m_i=\operatorname{median}{c}(D{i,c})$$
For scope $g$, calculate the station-balanced segment summary:

$$M_g=\operatorname{median}_{i\in I_g}(m_i)$$

In words, each peer is first reduced to one typical paired difference, $m_i$, and the segment summary $M_g$ is then the median across those peer values. A peer with many Joint observations therefore cannot dominate the station-balanced segment merely through volume.

The observation-level median of all $D_{i,c}$ is retained separately. It answers a different question because peers with more Joint observations receive more weight.

Sequential TX A/B

Retain exact-identity reports matching Target or Reference schedule phases.
Pair planned starts one-to-one by nearest cyclic separation under the common Repeat Interval.
Require both planned starts to lie within the run window.
Within each peer and scheduled pair, calculate a micro-median for each side.
Calculate Pair Delta when both micro-medians exist; otherwise retain the pair as one-sided evidence.
Apply the minimum complete-pair count per peer.
Calculate peer and segment medians as above.

The micro-median protects a scheduled side from duplicate-like repeated rows but does not make the two sequential transmissions simultaneous.

Local Median Neighborhood

For each remote peer-cycle, WSPRadar first calculates one normalized SNR contribution per active local callsign + locator, then takes the exact median across contributing local identities. An absent local identity is omitted rather than assigned zero. Reference correction is applied before the local pool is aggregated. The Target is compared with this cycle/path median, after which peer and segment Delta-SNR medians are calculated.

Local Best Station

For each remote peer-cycle, WSPRadar selects the strongest qualifying corrected local contribution as the Reference. The resulting Reference is a changing upper envelope. It is neither a local mean nor a comparison with one fixed station.

Medians reduce sensitivity to isolated extreme values, quantized SNR outliers and duplicate-like bursts. They do not remove systematic calibration error, propagation bias or dependence across cycles and stations.

7.8 Geographic, temporal and selected-path summaries

7.8.1 Geographic summaries

Segment Inspector starts from the complete qualifying peer population in the active retained scope; table sorting, row selection and visibility controls do not change these summaries.

Performance distance profiles group peers by exact calculated distance from Target QTH. A deterministic width of 125, 250, 500 or 1,000 km is selected from the active distance span, with edges anchored at integer multiples from 0 km and the final selected upper boundary included. Disjoint selected ranges retain missing gaps rather than treating them as zero evidence.

For each distance bin, WSPRadar calculates:

at-least-once Peer Reach;
station-balanced Decode Rate;
Opportunity-level Decode Rate; and
successful Target SNR, first reduced to one median per peer and then summarized across peer medians.

For successful-SNR spread, three or more peer medians produce an IQR, two produce a min–max interval, and one produces a single point. Counter-only peers receive no synthetic SNR. Distance inherits the precision of the reported Maidenhead locator and is not survey-grade positioning.

Benchmark geographic summaries use one peer median Delta SNR per qualifying identity and then the segment median of those peer medians. Observation-level Delta SNR remains available as a separately weighted distribution.

7.8.2 Benchmark evidence coverage

For station $i$ in bin $b$, let Only Target, Joint and Only Reference counts be $T_{i,b}$, $J_{i,b}$ and $R_{i,b}$, with:

$$N_{i,b}=T_{i,b}+J_{i,b}+R_{i,b}$$

A contributing station supplies one split support vote:

$$v_{T,i,b}=\frac{T_{i,b}}{N_{i,b}},\qquad v_{J,i,b}=\frac{J_{i,b}}{N_{i,b}},\qquad v_{R,i,b}=\frac{R_{i,b}}{N_{i,b}}$$

The station-balanced Joint Evidence Share is:

$$JES_{station}(b)=100%\times\operatorname{mean}_{i}\left(\frac{J_{i,b}}{N_{i,b}}\right)$$

The outcome-level share is:

$$JES_{outcome}(b)=100%\times\frac{\sum_iJ_{i,b}}{\sum_iN_{i,b}}$$

The station-balanced form first asks what fraction of each peer's retained units were Joint and then averages those fractions. The outcome-level form simply pools all retained units before taking the Joint fraction. The same distinction — equal peer weight versus equal evidence-unit weight — appears elsewhere in WSPRadar.

The first gives every contributing peer equal weight; the second gives every retained comparison unit equal weight. Joint Evidence Share measures pairability — the fraction of retained evidence that can contribute Delta SNR. It is not a Target win rate.

Under the Target-Active Gate, Only Target and Only Reference are directional and asymmetric. Sequential TX A/B instead uses deterministic complete or one-sided scheduled pairs, but a one-sided pair still has no Pair Delta.

7.8.3 Temporal summaries and UTC folding

Chronological views preserve the actual sequence of the run across the full selected UTC window using the selected time-bin width. Bins begin at the selected start; the final interval may be shorter, and intervals without evidence remain blank rather than becoming 0 dB. UTC-hour views fold evidence from represented dates onto fixed one-hour slots to describe recurring time-of-day structure.

Offered chronological widths are governed by the complete run duration, not by the observed evidence span: runs through 6 hours default to 10m; longer runs through 24 hours default to 30m; runs longer than 24 hours default to 12h. The offered sets are listed in Section 4.5, including the 2h choice in every duration tier. An explicitly loaded valid legacy 5m or 15m choice is preserved without making those widths normal new choices.

For Performance successful-SNR deviation, a peer enters the anomaly population only when it has at least three successful normalized Target-SNR observations in the complete run window. Its baseline is the median of those successes. Each successful observation contributes:

$$A_{i,c}=SNR_{i,c}-\operatorname{median}_{c'}(SNR_{i,c'})$$

Here $c$ identifies the current successful observation and $c'$ indexes all successful observations for peer $i$ in the complete run window that define its baseline. Thus 0 dB means “at this path's own usual successful level,” not Target–Reference equality. A positive anomaly is a stronger-than-usual successful decode for that path, and a negative anomaly is weaker than usual.

Chronologically, each peer contributes at most one median anomaly per selected bin. In the UTC-folded view, each peer contributes one median per date and UTC hour before those peer-date-hour values are summarized across the folded population. This prevents prolific peers or dates from dominating through raw row count.

Performance temporal support uses the same qualifying peers but retains all confirmed opportunities, including peers omitted from the successful-SNR anomaly layer. In a chronological bin, each peer contributes one split vote according to its within-bin Decode Rate. The station-support total is therefore the number of contributing peers, while the split ratio reproduces the station-balanced rate. The opportunity-support total is the raw confirmed-opportunity count, and its split ratio reproduces the Opportunity-level rate.

For each folded UTC hour, station support is the average number of distinct peer-date-hour presences over represented dates whose hour slot overlaps the analysis window. The folded station-balanced rate is calculated by pooling each peer's outcomes at that UTC hour across represented dates, calculating one rate per peer, and then giving each peer equal weight. Folded opportunity counts are pooled outcome totals divided by the corresponding represented-date denominator.

