Align artifact with the SSRN paper; make refutation conditions executable

[AMBRA7592]

Summary

This PR closes every gap found in a full audit of the artifact against the companion paper (Brandes 2026, Rendering as Audit, SSRN 6612663), and elevates the page from describing its verification claims to executing them. All scored values, confidence levels, flag placements, and the E-1 prediction remain identical to Figure 1/§4 of the paper — nothing here extends or preempts the paper.

Alignment fixes (repo ↔ paper inconsistencies)

1. The † flag is now genuinely protocol-derived. The page and paper (§3.3, §4) state the merged-unresolved flag is "not manually applied" but "a protocol-generated rendering consequence" — yet the code hard-coded it as flag: "MU" in the data, and the footer invites View-Source auditing "per Principle 3". A code auditor would have concluded the artifact fails its own P3 test. The rule is now an executable condition (PROTOCOL.MU.applies) applied to recorded findings; no flag exists anywhere in the data.
2. κ is computed, not hard-coded. Composite values were HTML literals; they're now summed from the nine rendered components at render time, with the paper's published 0.58/0.61 kept as fixtures the self-audit checks against (Table 1 row 1: "aggregation opacity").
3. Refutation scope corrected. README/page invited evaluation against "the affordances claimed in §6" wholesale — but §6.3 of the paper explicitly scopes A6/A8 as beyond the current principle set, so that wording set the artifact up to be "refuted" on tasks the paper never claimed. Coverage is now stated precisely: A1–A5 + A7 at two-case scale; A6/A8 out of scope; the §4 full tool (panel, coupling matrix, scalar diagnostics) explicitly not part of the demo.
4. Bar overflow fixed. Egypt line 9 (0.18 × 900 = 162px) overflowed its 160px container; bars now use a declared shared axis (0–0.20) with an overflow guard, documented in the legend.
5. Prediction registry honesty. "As of rendering date" (meaningless on a static page) → an explicit as-of date tied to the registry's last substantive update (2026-04-20), plus a review log that distinguishes pre-publication history (2024 seal, 2025 review point) from what is publicly verifiable via commit history (§5's standard applied to the artifact itself).

Additions

• Self-audit panel — six refutation conditions executed in-browser on every load (decomposition rendered; κ = component sum = paper values; provenance present; flags rule-derived and matching Figure 1 exactly; columns aligned; registry complete/dated), framed per §6.3 as affordance tests, not rigor tests.
• Bidirectional linking: SSRN links in header/intro/footer; CITATION.cff with the paper as preferred citation (enables GitHub's "Cite this repository").
• Machine-readable JSON view serialized from the live objects (what you export is what was rendered), incl. the serialized rule predicate.
• Accessibility: full keyboard operation (Tab/Enter/Space/Esc), ARIA table semantics, live-region inspector, mobile column labels, noscript notice.
• CSP blocking all network requests — the "no tracking, no dependencies" claim is now enforced, not asserted.
• CHANGELOG.md (every change annotated with the paper section motivating it — the anti-drift mechanism) and a refutation-report issue template operationalizing §9's invitation.

Verification

Exercised end-to-end in headless Chromium: 17/17 checks pass, zero console errors (CSP clean). Verified: 9 rows render by default; κ computes to 0.58/0.61; exactly two derived flags on line 9; self-audit reports 6/6; hover/click/focus/Enter all drive the inspector; Esc clears; registry renders seal/cadence/log; JSON view parses and matches; max bar width 144px; and a tamper test (mutating a component value in-page) correctly trips the self-audit to 5/6 with the failure banner.

Deliberately unchanged

• All data values (paper fixtures), the visual design language, the single-file no-dependency architecture, the MIT license, and the live URL cited in §9 of the paper.
• No panel view, no additional cases, no coupling matrix — those would preempt the paper, which defers panel-level principles (§6.3).

Suggested follow-ups for the author (not in this PR)

• After merge, tag the release (e.g. v1.1) and consider a Zenodo DOI + repo topics/homepage for citability/discoverability.
• Record the outcome of the 2025-09-12 E-1 review in the registry's review log (only you can attest it).
• When next revising the SSRN paper, consider citing the tagged artifact version (or DOI) alongside the live URL.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SpETgXVZ59E5U8SXqY17ix

---

Generated by Claude Code