Repository navigation
Add live research replay batch to Judge - #285
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 992cda0fbc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| <section className="judge-section judge-published-research" aria-labelledby="published-research-title"> | ||
| <div className="judge-section-heading"><span className="judge-kicker">06 / OWNER-PUBLISHED RESEARCH</span><h2 id="published-research-title">Two real runs, ready for review.</h2><p>These owner-published records preserve the original question, captured sources, instrument context, risk gates, and process trace. They are research snapshots, not audited performance.</p></div> | ||
| <div className="judge-section-heading"><span className="judge-kicker">06 / OWNER-PUBLISHED RESEARCH</span><h2 id="published-research-title">Seven live records. One clear evidence boundary.</h2><p>Five new token reviews join the two earlier public examples. Every card opens a real owner-published snapshot with captured sources, instrument checks, and risk gates. Links expire 30 days after publication and can be revoked.</p></div> | ||
| <p className="judge-lede">Across the five new runs, 40 issuer headlines were collected per company from 17–30 publishers. AI synthesis was unavailable in each, so these are transparent fallback records rather than completed analyst theses. No directional assessment or trade handoff was made; each replay shows quote freshness and unmeasured token-market checks.</p> |
There was a problem hiding this comment.
For all five AI-unavailable fallback runs, this claims that the replay exposes quote freshness, but publication drops tokenizedMarketReview and serializes only the stock-perpetual fields from market.perpSymbol, market.lastPrice, and market.providerTimestamp (app/api/research/replays/route.ts:65). The replay renderer then displays only that price as a generic “Market snapshot,” without its symbol or timestamp (app/replay/[token]/page.tsx:25). Consequently, judges cannot verify the advertised stale/fresh rToken labels and may mistake a perpetual price for the token snapshot; include the tokenized-spot status and timestamp in the public payload/UI or remove this claim.
Useful? React with 👍 / 👎.
| <div className="judge-lane-grid"> | ||
| <Link href="/replay/c2ffbe89-74a5-4547-bf89-5952e37efeaa"><span>RNVDA · TOKENIZED MARKET</span><strong>Is the rNVDA market worth reviewing?</strong><small>Inspect what the run found about NVIDIA coverage, the instrument context, and the evidence limitations.</small><em>Open published replay ↗</em></Link> | ||
| <Link href="/replay/91bb478a-7977-4b34-b32d-b1f558a8af75"><span>NVDA + TSLA · COMPARISON</span><strong>Compare the latest evidence and risks.</strong><small>Review the two issuer assessments, their cited public sources, and separate decision gates.</small><em>Open published replay ↗</em></Link> | ||
| <Link href="/replay/c5f5b22e-c995-4a5e-8193-5737c4d413b8"><span>RTSLA · EV / AUTOMOTIVE</span><strong>Would you review the rTSLA market?</strong><small>40 issuer headlines across 30 publishers. The token quote was stale; depth, spread, and premium/discount were not measured.</small><em>Inspect published replay ↗</em></Link> |
There was a problem hiding this comment.
Thirty days after these records were published—or immediately if the owner revokes them—each hard-coded card will lead to a 404 because the replay route only selects unrevoked, unexpired rows. Republishing does not repair these URLs: POST /api/research/replays creates a new public_id after expiry. The permanent Judge page, README, and evidence manifest will therefore continue advertising dead “live” evidence unless the links are resolved dynamically, replaced with durable curated artifacts, or removed when they expire.
Useful? React with 👍 / 👎.
| <Link className="judge-secondary" href="/demo">Read the product walkthrough <span>→</span></Link> | ||
| <Link className="judge-secondary" href="/evidence">Inspect evidence manifest <span>↗</span></Link> | ||
| <Link className="judge-secondary" href="/replay">Open labeled research replay <span>↗</span></Link> | ||
| <Link className="judge-secondary" href="/replay/c5f5b22e-c995-4a5e-8193-5737c4d413b8">Read a live rTSLA review <span>↗</span></Link> |
There was a problem hiding this comment.
When a judge already has a Tidelight session belonging to anyone other than the publishing owner, this new hero link returns 404 instead of the advertised public replay. Both the page and JSON endpoint query through the session-aware Supabase client, while 20261021123000_public_research_replays.sql grants the active-link RLS policy only to anon; authenticated sessions receive only the owner policy. Extend the active-link policy to authenticated users or deliberately perform this public lookup anonymously so the no-login Judge flow works regardless of existing cookies.
Useful? React with 👍 / 👎.
Summary
Verification