For Performance, a represented UTC date is a date with at least one qualifying confirmed opportunity somewhere in the active scope and selected window. A represented date-hour inside the window contributes zero when it has no evidence; a date-hour outside the window is excluded. A partially overlapping first or last hour counts as one represented slot rather than receiving exposure weighting, so boundary-hour averages can be depressed. UTC-hour folding requires at least two represented dates.

Benchmark temporal Delta SNR uses retained Joint observations or complete scheduled pairs. If no paired values remain, the Delta SNR panel still shows the full selected UTC window and states that paired Δ SNR evidence is absent. This means no retained Joint observation or complete scheduled pair remains in the displayed scope; it does not by itself mean that the data source returned no observations, and temporal coverage can still show one-sided outcomes. Chronological bins summarize raw paired values in actual time; UTC-hour bins summarize the same paired population by hour across dates represented by retained Benchmark evidence. Benchmark temporal coverage uses all retained Only Target, Joint and Only Reference units and the two Joint Evidence Share summaries above. Benchmark folding likewise requires at least two represented evidence dates.

7.8.4 Selected-path summaries

Selected Station Evidence filters the active retained scope to one exact peer identity without changing the upstream analysis population.

For Performance, the selected path reports:

actual normalized successful Target SNR in chronological bins;
one date-hour median per represented date in the folded SNR profile;
successful/counter opportunity counts; and
Decode Rate through time.

With one peer, station-balanced and Opportunity-level Decode Rate are numerically identical within a populated bin; the separate support counts still distinguish path presence from evidence volume.

For Benchmark, the selected path reports observation-level Delta SNR for each Joint unit or complete scheduled pair and separately reports Only Target, Joint and Only Reference coverage. Changing the selected path or display bin changes only the retained-evidence view, not matching, eligibility or aggregation upstream.

Drill-Down can temporarily restrict this same selected-path evidence to one centered 1h, 3h, 6h, 12h or 24h interval before ordinary table filters run. Its focused metric recipe retains one scientific unit at its native coordinate: one consolidated Joint Spot and actual Delta SNR at canonical cycle UTC for simultaneous Benchmark; one complete Scheduled Pair and actual Pair Delta SNR at planned Target-start UTC for sequential TX A/B; or one successful confirmed opportunity and actual normalized Target SNR at canonical cycle UTC for Performance. Thus “native” describes processed retained evidence after consolidation, matching and scientific filters, not untouched provider rows. The focused metric recipe contains neither temporal-bin medians or quartiles nor a density grid, colorbar, full-run median or folded profile. Companion outcome/coverage panels may retain their chronological aggregation, and the segment and full-window selected-path recipes remain unchanged density summaries. Sequential membership still keeps both Target and Reference component rows of each admitted pair together in the table.

Candidate-linked Outlier Focus uses the full retained pre-event flank, guarded provisional episode and post-event flank and may exceed 24 hours. The Benchmark overlay uses the already completed detector model rather than redetecting from the focused subset. For every reported candidate intersecting the focus window, it marks with the same * each native unit that individually meets both $D_{\min}$ and $Z_{\min}$ against that candidate's final baseline and robust spread; weaker grouped units retained between strong anchors remain ordinary dots. A muted Focused episode band distinguishes the selected candidate's reported retained-evidence interval. The renderer pads each end by half one native evidence-unit width and clips the band to the focused window, making a one-unit impulse visible without representing unobserved physical duration or a confidence interval. Expected local Delta SNR across the focus, the pre/post flank medians over their respective support intervals, symmetric guide boundaries for robust-z magnitudes 1, 2, 3 and $Z_{\min}$, and the absolute-departure boundary $D_{\min}$ all belong only to that focused episode; another starred candidate can have a different baseline and robust spread. From the detector definition in Section 7.11, a guide of magnitude k lies at the local baseline plus or minus k × robust spread / 0.6745. These guides visualize detector coordinates; they are not standard deviations, confidence intervals or independent qualification tests, and crossing one guide alone is insufficient to qualify a candidate. Focus selection and candidate provenance are presentation state only; they do not alter AnalysisContext, matching, eligibility, detector results, the provider query, saved configuration or public URL.

7.8.5 Descriptive spread and visualization transforms

IQR and min–max displays are descriptive spread summaries, not confidence intervals. An IQR band is drawn only where at least five values contribute to the relevant bin; the median remains available with fewer values. Empty bins remain missing rather than becoming synthetic zero observations.

Benchmark histograms normally use 1 dB bins, use 0.5 dB only for a clear half-dB lattice, and coarsen broad ranges to keep the number of bins bounded. Temporal density cells use integer-dB classes. Each density panel is normalized independently:

$$D_{relative}=100\times\frac{n_{cell}}{\max(n_{cell,panel})}$$

Here $n_{cell}$ is the evidence count in one density cell. Dividing by the most populated cell converts the panel to a relative-density display while leaving the underlying counts unchanged.

Thus 100 means the most populated cell in that panel, not 100% of all evidence. Density colors cannot compare absolute evidence volume between independently normalized panels; support counts provide that information.

Benchmark temporal and histogram views use a presentation-only monotonic scale centered on the scope median $M$. For a broad range, equal visual steps are anchored at $M$, $M\pm3$, $M\pm6$, $M\pm10$, $M\pm20$ and $M\pm30$ dB, with a tail anchor at $M\pm60$ dB and extrapolation when required. When every required deviation is at most 10 dB, the tighter anchors are $M$, $M\pm1$, $M\pm3$, $M\pm6$ and $M\pm10$ dB, with continuation anchors at $M\pm20$ and $M\pm40$ dB. The required range includes the applicable raw histogram or rounded heatmap-bin edges, a minimum 3 dB half-span and absolute 0 dB, so Target–Reference equality remains visible. The anchor mapping changes displayed spacing only: raw Delta SNR values, bin membership, counts, medians and quartiles remain unchanged. Because the vertical mapping is nonlinear, histogram bar length against its percentage axis — not displayed area — is the quantitative encoding.

Performance successful-SNR views remain on a linear dB axis.

7.9 Geography, solar classification and population filters

Distance and azimuth are calculated from the configured Target QTH and reported peer locators using a spherical Earth radius of 6371 km. The map uses an azimuthal equidistant projection centered on Target QTH, with radial boundaries at 2500, 5000, 10000, 15000, 20000 and 22000 km and 22.5-degree azimuth sectors.

Reported locators represent grid cells rather than measured antenna coordinates. Geographic summaries are internally consistent with those inputs but should not be interpreted as survey-grade position or direct take-off-angle measurement.

