Present recurrence findings in ATMAI design language v1.2

[abrahamabel]

The retained recurrence report was readable but did not use ATMAI's current visual language or offer individual-seed exploration. This PR adds a derived ATMAI v1.2 presentation of the same frozen findings, with paper reading surfaces, nested ink evidence panels, Archivo headings, technical readouts and one violet observed-result region.

The interactive HTML includes all 20 seeds and four arms, selectable representation/topographic/component measurements, and an accessible table of the plotted values. A Markdown findings page and SVG overview render directly on GitHub. The root README links to both formats.

The prior-A return advantage, higher topographic error, preceding-B adaptation cost, explicitly post hoc 5.32% combined-loss disadvantage, recovery censoring, and observed Linux exact-replay failure retain visible treatment. CI status in the report is a dated observation of research revision 9a24419, not live presentation-branch status.

This is stacked on #9 and contains presentation changes only. Merge #8 and #9 before retargeting this PR to main. All completed experiments, canonical source and reproducibility.json remain unchanged.

Provenance and regeneration

The report is original repository-owned implementation of the user-authorized ATMAI v1.2 specification. No unlicensed source implementation, logo artwork, vendor effects or font binaries were copied. The HTML uses the specification's Google Fonts setup with system fallbacks. PROVENANCE.md records source revisions/hashes and presentation boundaries.

python3 scripts/build_research_report.py verifies frozen files before deriving index.html and overview.svg. Its --check mode is part of both Linux and reference-runtime CI jobs and rejects stale generated presentation files.

Validation

• Repository validator and all 57 unit tests passed locally.
• JavaScript syntax check, deterministic presentation check and whitespace check passed.
• Independent review compared all 20 seeds × four arms × 32 windows and aggregate means with the retained JSON; no blocking evidence finding.
• Browser checked at desktop and narrow mobile layouts with no document overflow. Seed and metric selection updated the chart, caption and all 32 plotted-value rows; the disclosure opened correctly. No browser console errors or warnings were observed.
• Frozen experiment, engine and registry bytes remain unchanged.
• Final branch CI at 928467f: passed. Linux and macOS validation, derived-presentation checks and all 57 tests passed; all three full experiments reproduced exactly on the reference runtime.

Preview: docs/reports/recurrence-v1.2/index.html. GitHub-native findings: docs/reports/recurrence-v1.2/README.md.

---

Note

Low Risk
Documentation and derived HTML/SVG from frozen evidence; no algorithm, experiment, or registry changes. CI only adds a deterministic artifact check.

Overview
Adds a derived ATMAI Design System v1.2 presentation of the frozen capacity-controlled recurrence study (exp-20260907-8313df3e), without changing experiments, the canonical engine, or retained JSON.

docs/reports/recurrence-v1.2/ includes a GitHub-friendly README.md and overview.svg, an interactive index.html (seed/metric timeline chart, tables, reproduction notes), and PROVENANCE.md documenting design sources, evidence hashes, and rebuild commands. The root README now links to this bundle instead of only the lab’s original report.html.

scripts/build_research_report.py verifies frozen registry/files, reads the retained results JSON, and deterministically writes index.html and overview.svg; --check fails CI if committed outputs are stale. Repository integrity workflows on Linux and macOS run that check alongside existing validation and tests.

Unfavorable findings (topology cost, post hoc combined loss, Linux exact-replay failure) remain explicit in the copy; this is presentation and integrity tooling only.

^{Reviewed by Cursor Bugbot for commit 928467f. Bugbot is set up for automated code reviews on this repo. Configure here.}

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.