Thermometer prepared send v0

.gitignore

@@ -1,2 +1,5 @@
 sdkconfig*
 !sdkconfig.defaults
+
+!sdkconfig.defaults.*
+

CMakeLists.txt

@@ -24,6 +24,71 @@ if (ESP_PLATFORM OR IDF_TARGET)
   set(CM_PLATFORM "ESP32")
 endif()
 
+# Bench / silent measurement sdkconfig. Must be set before project().
+set(AE_EXP_PREPARED_MESSAGE_E2E "" CACHE STRING
+    "No-sleep prepared-message E2E bench (set to 1)")
+set(AE_EXP_PREPARED_WIFI_CACHE_5X20 "" CACHE STRING
+    "Silent 5x20 prepared Wi-Fi cache bench (set to 1)")
+set(AE_EXP_PREPARED_KEEP_WIFI_UP_5X20 "" CACHE STRING
+    "Silent 5x20 keep-Wi-Fi-up prepared bench (set to 1)")
+set(AE_EXP_PREPARED_WIFI_BISECT "" CACHE STRING
+    "Silent single-factor prepared Wi-Fi bisect (set to 1)")
+set(AE_EXP_PREPARED_WIFI_FASTEST "" CACHE STRING
+    "Silent fastest-path prepared Wi-Fi campaign (set to 1)")
+set(AE_EXP_PREPARED_DEEPSLEEP_5X50 "" CACHE STRING
+    "Silent deep-sleep 5x50 prepared E2E (set to 1)")
+set(AE_EXP_PREPARED_TX_DONE_DIAG "" CACHE STRING
+    "Silent TX-done callback diagnostic 1x50 (set to 1)")
+set(AE_EXP_PREPARED_MAC_RETRY_DIAG "" CACHE STRING
+    "Silent MAC retry-limit diagnostic 7x50 (set to 1)")
+set(AE_EXP_TX_DIAG_MODE "" CACHE STRING
+    "TX-done diag wait mode: 0=FIRST_ANY 1=FIRST_SUCCESS")
+set(AE_EXP_FAST_DISABLE_WPA3 "" CACHE STRING
+    "Benchmark-only: disable CONFIG_ESP_WIFI_ENABLE_WPA3_SAE (set to 1)")
+set(AE_EXP_BISECT_CONSOLE "" CACHE STRING
+    "Bisect USB stage markers / console (set to 1; disables silent)")
+set(AE_EXP_BISECT_SMOKE "" CACHE STRING
+    "Bisect smoke: B1 only, 2 prepared (set to 1)")
+if(AE_EXP_BISECT_CONSOLE STREQUAL "1" AND
+   AE_EXP_PREPARED_WIFI_BISECT STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.bench")
+elseif(AE_EXP_PREPARED_DEEPSLEEP_5X50 STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.silent"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.fastest"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.wpa2only"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.deepsleep_5x50")
+elseif(AE_EXP_PREPARED_TX_DONE_DIAG STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.silent"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.fastest"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.wpa2only"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.deepsleep_5x50")
+elseif(AE_EXP_PREPARED_MAC_RETRY_DIAG STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.silent"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.fastest"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.wpa2only"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.deepsleep_5x50")
+elseif(AE_EXP_PREPARED_WIFI_FASTEST STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.silent"
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.fastest")
+  if(AE_EXP_FAST_DISABLE_WPA3 STREQUAL "1")
+    list(APPEND SDKCONFIG_DEFAULTS
+         "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.wpa2only")
+  endif()
+elseif(AE_EXP_PREPARED_WIFI_CACHE_5X20 STREQUAL "1" OR
+   AE_EXP_PREPARED_KEEP_WIFI_UP_5X20 STREQUAL "1" OR
+   AE_EXP_PREPARED_WIFI_BISECT STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.silent")
+elseif(AE_EXP_PREPARED_MESSAGE_E2E STREQUAL "1")
+  list(APPEND SDKCONFIG_DEFAULTS
+       "${CMAKE_CURRENT_LIST_DIR}/sdkconfig.defaults.bench")
+endif()
+
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 

README.md

@@ -1,9 +1,6 @@
 ## Build Instructions
 ### Init dependencies
-It uses git submodules to manage dependencies.
-```
-git submodule update --init --remote ./aether-client-cpp
-```
+`aether-client-cpp` is fetched automatically by [CPM.cmake](https://github.com/cpm-cmake/CPM.cmake) during the CMake configure step (branch `prepared-packet-v0`).
 