Maximum peer distance from Target (km) removes peers at or beyond the selected distance before scientific aggregation and processed-evidence export. Map segments, support counts, Segment Inspector and exports therefore use one retained peer population. Inspector selections can narrow that population but cannot restore excluded rows.

Two rules precede the geographic scope:

Target-active conditioning remains global, so an out-of-scope peer can prove Target operation without becoming a scoped outcome.
When moving-station exclusion is enabled, changing-location callsigns are identified in the otherwise eligible global population before distance scope is applied.

Solar classification uses solar elevation at Target QTH. Same-cycle evidence uses the cycle timestamp. Scheduled TX A/B uses the midpoint between the planned Target and Reference starts so one pair cannot be split across solar classes.

The archive row limit and the controls that can reduce the retrieved source population are operational matters documented in Section 5.6; they do not change the scientific summaries after the retained population has been formed.

7.10 Dependence, uncertainty and validation scope

WSPRadar observations are clustered rather than independent. In ordinary station terms, 1,000 spots are not the same as 1,000 unrelated experiments. Repeated cycles from one peer share hardware and path characteristics; stations in nearby regions share propagation; time bins are autocorrelated; and one ionospheric or interference event can affect many observations simultaneously. A large row count is therefore not an independent sample size.

Station balancing reduces domination by prolific peers, and medians reduce sensitivity to isolated outliers. Neither creates independence, removes systematic bias nor supplies a sampling distribution. IQRs describe within-run spread and are not uncertainty intervals.

WSPRadar currently reports descriptive summaries of retained evidence. It does not automatically calculate standard errors, confidence intervals, p-values, statistical power or causal effects. The summaries can be exact for the retained evidence while uncertainty about a wider population or future run remains unresolved. Naive inferential calculations that treat every spot or pair as independent would generally understate that uncertainty.

Scientific support should therefore be described at several levels:

evidence depth: number of opportunities, Joint units or scheduled pairs;
evidence breadth: number and geographic diversity of peer identities;
within-run consistency: agreement across station-balanced, observation-level, geographic and temporal summaries;
experimental repeatability: recurrence in a new suitably controlled run; and
experimental control: calibration, crossover, reversed schedule or independent measurement appropriate to the claim.

Empirical software-validation audits are not timeless method definitions. Any reported validation statistic should identify its datasets, date, WSPRadar version or source revision, and calculation method. Without that provenance it should be removed from the normative manual or labelled explicitly as a dated validation check.

7.11 Robust local-baseline Delta SNR event detection

The detector's analysis target is a temporary same-sign departure in one path's paired Delta SNR from a stable local expected value. It is not a ranking of the largest raw values and does not estimate an event probability. Section 2.5 explains when and how an operator should use the diagnostic; this section defines the exact scientific construction.

The notation is local to this section except for the three control symbols already introduced in Section 4.6: $D_{\min}$ is Minimum absolute ΔSNR departure (dB), $Z_{\min}$ is Minimum robust z-score, and $H_{\max}$ is Maximum pre/post baseline difference (dB). No notation elsewhere in Chapter 7 is redefined.

Symbol	English mnemonic and meaning in this section
$i$	one exact peer `callsign + locator` path identity
$u$	one native paired unit: a same-cycle Joint Spot or complete Scheduled Pair
$k$	one UTC-aligned 10-minute baseline-cell index
$D_{i,u}$	corrected paired Target-minus-Reference Delta SNR for unit $u$, following Section 7.5
$\widetilde D_{i,k}$	median Delta SNR in populated baseline cell $k$
$\mathcal{B}{\mathrm{pre}},\mathcal{B}{\mathrm{post}}$	retained pre-event and post-event Baseline evidence values
$B_{\mathrm{pre}},B_{\mathrm{post}},B$	pre-event, post-event and final local Baseline
$B^P_{i,k}$	Pilot Baseline for path $i$ and cell $k$
$r^P_{i,u},r_{i,u}$	Pilot and final Residual for native unit $u$
$\mathcal{V},S_{\mathrm{robust}}$	centred flank Variability sample and robust local Scale
$z_{i,u}$	robust z-score of one native paired unit
$C_i,G_i,F$	typical path Cadence, maximum internal Gap and provisional grouping Floor
$W_i^P,W_i^B$	Pilot- and final-Baseline exclusion Widths
$E,m_E,z_E,\operatorname{agree}(E)$	candidate Event, its Median residual, event robust z-score and sign-agreement fraction

1. Evidence and resolution. Only native paired units supply detector Delta SNR. One-sided outcomes have no paired value and cannot qualify an event, although their times contribute to cadence estimation and the outcomes remain diagnostic context. Detection runs independently for each path $i$ before Temporal Evidence display aggregation. Changing a display bin cannot create, merge, split or remove an event. When Report ΔSNR outlier candidates is off, the detector is not run and no outlier semantics are added to the result.

Native values are reduced to $\widetilde D_{i,k}$ for baseline and robust-scale estimation. Candidate grouping and reported boundaries retain the native paired-unit times.

2. Stable local baseline and robust variability. For the candidate under evaluation, the detector examines populated 10-minute cells up to six hours before and six hours after it. The candidate and its exclusion width are omitted. Each flank must contain at least four populated cells; missing cells are never imputed. The retained cell values form $\mathcal{B}{\mathrm{pre}}$ and $\mathcal{B}{\mathrm{post}}$:

$$ B_{\mathrm{pre}}=\operatorname{median}(\mathcal{B}_{\mathrm{pre}}),\qquad B_{\mathrm{post}}=\operatorname{median}(\mathcal{B}_{\mathrm{post}}) $$

The final expected local Delta SNR gives the two flanks equal weight, while the stability gate limits their disagreement:

$$ B=\frac{B_{\mathrm{pre}}+B_{\mathrm{post}}}{2},\qquad \left|B_{\mathrm{pre}}-B_{\mathrm{post}}\right|\leq H_{\max} $$

Equal flank weighting prevents the side with more populated cells from dominating. If either flank lacks support or the stability gate fails, the path remains unclassified for that candidate and no event is reported.

To measure nearby variability without treating a baseline shift as noise, each flank is centred on its own baseline:

$$ \mathcal{V}=\left{x-B_{\mathrm{pre}}:x\in\mathcal{B}_{\mathrm{pre}}\right} \cup\left{x-B_{\mathrm{post}}:x\in\mathcal{B}_{\mathrm{post}}\right} $$

The robust local scale is:

$$ S_{\mathrm{robust}}= \begin{cases} \operatorname{MAD}(\mathcal{V}), & \operatorname{MAD}(\mathcal{V})>0,\\ \frac{1}{2}\operatorname{IQR}(\mathcal{V}), & \operatorname{MAD}(\mathcal{V})=0\ \land\ \operatorname{IQR}(\mathcal{V})>0,\\ 0.5\ \mathrm{dB}, & \operatorname{MAD}(\mathcal{V})=0\ \land\ \operatorname{IQR}(\mathcal{V})=0. \end{cases} $$

MAD is the median absolute deviation and IQR is the interquartile range. The 0.5 dB floor prevents division by zero for locally quantized evidence. Against the final baseline, one native unit has:

$$ r_{i,u}=D_{i,u}-B,\qquad z_{i,u}=0.6745\frac{r_{i,u}}{S_{\mathrm{robust}}} $$

A positive residual is above the expected local Target-minus-Reference Delta SNR and a negative residual is below it; this residual sign, not the sign of raw $D_{i,u}$ relative to 0 dB, drives grouping and qualification. The factor 0.6745 supplies conventional modified-score scaling when MAD is active. The score remains descriptive rather than a calibrated probability, p-value or Gaussian significance level.

