[chin0312]

Summary

PR #23 has been merged. Production assets are NVDAx, SPYx, and AAPLx; QQQx and TSLAx remain research-only.

This PR keeps the approved Evidence State v2 backend contract and completes its human-facing Console integration:

• The public Console names the model "Valtide Model" or "Model"; API model IDs remain unchanged.
• Current findings and valuation-range copy use concise risk language without repeated model-dependence explanations. Number Line and timeline chart now share Observed xStock, Valtide Fair Value, X-Perp, and coverage-derived Valuation Range; Underlying Anchor remains context, not a plotted market marker.
• Policy rows are directly visible; assets with no optional onchain binding show "No Policy Connected" without X Layer failure chrome.
• Observation audit timestamps and provenance labels are human-readable; v2 hides implementation-only calibration/rule fields while old records remain marked as an earlier rule.
• The existing Demo, asset isolation, detector-score Model Distance, chart bindings, three-asset picker, and v1 audit preservation remain intact.

Evidence State v2 mapping

James's frozen SUPPORT/WATCH/REVIEW bands, model fits, calibration, q80/q95 thresholds, and research outputs are unchanged.

• SUPPORT + healthy exact-time X-Perp + quality gates -> SUPPORTED. X-Perp direction inside SUPPORT is descriptive only.
• WATCH -> INCONCLUSIVE.
• REVIEW + X-Perp closer to xStock -> INCONCLUSIVE.
• REVIEW + X-Perp closer to P1a-C -> CHALLENGED.
• Missing/stale/invalid sources or quality failures -> INCONCLUSIVE.

Research evidence

The following are James's original frozen exposed-test results, not fresh-panel metrics:

• NVDAx SUPPORT: n=10,369; raw xStock MAE about 4.28 bps; p95 absolute error about 13.64 bps; relative-q95 tail rate about 0.395%. REVIEW raw MAE about 18.91 bps (about 4.4x SUPPORT).
• SPYx SUPPORT: n=11,551; raw xStock MAE about 3.38 bps; p95 absolute error about 12.69 bps; relative-q95 tail rate 0%. REVIEW raw MAE about 125.74 bps (about 37.3x SUPPORT).
• AAPLx SUPPORT: n=11,195; raw xStock MAE about 5.41 bps; p95 absolute error about 27.04 bps; relative-q95 tail rate about 0.027%. REVIEW raw MAE about 61.82 bps (about 11.4x SUPPORT).

The exact fresh canonical CSV bytes were unavailable locally, so the full fresh SUPPORT-band truth metrics were not independently recomputed. This is a follow-up validation limitation, not a merge blocker: the frozen research bands and thresholds are unchanged, and the removed X-Perp directional preference within SUPPORT was not part of James's validated detector rule.

The expected v2 panel totals below are arithmetic derivations from committed frozen aggregates, not a new row-level replay:

• NVDAx: 3,731 SUPPORTED / 319 INCONCLUSIVE / 60 CHALLENGED.
• SPYx: 4,009 SUPPORTED / 30 INCONCLUSIVE / 71 CHALLENGED.
• AAPLx: 4,001 SUPPORTED / 102 INCONCLUSIVE / 7 CHALLENGED.

Publisher safety

Publisher compatibility expects v2 and checks the result's evidence_semantics explicitly. Existing deployment bindings do not declare v2 publication compatibility, so writes remain fail-closed. This PR does not change deployment manifests, contracts, or publication configuration; no transaction was sent.

Validation

• npm test: 53 passed.
• npm run build: TypeScript and Vite production build passed; existing large-chunk advisory remains.
• Exact-head GitHub Actions run 37514669807 is green; Vercel preview completed successfully.

Scope

Only apps/web/** changed. CSS, frontend layout system, backend, quant artifacts, docs, contracts, deployments, Landing, Docs, and Methodology were not changed. No production deployment or mutation was performed.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
valtide	Ready	Preview	Oct 6, 2026 8:13pm UTC