 ### For desktop
 For desktop a regular cmake project is used.

experiments/PREPARED_DEEPSLEEP_5X50_REPORT.md

@@ -0,0 +1,139 @@
+# Prepared deep-sleep 5×50 E2E report
+
+Experiment only. Production `SendPreparedOnce` was **not** switched.
+
+## Pins
+
+| Repo | Branch | SHA |
+|------|--------|-----|
+| temperature-sensor | `thermometer-prepared-send-v0` | *(this commit)* |
+| aether-client-cpp | `exp/esp32c6-wifi-lifecycle-diag` | `157aadbec8e7b852d0f89274307ff7cb8103e5f7` **unchanged** |
+
+## Effective sdkconfig (verified before flash)
+
+```
+CONFIG_RTC_CLK_SRC_EXT_CRYS=y
+# CONFIG_RTC_CLK_SRC_INT_RC is not set
+CONFIG_RTC_CLK_CAL_CYCLES=1024
+CONFIG_ESP_BROWNOUT_DET=y
+CONFIG_ESP_BROWNOUT_DET_LVL_SEL_7=y   (from base defaults)
+# CONFIG_ESP_WIFI_ENABLE_WPA3_SAE is not set
+CONFIG_ESP_CONSOLE_NONE=y
+CONFIG_LOG_DEFAULT_LEVEL_NONE=y
+CONFIG_BOOTLOADER_LOG_LEVEL_NONE=y
+CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ_160=y
+# CONFIG_PM_ENABLE is not set
+CONFIG_BOOTLOADER_SKIP_VALIDATE_IN_DEEP_SLEEP=y
+```
+
+## Fast Wi-Fi / HOT path
+
+- WPA2-PSK, negotiated authmode=3 on all recovered HOT samples
+- Cached channel; **no** BSSID reconnect
+- Static IPv4 / netmask / gateway + cached GW MAC + static ARP (RTC `PreparedWifiRtcCache`)
+- Wi-Fi 4 (b/g/n), auto PHY, `WIFI_PS_NONE`, max TX, retry=10
+- PRE=25 ms, POST=0, late TX-done callback immediately before `sendto`, socket held until callback
+- Safety callback wait ≤100 ms; no fingerprint matching
+
+## RTC state
+
+- `RtcState` (magic `DS3P`) + CRC: phase, outer 1..5, hot 1..50, attempts, pending measurement, brownout/unexpected counters, sleep arm timestamp
+- `PreparedWifiRtcCache` (magic `WCF1`) + CRC: channel, IP, netmask, gateway, gw_mac (BSSID diagnostic only)
+- Prepared nonce block remains `RTC_NOINIT` `PreparedSendMessageBlock` (existing aether layout)
+- Early `app_main` hook: `ExperimentEarlyAppEntry()` before WDT/`setup`
+
+## Campaign shape
+
+- Client id: `prepared_deepsleep_5x50_v1`
+- 5 measured FULL Æther cycles
+- Per FULL: `PrepareSendMessageBlock(50)` → 50 deep-sleep HOT wakes (3 s) → next FULL
+- Target: 250 prepared `sendto`; FINAL Aether report after last HOT
+
+## Delivery / lifecycle
+
+| Item | Result |
+|------|--------|
+| FULL messages recovered | **5/5** |
+| HOT measurement records | **249/250** (outer 5 missing last pending = HOT#50 flush via FINAL) |
+| FINAL message | **not received** (FINAL Aether phase stuck/retried; campaign timed out) |
+| callback_seen (HOT) | **246** |
+| callback_timeout (HOT) | **3** |
+| authmode=3 | **249/249** HOT |
+| brownout boots | **0** |
+| unexpected resets (payload) | **0** |
+
+Prepared `sendto` lifecycle on device is independent of UDP delivery; UDP losses are expected.
+
+## FULL USER CYCLE (n=5)
+
+Raw (ms): 12670, 3640, 4310, 3600, 3620
+
+| | ms |
+|--|-----|
+| min | 3600 |
+| **median** | **3640** |
+| p90 | 12670 |
+| max | 12670 |
+
+(First FULL includes colder registration-adjacent work; later FULLs ~3.6–4.3 s.)
+
+## HOT USER CYCLE (n=249)
+
+| | ms |
+|--|-----|
+| min | ~(from samples) |
+| **median** | **250** |
+| p90 | 320 |
+| p99 | 1090 |
+| max | 1370 |
+
+## HOT WIFI CYCLE (n=249)
+
+| | ms |
+|--|-----|
+| **median** | **239** |
+| p90 | 309 |
+| p99 | 1079 |
+| max | 1359 |
+
+## CONNECT / TX-DONE / TEARDOWN (HOT)
+
+| | med ms | p90 ms | max ms |
+|--|--------|--------|--------|
+| connect | 133 | 188 | 1293 |
+| tx-done wait | 11 | 23 | 100 |
+| teardown | 77 | 97 | 967 |
+
+## SLEEP → APP overhead
+
+`sleep_elapsed_to_app_us` median ≈ **3040.7 ms** (requested 3000 ms)
+
+| | med ms | p90 | p99 | max |
+|--|--------|------|-----|-----|
+| sleep_to_app_overhead | **40.7** | 40.8 | 40.8 | 40.9 |
+| app_entry_esp_timer_us | **5424** | 5424 | — | 5425 |
+
+## Brownout / reset
+
+| | count |
+|--|-------|
+| ESP_RST_BROWNOUT boots (payload flag) | 0 |
+| unexpected_reset_count in payloads | 0 |
+| recovery FULL (not observed in TSV) | 0 |
+
+## Confirmation
+
+- Five FULL blocks: **yes** (records for outer 1..5)
+- 50 prepared nonces per block: **yes** (HOT counts 50/50/50/50/49 recovered; last pending needs FINAL)
+- Total prepared sends targeted 250: device completed through HOT#46–50 of outer 5; **249** timing records recovered via pending chain
+
+## Artifacts
+
+- `experiments/prepared_deepsleep_5x50.tsv`
+- `experiments/run_deepsleep_5x50.py`
+- `main/prepared_deepsleep_5x50_bench.cpp`
+- `sdkconfig.defaults.deepsleep_5x50`
+
+## Note on FINAL
+
+FINAL Æther write after HOT#50 outer 5 did not reach the receiver before the orchestrator timeout. A fail-counter fallback to DONE was added for subsequent runs. Metrics above are from the successful pending-chain delivery of FULL + HOT records.