3. Cadence-aware pilot grouping. The typical path cadence $C_i$ is estimated in minutes from unique eligible paired and one-sided outcome times. Positive intervals no longer than 45 minutes are retained, and at least two are required; otherwise the configured paired-unit cadence supplies $C_i$. The maximum internal gap is:

$$G_i=\min\left(45,\max\left(15,1.5C_i\right)\right)\ \mathrm{minutes}$$

Thus sparse paths receive a cadence-aware grouping allowance, but no event can bridge more than 45 minutes and no estimated cadence reduces the allowance below 15 minutes.

To seed candidates without allowing the point under test to define its own expected value, each supported cell receives a Pilot Baseline $B^P_{i,k}$. Its exclusion width and the permissive grouping Floor are:

$$ W_i^P=\max(60,2C_i)\ \mathrm{minutes},\qquad F=\min(1\ \mathrm{dB},D_{\min}) $$

The pilot fit uses the same six-hour flank reach, four-populated-cell minimum on each side and $H_{\max}$ stability gate. A native unit $u$ in cell $k(u)$ receives a pilot residual only when that cell has supported flanks:

$$r^P_{i,u}=D_{i,u}-B^P_{i,k(u)}$$

Nearby pilot residuals of the same sign and magnitude at least $F$ form a provisional event. Grouping stops at a gap greater than $G_i$, an opposite-sign member or a supported return to baseline. One supported neutral unit may bridge the evidence when the same-sign departure resumes; a second consecutive supported neutral unit ends the event. Unsupported units remain unclassified and are not imputed. The low Floor finds continuity; it does not relax final qualification.

4. Candidate-excluded final baseline. The complete provisional event is excluded from the final baseline fit with width:

$$W_i^B=\max(10,C_i)\ \mathrm{minutes}$$

The final fit uses the baseline and robust-scale construction in Step 2. Against that shared final baseline, same-sign shoulders with residual magnitude at least 1 dB may expand the event for up to three passes. The complete expanded interval remains excluded at every refit, preventing the candidate from pulling its own expected value toward the excursion.

5. Qualification, strong-core rescue and reported boundaries. For one refined candidate Event $E$, define its Median residual, event robust z-score and sign-agreement fraction as:

$$ m_E=\operatorname{median}_{u\in E}(r_{i,u}),\qquad z_E=0.6745\frac{m_E}{S_{\mathrm{robust}}} $$

$$ \operatorname{agree}(E)= \frac{\left|\left{u\in E:\operatorname{sign}(r_{i,u})=\operatorname{sign}(m_E)\right}\right|}{|E|} $$

The event qualifies only when every gate holds:

$$ |m_E|\geq D_{\min},\qquad |z_E|\geq Z_{\min},\qquad \left|B_{\mathrm{pre}}-B_{\mathrm{post}}\right|\leq H_{\max},\qquad \operatorname{agree}(E)\geq\frac{2}{3} $$

Every duration and eventual class uses these same gates. There is no duration bonus, accumulated-evidence discount or weaker threshold for a sustained event.

If the complete refined event cannot yield a qualifying strongly anchored interval, it is split at final-baseline returns into contiguous same-sign sections without neutral bridging. Each section is tested against the same gates while reusing the complete candidate's final baseline, robust scale and flank support. This can rescue a strong core suppressed by weak surroundings without allowing the smaller section to select a more favorable reference.

After qualification, a native unit is a strong boundary anchor only when:

$$ \operatorname{sign}(r_{i,u})=\operatorname{sign}(m_E),\qquad |r_{i,u}|\geq D_{\min},\qquad |z_{i,u}|\geq Z_{\min} $$

The reported interval is trimmed to the first and last strong anchors. Units already grouped between them remain internal evidence even when they individually miss an anchor threshold. The trimmed interval is rebuilt and retested against every event-level gate using the same final baseline, robust scale and flank support. Weak leading and trailing units are removed; weaker internal bridges can remain. One surviving anchor becomes a Spot impulse. If no strongly anchored interval passes, no event is reported. The lower grouping Floor and higher boundary requirements therefore produce hysteretic event boundaries.

The all-path temporal marker is selected only from these strong anchors: within each cross-path review event, the individually qualifying native unit with the greatest absolute residual supplies the *. An unsupported or nonqualifying episode peak cannot become that plot representative. Marker selection is separate from Largest single-cycle departure, which remains the true greatest-absolute retained residual in each path event for the report and export.

6. Descriptive class and cross-path context. Classification occurs after trimming:

Spot impulse: one retained native paired unit;
Sustained excursion: at least three retained native paired units spanning at least 30 minutes; and
Short burst: every other retained multi-unit event.

The classes describe temporal evidence shape and do not change qualification. The displayed first-to-last interval spans observed retained units; it does not establish uninterrupted behavior between them.

Path events qualify independently. Same-sign events are placed in one review card when they overlap or lie within the greater of 10 minutes and half the configured paired-unit cadence. Context is path-specific for one path, directionally coherent for several paths in adjacent compass sectors, scope-wide for several separated directions, or multiple paths when direction is unavailable for at least one contributor. Cross-path context does not alter path qualification and is not an independence or significance calculation.

Target/Reference decomposition and nearby one-sided outcomes are retained as internal diagnostics and can be investigated through the paired evidence and Drill-Down. They do not qualify, extend or strengthen an event. The detector is therefore a deterministic descriptive classifier of the retained paired evidence. Its event count is not an independent sample size, it performs no multiple-event significance correction, and causal attribution still requires the experiment control described in Chapters 2, 3 and 8.

8. Evidence-Matched Claims and Reproducibility

WSPRadar supports bounded descriptive and comparative claims about retained observational evidence. Strong reporting identifies the conditioned population, reported summary and weighting, support, experiment design and remaining unobserved or uncontrolled variables.

