End-to-end site validation and regression repair after Results/Changelog, leverage, and chart work

Objective

Maintenance_execution for qm-yn3d: full end-to-end validation of the static site after the Results/Changelog, strategy-class separation, leverage/rolling-lookback, and experiment-chart changes. Constraints: profile maintenance with modifiers web and public_artifact — do not mutate data/market.sqlite, do not log secrets, do not hand-edit docs/research/catalog-canary/**, docs/strategies/index.html, or docs/.shared-artifact-manifest.json; use only canonical producers.

Approach

• Served docs/ locally via python3 -m http.server --directory docs --bind 127.0.0.1 8788 (8787/8765 were occupied by unrelated bundlers; 8788 is the validated origin). Smoked index.html, results.html, allocatesmartly-strategies.html, other-strategies.html, allocatesmartly-meta.html, other-meta.html, meta-strategies.html, experiments.html, research.html, catalog.html, plus 2–3 detail pages each of results/, strategies/, meta-strategies/, experiments/, research/catalog-canary/ via HTTP 200 checks over 22 URLs.
• Browser-exercised with real Playwright Chromium (headless, --no-sandbox) at desktop 1280x800 and mobile 375x812. For each viewport visited the same 23 pages, waited for networkidle, measured canvas/svg/table presence, captured full-page screenshots to /tmp/qm-yn3d-evidence/{desktop|mobile}-*.png, and recorded console errors and 4xx responses to summary.json. Used the same harness to verify Historical Results/Changelog newest-first ordering against site-data/results.json (517 entries, first qm-b6tc 2026-08-22T15:49:00Z), strategy-class separation against site-data/catalog.json (133 family=as vs 6 family=other), and JS routing via docs/assets/js/site.js.
• Checked Meta pages (3) for charts/canvas (5–6), tables (2–4), allocations, and links; checked experiment pages (3) for signal plots (canvas/svg present, 2 each, no JS errors); verified research/catalog-canary/index.html manifest and payload rendering.
• Ran gates:
• python3 research/scripts/check_static_reports.py docs → local bundle OK: 1415 files; 183 canary routes
• python3 -m research.artifacts check --mode full --offline → shared artifacts deterministic and current
• python3 -m pytest -q tests/test_reports_coverage.py tests/test_reports_export.py tests/test_reports_pages.py tests/test_reports_scaffold.py tests/test_site_ia.py tests/test_site_internal_links.py tests/test_as_verify.py → 118 passed

Regression Found and Fixed

Playwright reported 6 failures: every docs/results/*.html detail page loaded results/assets/css/site.css (404) instead of ../assets/css/site.css, with console error "Failed to load resource: the server responded with a status of 404". Root cause was research/reports/results.py::_render_detail using render_site_header(current="results") with empty root_prefix and inline <link rel="stylesheet" href="assets/css/site.css" />. The fix was a producer-only change:

• research/reports/results.py — _render_detail now calls render_site_header(root_prefix="../", current="results") and emits <link rel="stylesheet" href="../assets/css/site.css" />, matching the experiments/detail pattern (root_prefix="../").

No hand-edits to generated docs/ were made; the site was rebuilt via canonical producers:

• python3 -m research.reports.results (equivalently python3 -m research.reports build --render)

Re-ran the 23-page Playwright harness after rebuild: all 46 viewport-page loads returned 200 with 0 failed requests and 0 console errors; strategies/<em>.html retained 6 canvases/2 tables, meta-strategies/meta-</em>-lev-<em>.html 5 canvases/4 tables, experiments/</em>.html 2 canvases each, with correct payload payloads.

Files Changed

• research/reports/results.py — fix Results-detail asset/header prefix (CSS and render_site_header now ../)
• research/results/qm-yn3d.md — this durable task note (new)
• docs/site-data/results.json — regenerated (517 → 518 entries; newest qm-yn3d)
• docs/results.html — regenerated index (518 entries, newest-first)
• docs/results/qm-yn3d.html — new detail page for this validation
• docs/results/*.html (517 existing detail pages) — regenerated with corrected ../assets/css/site.css and ../index.html brand link

No changes to data/market.sqlite, docs/research/catalog-canary/**, docs/strategies/index.html, or docs/.shared-artifact-manifest.json beyond deterministic rebuilds.

Validation

• python3 research/scripts/check_static_reports.py docs — 1415 files OK post-rebuild
• python3 -m research.artifacts check --mode full --offline — deterministic (11 artifacts)
• Focused pytest suite (reports/site/as_verify) — 118 passed
• HTTP smoke — 22 URLs 200 OK, site-data/results.json newest-first, site-data/catalog.json family/tab/source counters verified
• Playwright browser evidence at 1280x800 and 375x812 — 46 screenshots in /tmp/qm-yn3d-evidence/, summary.json with 0 404s post-fix, charts/tables proven via canvas/svg counts:
• strategies/aaa.html, haa-simple.html, sixty-forty.html — 6 canvases each
• meta-strategies/meta-target-risk-6040-lev-2.html, meta-target-cagr-sp500-lev-2.html — 5 canvases, 4 tables
• experiments/qm-kdnj.2.html, qm-wpx5.html, qm-aaaz.html — 2 canvases each (signal plots) with no JS errors
• results/qm-b6tc.html, qm-wpx5.html, qm-g7jk.html — body renders, metadata badges correct, ../assets/css/site.css loads
• research/catalog-canary/index.html — canary table renders

Results

All acceptance criteria were exercised with executable browser proof at both viewports; gates pass without drift; the single Results-detail CSS regression was repaired via the producer and reverified; research/results/qm-yn3d.md and docs/site-data/results.json now reflect the validation and appear newest-first in results.html.

Risks / Limitations

The full python3 -m pytest -q suite is heavy and times out at 300s in this environment; the smallest-relevant focused suites were used per Main's guidance (reports + site + as_verify) and the browser harness covers the remaining web-contract proof. Existing data/market.sqlite (74.9 MiB) was preserved read-only throughout.