Add auditable Evidence One-Pager

.github/workflows/commercial-research-beta.yml

@@ -10,7 +10,7 @@ permissions:
 jobs:
   local-engineering-gate:
     runs-on: ubuntu-latest
-    timeout-minutes: 30
+    timeout-minutes: 60
     env:
       PYTHONDONTWRITEBYTECODE: "1"
       PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}

.superpowers/sdd/2026-08-16-evidence-one-pager/design-qa.md

@@ -0,0 +1,124 @@
+# Evidence One-Pager design QA
+
+## Result
+
+`final result: passed`
+
+All saved images were inspected. There are zero open P0, P1, or P2 findings.
+One P3 observation is recorded as intentional polish rather than a blocking
+defect.
+
+## Scope and method
+
+- Implementation HEAD: `129129b493265b227ebcbb6f8670a74931df0ec0`.
+- State inspected: the deterministic standalone complete-state fixture created
+  with the existing `_synthetic_brief("complete")` test seam.
+- Fixture HTML:
+  `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/complete.html`,
+  SHA-256
+  `861a8e7048eb00db3fee216a42b1cd989d7236ea6f3351094524bb1c7d870b9d`.
+- Accepted screenshots were captured with the repository's Playwright/Chrome
+  setup, Google Chrome, browser zoom `100%`, DPR `1`, visual viewport scale
+  `1`, and screenshot `scale="device"`.
+- The desktop implementation and the approved reference were compared at the
+  same native `2310x1504` pixel dimensions. Neither panel was cropped, resized,
+  stretched, or content-aware edited.
+- The current captures were accepted only after their PNG signatures, exact
+  IHDR dimensions, viewport metadata, overflow, and browser-error fields were
+  verified.
+
+## Artifact identities
+
+| Artifact | Exact path | SHA-256 | Verified geometry |
+| --- | --- | --- | --- |
+| Approved reference | `/var/folders/cw/xfqgmp_57rn7nn3fq68z_6280000gn/T/codex-clipboard-80b40520-4c8b-493e-89af-a87e159e329b.png` | `d467ce50f7803b3a269b5cfd748a87c1ce4a269345943ca6993d365056c72d59` | PNG IHDR `2310x1504` |
+| Desktop implementation | `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/implementation-desktop-2310x1504.png` | `6395a7b396e0cf014a61d31c1f1ca87ddad03bdfca1dc66d9dd14c5f10091990` | PNG IHDR and CSS viewport `2310x1504` |
+| Phone implementation | `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/implementation-phone-390x844.png` | `47c9232768b6030f42c6d5dbfe2c1757f2dcd9097efeaee0ccd9fc1cc297aa47` | PNG IHDR and CSS viewport `390x844` |
+| Labeled comparison | `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/reference-vs-implementation-2310x1504.png` | `e2514378e1201fc74b5649d1943f6bb6b343370cc7f974c20527ccbd046b34c1` | PNG IHDR `4700x1624`; equal `2310x1504` panels, `80px` gutter, `120px` label band |
+
+Supporting evidence:
+
+- Capture metadata:
+  `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/capture-metadata.json`,
+  SHA-256
+  `9c177008c08fc4f4280a64d72463c51eea3011a84b7407b17219f4d7e96cf6c8`.
+- Composite metadata:
+  `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/composite-metadata.json`,
+  SHA-256
+  `9f58c1f02499bc588dbf0494719089028c6720f7a6a1a45cdad57c05e961f171`.
+- Pixel verification:
+  `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/pixel-verification.log`,
+  SHA-256
+  `481d3326573303ebd884cb72a4539512b0f8e5c59cbc248daa67530aabf7d1bd`.
+  It records both composite panels as raw-pixel-equal to their source images
+  and confirms that labels sit outside the image content.
+- Hash ledger:
+  `/tmp/stock-evidence-one-pager-design-qa-129129b493265b227ebcbb6f8670a74931df0ec0/visual-hashes.tsv`,
+  SHA-256
+  `72990560d766305dff6e23e83de3b40f5fb10972deb39da1ca341efbeff18f7e`.
+
+## Findings by severity
+
+| Severity | Count | Finding and disposition |
+| --- | ---: | --- |
+| P0 | 0 | No blocker, unusable flow, or evidence-boundary breach observed. |
+| P1 | 0 | No major hierarchy, readability, clipping, overflow, or interaction defect observed. |
+| P2 | 0 | No material responsive, spacing, state, or visual-consistency defect observed. |
+| P3 | 1 | At `2310px` wide, the implementation retains the existing centered, narrower Research Brief shell and therefore uses more surrounding whitespace than the dense reference canvas. This is intentional: the approved design preserves the existing report, avoids a fixed dense dashboard canvas, and permits normal scrolling. No change is required for this slice. |
+
+## Visual assessment
+
+- The summary has a distinct dark surface, amber emphasis, legible hierarchy,
+  explicit state text, semantic cards, and a clear handoff to the complete
+  Research Brief below it.
+- State meaning does not depend on color alone. Visible state labels and card
+  text remain present in the complete-state capture.
+- The complete report remains in the same offline document and is visibly
+  reachable by normal vertical scrolling.
+- The phone capture reflows to one column. Identity copy wraps without
+  collision, and measured document and one-pager horizontal overflow are both
+  `0px`.
+- Desktop and phone captures recorded no console errors and no page errors.
+
+## Intentional differences from the reference
+
+The implementation follows the reference's visual hierarchy without copying
+claims that the frozen evidence does not support:
+
+- no `Certified` badge or certification claim;
+- no target price, upside, current-price, or spot-price framing;
+- no probability, percentile, or confidence-distribution claim;
+- no capital-allocation instruction; and
+- no buy, sell, own, invest, or other action language.
+
+Instead, it shows already-frozen evidence, scenario assumptions, independent
+state disclosures, blockers, process status, and the next research task. The
+scenario naming uses `Bull`, not `Blue Sky`, and the artifact remains English
+only for this bounded slice.
+
+## Phone, zoom, and contrast evidence
+
+- Direct capture evidence: desktop `2310x1504@1` and phone `390x844@1`, both
+  at browser zoom `100%`, visual viewport scale `1`, with zero document or
+  one-pager overflow.
+- The bridged standalone aa6 browser packet, result SHA-256
+  `2eba7b2a354ba0bcc2cdedb5119f07e21879023722472c79fa64f0b45fcd80b2`,
+  covers the complete 24-cell matrix at
+  `100%`, `200%`, and `400%` zoom with zero failed assertions.
+- The fresh Research accessibility packet on implementation HEAD, SHA-256
+  `52c62631274d7f28027f648a98144133cf271b9bf6594cba550d6498bcfd5946`,
+  covers existing desktop and phone
+  routes plus all three Workbench one-pager cells. The Workbench cells recorded
+  minimum text/link contrast `16.644346951453382`, minimum boundary contrast
+  `3.6594143300026367`, a `44px` download target, zero overflow, and no
+  console/page errors.
+- The same browser evidence includes forced-colors and print assertions. These
+  are deterministic automated checks, not a claim of human or screen-reader
+  validation and not a complete WCAG conformance audit.
+
+## Final decision
+
+The inspected desktop and phone implementation is visually coherent,
+responsive, and faithful to the approved evidence boundary. With every P0,
+P1, and P2 closed, the visual QA gate passes. The single P3 note is an
+intentional shell-width difference and remains non-blocking.