8.1 Claim classes and evidence-matched wording

Claim class	What WSPRadar can support	Additional requirement for a stronger claim
Descriptive	Reach, Decode Rate, successful SNR, Delta SNR, Decode Outcomes and where they appeared in the selected evidence.	State the population, weighting, scope and support.
Comparative	Target-versus-Reference difference under the selected Benchmark design.	State what the Reference represents and the matched subset.
Component attribution	A difference associated with a local path or component.	Controlled Hardware A/B, calibration and preferably crossover/reversal.
Causal	The tested change caused the observed effect.	A design that controls plausible alternatives; WSPRadar summaries alone are insufficient.
Inferential	Confidence, significance or a population-general effect.	A justified dependence model and inferential analysis not currently supplied by WSPRadar.

Use the result type that matches the statement:

Performance supports the Target's conditional behavior within independently confirmed opportunities and its at-least-once reach during the selected window.
Benchmark Delta SNR supports paired Target-minus-Reference description within the Joint subset.
Decode Outcomes support statements about pairability and one-sided evidence.
Distance or direction structure supports statements about observed path segments, not direct radiation angle or gain pattern.
Local Neighborhood Benchmark supports statements relative to the selected dynamic local definition, not a permanent station ranking.

Avoid	Evidence-matched wording
“Antenna A has 3 dBi more gain.”	“Path A produced a +3.0 dB station-balanced median Delta SNR against B for the paired evidence in this band, window and segment.”
“My receiver sensitivity is 72%.”	“The Target receiver's station-balanced Decode Rate was 72% among qualifying peer-cycles independently confirmed elsewhere.”
“Performance should be close to 100%.”	“Decode Rate is conditional on independently confirmed opportunities; 100% is not an expected baseline.”
“A is statistically significantly better.”	“The descriptive paired median favored A in the selected evidence; no significance test was performed.”
“The antenna has a lower take-off angle.”	“The observed advantage was concentrated in the specified longer-distance segments; radiation angle was not measured.”
“A is more efficient because it had more exclusive decodes.”	“A produced more one-sided decode evidence under the documented power, schedule and network conditions; efficiency was not isolated.”
“The local median is the average local station.”	“The Reference was the cycle/path median of one contribution per active local callsign-plus-locator identity.”

8.2 Interpretation boundaries

WSPRadar does not directly measure:

antenna gain in dBi or radiation efficiency;
take-off angle or propagation mode;
calibrated receiver sensitivity or absolute field strength;
every attempted transmission or complete failure log;
independent sample size, confidence intervals or statistical significance; or
causation.

Important data and design boundaries include:

user-supplied callsigns, locators and powers can be wrong;
archives contain successful decodes rather than complete attempt logs;
Performance is conditioned on independently observable opportunities;
Target-active conditioning is asymmetric;
successful Target SNR is censored to successful decodes;
Benchmark Delta SNR is selected on Joint observation of both sides;
one-sided evidence has no missing-side SNR;
simultaneous TX retains power, frequency-response, isolation and coupling differences between chains;
sequential TX remains time-separated;
station hardware, software, terrain, local noise, polarization and propagation remain coupled unless the experiment controls them;
observations are clustered across station, time, geography and propagation; and
upstream records and availability can change after the original run.

These boundaries define what the summaries describe; they do not make the observations useless. Broad, internally consistent and experimentally repeatable evidence can be operationally persuasive while remaining descriptive.

8.3 Reporting and reproducibility checklist

For a serious analysis, preserve three layers.

1. Analysis definition

WSPRadar application version and, where available, source revision;
RX/TX Direction, result type and Benchmark design;
exact Target and Reference identities and locators;
band and effective UTC boundaries;
geographic scope, solar state, exclusions and evidence thresholds;
Reference correction purpose, signed value and calibration basis;
primary predeclared evaluation scope and any sensitivity analyses; and
whether the run was exploratory or confirmatory.

When a Delta SNR outlier candidate contributes to the conclusion, also record that reporting was enabled, the three detector thresholds, detector version, exact path, UTC interval, descriptive event class and whether the event was identified exploratorily or assessed under a predeclared confirmatory setup.

2. Evidence supporting the conclusion

reported summary and weighting level;
qualifying peers and opportunities for Performance;
Joint peers and Joint spots/pairs for Benchmark;
station-level and observation-level summaries;
Joint Evidence Share and relevant one-sided Decode Outcomes;
geographic/temporal scope and any influential identity or short interval; and
within-run consistency versus repetition in a separate run.

3. External experiment record

physical antenna, feedline and RF-path arrangement;
switch/splitter topology and identity-to-path mapping;
transmitter, receiver, decoder and software versions;
actual transmit power and WSPR reporting basis;
calibration measurements and reference plane;
actual schedule, interruptions, crossovers and reversed assignments; and
faults, interference, weather or intended changes relevant to interpretation.

Retain the original export package as the evidence record for that run. A later retrieval can reflect upstream corrections or a newer WSPRadar version.

8.4 Analysis export package

Prepare All Results for Download builds a package from the completed run and current inspection selections. A typical package contains:

config/
  wspradar_config.config
  run_metadata.json
benchmark/
  figure_map_highres.png
  figure_segment_insight.png
  figure_segment_temporal_evidence.png
  figure_segment_temporal_coverage.png
  figure_selected_station_evidence.png
  figure_selected_station_coverage.png
  figure_drilldown_zoom_delta_snr_evidence.png
  figure_drilldown_zoom_coverage.png
  table_station_insights_current_segment.csv
  table_drilldown_selected_stations.csv
  table_drilldown_all_stations_current_segment.csv
  table_delta_snr_outlier_event_paths.csv
  table_delta_snr_outlier_paired_evidence.csv
  analysis_cache.parquet
performance/
  figure_map_highres.png
  figure_segment_insight.png
  figure_segment_temporal_snr_deviation.png
  figure_segment_temporal_evidence.png
  figure_selected_station_snr_evidence.png
  figure_selected_station_temporal_evidence.png
  figure_drilldown_zoom_snr_evidence.png
  figure_drilldown_zoom_temporal_evidence.png
  table_station_insights_current_segment.csv
  table_drilldown_selected_stations.csv
  table_drilldown_all_stations_current_segment.csv
  analysis_cache.parquet

Files without an applicable result or selected station can be absent.

Artifact	Scientific content and scope
`wspradar_config.config`	Versioned runnable definition and durable result-view settings.
`run_metadata.json`	Application/export provenance, Direction, band, time selection, Benchmark/correction definition, filters, thresholds and inspection selections.
`analysis_cache.parquet`	Processed retained evidence after scientific filters and geographic scope; not an untouched upstream dump.
`table_station_insights_current_segment.csv`	Per-peer summaries for the active Segment Inspector scope.
Drill-Down CSV files	Row-level retained evidence for selected or active-scope identities.
Delta-SNR outlier CSV files	When optional outlier reporting is enabled, qualified path-event summaries and their chronological native paired evidence for the active Segment Inspector scope.
Map and segment figures	Geographic and segment-level descriptive summaries for the completed result.
Temporal figures	Chronological and UTC-folded summaries for the active segment.
Selected-station figures	One exact selected peer identity in normal use; while optional Delta SNR outlier reporting enables an ordered multi-path selection, Benchmark can instead export the corresponding pooled multi-path Delta SNR view.
Drill-Down focus figures	Optional native-time metric and chronological companion figures for the exact selected station and active manual or candidate-linked focus interval; they supplement rather than replace the full-run selected-station figures.