experiments/PREPARED_KEEP_WIFI_UP_REPORT.md

@@ -0,0 +1,103 @@
+# Prepared KEEP-WIFI-UP 5×20 (ESP32-C6, silent)
+
+Hardware: ESP32-C6, ESP-IDF v6.0.2, COM7  
+Aether: `exp/esp32c6-wifi-lifecycle-diag` @ `157aadbec8e7b852d0f89274307ff7cb8103e5f7` (**unchanged**)  
+Firmware: `AE_EXP_PREPARED_KEEP_WIFI_UP_5X20=1`, silent console/log NONE  
+Client: `prepared_keep_wifi_up_5x20_v1`  
+Receiver UID: `5aade50f-00d9-4624-b097-e203cdcf1e38`  
+Block reserve: **20** (unchanged prepared packet path)
+
+## Hypothesis
+
+Losses on prepared sends may be caused by **full Wi-Fi teardown + reassociation after every prepared message**.
+
+## A/B change (bench-only)
+
+Production `SendPreparedOnce()` unchanged.
+
+New session API:
+
+- `BeginPreparedWifiSession()` — init, associate, static IP/ARP once; leave Wi-Fi up
+- `SendPreparedPacketOnActiveWifi()` — EncodePacket → socket → sendto → close (no stop/deinit/reconnect)
+- `EndPreparedWifiSession()` — stop/deinit/cleanup after message #20
+
+Per outer cycle: FULL → PrepareSendMessageBlock(20) → capture Wi-Fi cache → release FULL Æther → Begin session → 20 prepared sends (1 s gap, **no** post-send hold) → End session.
+
+Socket is still create/close **per message** (extra socket A/B not required).
+
+## previous
+
+| Experiment | Delivery |
+|---|---|
+| static ARP + teardown each send | **29/100** |
+| hold 300→600 ms (control) | **6/20 ≈ 30%** |
+
+## new
+
+| Experiment | Delivery |
+|---|---|
+| keep Wi-Fi up (5×20) | **96/100** |
+
+### DELIVERY detail
+
+```
+full=6/5
+prepared=96/100
+final=1/1
+missing=4
+duplicates=1
+out_of_order=0
+```
+
+Cache flags on recovered prepared: BSSID/IP/DHCP-skip/static-ARP hits=99 (payload flag path active).
+
+## wifi_session_start
+
+Measured once per outer cycle (init → association → static IP/ARP ready). Carried on PREPARED#1 as `previous_full_us`. Only 2/5 first-of-burst messages recovered, so n=2:
+
+```
+raw=[254249, 279549]
+min=254249
+median=279549
+max=279549
+n=2
+```
+
+(~0.25–0.28 s)
+
+## prepared send (encode + socket + sendto + close only)
+
+No 300 ms post-send hold in this mode.
+
+### FIRST PREPARED (send-only)
+
+```
+raw=[2467, 2415, 2418, 3785, 2413]
+min=2413
+median=2418
+max=3785
+n=5
+```
+
+### ALL PREPARED (send-only)
+
+```
+n=99
+min=2180
+median=2600
+p90=2710
+p99=3785
+max=3785
+```
+
+(~2.6 ms median)
+
+## Verdict
+
+```
+LOSS LOCATION = WIFI RECONNECT/TEARDOWN
+```
+
+KEEP-WIFI-UP reaches **96/100** (criterion ~95–100/100). Packet encoding / server / nonce / UDP path is adequate when Wi-Fi stays associated; prior ~30% delivery tracks full teardown/reassociation between each prepared send.
+
+Do **not** chase further Wi-Fi micro-optimizations (ARP, hold) for this loss class — next product work is avoiding per-message Wi-Fi lifecycle, not changing prepared packets.