When a Drill-Down focus is active at preparation time, Performance can add figure_drilldown_zoom_snr_evidence.png and figure_drilldown_zoom_temporal_evidence.png; Benchmark can add figure_drilldown_zoom_delta_snr_evidence.png and figure_drilldown_zoom_coverage.png. Each title uses only the selected callsign (locator) followed by - Time Window: {start} to {end} UTC. The metric figure preserves the same individual retained native points as the browser focus rather than substituting temporal medians; the companion figure preserves the applicable chronological outcome or coverage recipe. run_metadata.json records one drilldown_zoom block with its schema version, callsign, locator, exact start_utc and end_utc, selected focus option, manual or outlier_focus origin and render contract. When a candidate overlay is present, its registered recipe and signature preserve the focused episode and every individually qualifying candidate unit in the window, the muted focused-episode band, local and pre/post baseline values, robust spread and method, robust-z threshold and absolute-departure threshold used by the exported guides. The guide coordinates remain specific to the focused episode even when another starred candidate in the exported window was assessed against another baseline or spread. No focus block or focus figure is included while focus is off or the one-station requirement is not met.

When Delta SNR outlier reporting is enabled, table_delta_snr_outlier_event_paths.csv contains one row per qualified path event. It records the combined review-event class and observed UTC bounds, cross-path context and departure direction, then identifies the exact path, direction, path-event class, paired-evidence timing and count, expected and observed Delta SNR, largest departure, robust score, pre/post baseline diagnostics and nearby Decode Outcome diagnostics. One exact path can occur in more than one row of the same combined review event when it contributes more than one qualifying timeframe. The qualifying-path count remains the number of distinct callsign + locator identities.

table_delta_snr_outlier_paired_evidence.csv contains one row per retained native paired unit inside those path events, in chronological order. Event ID and Path event ID link it to the summary. For stand-alone readability it repeats the observed event bounds, path, direction and path-event class; the combined event class remains only in the summary because it can differ from an individual path's class. The table includes Target SNR, already-corrected Reference SNR, Delta SNR, expected local Delta SNR, departure from that baseline, per-unit robust score, strong-anchor status and reported-boundary role. Timestamps use ISO UTC and numeric fields remain numeric rather than embedding signs or units in the cells.

The IDs are deterministic joins within one prepared package, not persistent identities across separately prepared analyses and not assertions that a physical event occurred. If reporting is enabled but no path event qualifies, both CSVs remain present with their headers and no invented finding row. run_metadata.json records the enabled state, detector version, three-threshold policy, export-schema version, result status, table filenames and event, path and evidence counts once. Applicable exported Benchmark figures retain the same candidate-marker recipe metadata. When reporting is disabled, both CSVs, all outlier metadata and all marker recipes are omitted entirely.

Selected public machine-readable contract names. This concise table identifies supported external names useful to operators and downstream consumers; it is not an exhaustive saved-configuration field, URL-parameter or export-metadata catalog. The formal JSON Schema (config/wspradar-config.schema.json) is authoritative for saved-configuration fields, while the supported public URL contract is versioned separately. Private implementation identifiers are deliberately omitted. These names are not vocabulary for explaining the scientific method.

Contract surface	Exact names	Meaning
Configuration format	schema version `1`	Current pre-production `.config` contract; invalid or unsupported files are rejected rather than silently reinterpreted.
Result-type values	`performance`, `benchmark`	Values emitted by new analysis URLs, configurations and exports.
Durable result-view blocks	`results_view.performance`, `results_view.benchmark`	Saved inspection preferences. Their presence does not create or run an additional result.
Result folders	`performance/`, `benchmark/`	Top-level result folders in the export package.
Outlier tables	`table_delta_snr_outlier_event_paths.csv`, `table_delta_snr_outlier_paired_evidence.csv`	Enabled-only Benchmark tables linked by package-local Event ID and Path event ID; absent when Delta-SNR outlier reporting is disabled.
Figure metadata	`selected_evidence_figures`, `benchmark_evidence_figures`, `benchmark_evidence_recipes`, `drilldown_zoom`	Stable mappings and exact transient zoom identity/bounds for applicable exported figures and Benchmark recipes.
Correction metadata	`benchmark_snr_correction_mode`, `benchmark_snr_correction_db`	The semantic correction choice and its numeric dB value.

The export package preserves the processed evidence and provenance recorded by WSPRadar. It does not contain authoritative external operating logs, physical setup measurements or unchanged upstream responses. Preserve those separately as described in Section 8.3.

8.5 Disclaimer

WSPRadar is experimental open-source software provided “as is” without warranty. Its source and methods can be audited, but accuracy, completeness, availability and fitness for purpose are not guaranteed. Do not base major financial or safety decisions on WSPRadar alone.

References

[Ref-1] Preprint. Zander, J. (2022). Simple HF antenna efficiency comparisons using the WSPR system. arXiv:2209.08989v1. doi:10.48550/arXiv.2209.08989.
[Ref-2] Peer-reviewed article. Vanhamel, J.; Machiels, W.; Lamy, H. (2022). Using the WSPR Mode for Antenna Performance Evaluation and Propagation Assessment on the 160-m Band. International Journal of Antennas and Propagation, 2022, 4809313. doi:10.1155/2022/4809313.
[Ref-3] Operator technical account. Toledo, S. / 4X6IZ (2010). Failure to Use WSPR to Compare Antennas.
[Ref-4] Amateur-radio technical article and club presentation. Milazzo, C. F. / KP4MD (2011). Using the Weak Signal Propagation Reporter Network to Compare Antenna Performance.
[Ref-5] Amateur-radio magazine article. Griffiths, G.; Squibb, N. J. (2017). Improving HF Band SNR from analysis of WSPR spots. Practical Wireless, October 2017, 23-26.
[Ref-6] Taylor, J. H.; Walker, B. (2010). WSPRing Around the World. QST, 94(11), 30-32.
[Ref-7] Peer-reviewed review article. Frissell, N. A. et al. (2023). Heliophysics and amateur radio: citizen science collaborations for atmospheric, ionospheric, and space physics research and operations. Frontiers in Astronomy and Space Sciences, 10, 1184171. doi:10.3389/fspas.2023.1184171.
[Ref-8] Official technical overview. ARRL, WSPR: message format, coding, duration, timing, occupied bandwidth and SNR reference. Accessed 2026-07-12.
[Ref-9] Peer-reviewed article. Lo, S.; Rankov, N.; Mitchell, C.; Witvliet, B. A.; Jayawardena, T. P.; Bust, G.; Liles, W.; Griffiths, G. (2022). A Systematic Study of 7 MHz Greyline Propagation Using Amateur Radio Beacon Signals. Atmosphere, 13(8), 1340. doi:10.3390/atmos13081340.
[Ref-10] Official data-service documentation. WSPR.live, Welcome to WSPR Live: database access, schema, mode-code mapping, raw-data and availability disclaimer, and data-update behavior. Accessed 2026-08-06.
[Ref-11] Official project website. WSPRDaemon, WSPR Daemon: multi-channel spot acquisition, WSPR/FST4W decoding and reporting, noise estimation, database/Grafana output, and services for third-party applications. Accessed 2026-08-06.
[Ref-12] Official operating documentation. WSJT-X 3.0.1 User Guide: WSPR message formats and decoder performance; Windows --rig-name file isolation; Audio settings and file locations. QRP Labs, QMX Operating Manual, firmware 1_03_000: Beacon Frame and Start scheduling and WSPR repetition guidance; Ultimate3S Operating Manual, firmware v3.12a: global Frame/Start behavior, sequential mode entries and per-entry Aux values; AN003: Ultimate3/3S relay-switched filters: filtered relay/driver interfacing and RF-off switching intervals. Accessed 2026-07-15.
[Ref-13] Conference paper. Griffiths, G.; Robinett, R. (2020). Aids to the Presentation and Analysis of WSPR Spots: TimescaleDB database and Grafana. ARRL/TAPR Digital Communications Conference 2020.
[Ref-14] Tool documentation. WSPR.Rocks, Help & Documentation: SpotQ, SQL access, duplicate analysis, maps, charts and heatmaps.
[Ref-15] Product documentation. SOTABEAMS, WSPRlite Classic / DXplorer: WSPR-based antenna-performance analysis and DX10 metric.
[Ref-16] Project documentation. WSPR-Station-Compare, project page referencing Vanhamel et al. and Zander.
[Ref-17] Tool documentation. Antenna Performance Analysis Tool, WSPR-based antenna report generator.
[Ref-18] Tool documentation. GM4EAU, WATT WSPR Analysis Tool: Excel/VBA reporting, mapping, filtering and timeline animation.

Part IV: Practical Supplements

This part collects optional parallel WSJT-X and simultaneous-TX setup procedures, sequential TX A/B scheduling and switching guidance, Reference-side calibration and the project license. Use the sections that apply to your station and experiment.

Appendix A: Parallel WSJT-X Instances

This procedure creates a second isolated WSJT-X instance, for example for simultaneous RX or TX Hardware A/B Test on Windows. The current WSJT-X guide documents --rig-name as the supported way to isolate each instance's settings and writable files. WSJT-X versions and installation paths can change, so verify the current guide if your menus differ. [Ref-12]

A.1 Create the second instance

Create a desktop shortcut for wsjtx.exe.
Open shortcut properties.
In the shortcut's Target field, add a distinct rig name outside the executable quotation marks. Use the actual executable path from your installation, for example: "C:\WSJTX\bin\wsjtx.exe" --rig-name=SDR
Start the shortcut once and close it. For --rig-name=SDR, Windows creates these isolated locations:
- settings: %LOCALAPPDATA%\WSJT-X - SDR\WSJT-X - SDR.ini
- log/writable directory: %LOCALAPPDATA%\WSJT-X - SDR\
- default saved-audio directory: %LOCALAPPDATA%\WSJT-X - SDR\save\

A.2 Clone the starting configuration if required

Close all WSJT-X instances.
Copy %LOCALAPPDATA%\WSJT-X\WSJT-X.ini.
Paste it into %LOCALAPPDATA%\WSJT-X - SDR\.
Rename the copy to %LOCALAPPDATA%\WSJT-X - SDR\WSJT-X - SDR.ini, replacing the newly initialized instance file if intended.

A.3 Separate every data path

A cloned configuration can still point both instances at the same audio input or storage path. That can duplicate decoding of the same audio stream or create file conflicts. In the second instance, verify:

Open File > Settings > Audio.
Under Soundcard, set Input to the intended independent receiver or audio device. The WSJT-X guide specifies 48,000 Hz, 16-bit audio-device configuration.
Set Save Directory to an instance-specific path, normally %LOCALAPPDATA%\WSJT-X - SDR\save\.
Set AzEl Directory to an instance-specific path, for example %LOCALAPPDATA%\WSJT-X - SDR\.
Open File > Settings > General and set the exact Reference callsign and locator used for reporting.
Return to the main WSPR screen, confirm the intended band and audio level, enable spot uploading when required, and verify that uploaded rows use the Reference identity.
Confirm clock synchronization for both instances.

Separate directories do not prove RF-path independence. Confirm empirically that both streams use the intended hardware.

A.4 Configure distinguishable simultaneous TX

For simultaneous TX Hardware A/B, isolation of settings is only the software foundation. Before radiating, verify the complete two-transmitter arrangement into suitable loads or through a safely engineered low-power test path:

Assign the exact Target callsign and QTH to one instance and the different exact Reference callsign to the other. Configure the Reference instance to report from the same test QTH; WSPRadar displays disabled Target and Reference Locator fields derived from the first four Target-QTH characters and matches both uploaded identities within that shared grid-4.
Route each instance to its intended radio, control interface and audio output. A copied configuration must not key or feed the wrong transmitter.
Use the normal WSPR dial frequency on both radios if appropriate, but assign separated audio TX offsets such as 1450 Hz and 1550 Hz. Inspect the waterfall and choose clear, non-overlapping positions rather than assuming those illustrative values are free.
Configure deliberate same-cycle starts. Independent randomized Tx Pct settings do not define a synchronized comparison schedule.
Verify frequency, actual RF power, spectral cleanliness, clock alignment and uploaded callsign/QTH/power for both paths before collecting evidence.
Confirm adequate isolation between active transmitters and antennas. Coupled power can desensitize or damage equipment and can create intermodulation or misleading spots; use appropriate filtering, spacing, power levels and RF engineering for the station.

For a small observed difference, repeat with exchanged audio-frequency assignments and perform a hardware crossover where practical. Preserve both runs separately; do not pool them until the role, correction and analysis scope are aligned.

Appendix B: Sequential TX A/B Scheduling and Switching

This appendix collects the practical schedule and switching guidance behind the TX Hardware A/B playbook. Exact UI controls are in Section 4.3, and exact scheduled-pair construction is in Sections 7.1 and 7.7.

B.1 Requirements for a valid scheduled experiment

For sequential TX A/B antenna tests, one transmitter feeding two RF paths through a controlled switch is normally preferable to two independent transmitters. Transmitter, frequency reference, WSPR chain, callsign, power setting and timing remain common.

Use one normal valid callsign for both paths and identify the paths through different deterministic UTC phases. Enter the transmissions that actually occur on each RF path:

Repeat Interval is each path's actual recurrence, not necessarily a transmitter's displayed Frame value.
Target Start and Reference Start are different even UTC phases below that interval.
Use the shortest practical separation compatible with reliable operation and an acceptable duty cycle.
Report actual power; do not encode path identity through false dBm values.
Verify clock synchronization and the physical schedule-to-path mapping before transmitting.

A deterministic scheduler or controller is required. Standard randomized WSJT-X transmit-percentage operation does not create a fixed A/B sequence.

B.2 WSPRadar Timed A/B Relay Switch

WSPRadar includes:

tools/Timed-AB-Relay-Switch

Currently published version-0.1 release package:

Download the Timed A/B Relay Switch release package

The repository helper uses the same schedule vocabulary and constraints as WSPRadar:

Repeat Interval is shared by Target and Reference and accepts 4, 6, 10, 12, 20, 30 or 60 min.
Target Start and Reference Start are different even UTC phases below that interval.
The default is Repeat Interval = 10, Target Start = 00, Reference Start = 02.

The relay selects each path before its configured start and holds the most recently selected path through unscheduled gaps. It does not switch at unused two-minute WSPR boundaries. Configure the helper and WSPRadar identically from the transmissions that actually occur on each RF path. If physical polarity is reversed, change whether relay ON means Target or swap the two Start assignments.

An optional lead time lets the RF path settle before every scheduled start. Manual physical relay ON/OFF control remains available independently of automatic scheduling. Existing version-0.1 modulo-4 configurations retain their old behavior as 4 / 00 / 02 or 4 / 02 / 00 when loaded. The helper targets common ATtiny45/V-USB HID relay boards with USB VID/PID 16c0:05df and uses the Python HID stack on Windows, Linux and macOS. Consult its README for current installation, permissions and options.

The linked version-0.1 package still contains the former fixed modulo-4 scheduler. Until a newer package is published, use the repository version for the configurable schedule described here.

Install from the tool directory:

py -3 -m pip install -r requirements-relay.txt

or on Linux/macOS:

python3 -m pip install -r requirements-relay.txt

Windows setup and dry run:

Start-Timed-AB-Relay-Switch.cmd --setup
Start-Timed-AB-Relay-Switch.cmd --dry-run

Linux/macOS setup and dry run:

chmod +x ./Start-Timed-AB-Relay-Switch.sh
./Start-Timed-AB-Relay-Switch.sh --setup
./Start-Timed-AB-Relay-Switch.sh --dry-run

A small USB relay should not normally switch RF directly. It should control a properly rated RF switch or relay system. Verify voltage, current, polarity, fail-safe state, RF power, isolation and interlocks.

B.3 Ultimate3S schedule example

The QRP Labs Ultimate3S can run a sequence of WSPR entries and apply a per-entry Aux output to external path-switching hardware. When a two-entry sequence begins at 00, a global 10-minute frame can use Target at 00, Reference at 02, then pause until the next sequence at 10; in WSPRadar this is Repeat Interval = 10, Target Start = 00, Reference Start = 02. The same arrangement with a 20-minute global frame gives each path a 20-minute recurrence while retaining two-minute A/B separation.

The Ultimate3S manual documents Start = 00 specially as "not used", so verify the displayed and observed UTC sequence and enter its actual phases rather than assuming a literal setting-to-time mapping. The Aux lines share display signals; use the documented filtered driver or relay interface and switch only in the RF-off interval [Ref-12].

B.4 QMX schedule examples

One QMX with Frame = 10, Start = 0 transmits at 00, 10, 20, 30, 40, 50. If an external switch alternates those transmissions between paths, Target is 00, 20, 40 and Reference is 10, 30, 50; each path repeats every 20 minutes. Enter Repeat Interval = 20, Target Start = 00, Reference Start = 10; do not enter 10 / 00 / 02.

A single QMX cannot produce an adjacent 00/02 pair followed by an eight-minute pause with that beacon scheduler. It can alternate adjacent paths only by transmitting every two minutes, which the QMX manual discourages as antisocial network use. Two independently scheduled QMX units with Frame = 10, Starts 00 and 02, do implement WSPRadar's 10 / 00 / 02 schedule, but their transmitter chains and actual powers must be controlled as separate hardware [Ref-12].

B.5 Verify mapping and preserve the experiment

Before transmitting:

test without RF power;
verify Target and Reference path polarity;
verify no transition occurs during a WSPR transmission;
use a dummy load or low-power continuity/SWR test;
document relay channel, polarity, lead time, actual on-air schedule, schedule assignment and path mapping.

Switch loss, isolation, connectors, feedline differences and antenna surroundings remain part of the result. Swapping antennas between switch paths can help separate antenna effects from path effects. Repeating the experiment with reversed schedule assignments can help expose timing or role-dependent effects.

Appendix C: Reference SNR Calibration

This procedure estimates a stable additive offset between receive chains or Reference-side paths.

Common input: feed both receive chains from one stable antenna through a suitable splitter and controlled cables.
Characterize the splitter: account for output imbalance and cable differences; swap outputs in a control run when practical.
Collect paired evidence: operate simultaneously across the intended signal levels without changing gain or decoder settings.
Derive the offset: use paired Delta SNR evidence and state whether the value was calculated from station-balanced summaries or raw pairs.
Check consistency: inspect by station, time and SNR. One constant is not defensible if offset changes with level, frequency, AGC or time.
Apply the sign: enter the observed target - reference offset with the same sign.
Validate: repeat or swap paths and confirm corrected common-input Delta is plausibly near zero.

Consistency across station, time and SNR views supports using one additive offset within the tested setup; it does not establish traceable laboratory accuracy. Splitter loss, mismatch, coupling and source instability can remain.

License

WSPRadar is licensed under the GNU Affero General Public License version 3 (AGPLv3). The repository LICENSE file is controlling.

Name		Name	Last commit message	Last commit date
.devcontainer		.devcontainer
.github		.github
.streamlit		.streamlit
.vscode		.vscode
config		config
core		core
docs		docs
img		img
scripts		scripts
tests		tests
tools		tools
ui		ui
.gitattributes		.gitattributes
.gitignore		.gitignore
AGENTS.md		AGENTS.md
AGENT_README.md		AGENT_README.md
CHANGELOG_DAILY.md		CHANGELOG_DAILY.md
CNAME		CNAME
LICENSE		LICENSE
README.md		README.md
app.py		app.py
i18n.py		i18n.py
index.html		index.html
packages.txt		packages.txt
pytest.ini		pytest.ini
requirements-dev.txt		requirements-dev.txt
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation