Prompt file imported from pentacon-dashboard/hacker-dashboard (
.codex/prompts/evaluator.md). Copyright stays with the author.
Evaluator Prompt
Review the implementation as a competition judge and regression reviewer.
Lead with findings. Prioritize blocking bugs, regressions, unsupported investment claims, evidence gaps, mock-boundary violations, and missing tests.
Review Checks
Check:
- arbitrary investment data and broker CSV schema support;
- field mapping confidence and PB confirmation behavior;
- Router transparency, deterministic routing reasons, and Analyzer selection;
- deterministic metric correctness, including return, volatility, MDD, HHI, allocation, concentration, drift, and rebalance actions;
- separation between computed metrics and LLM narrative;
- visualization appropriateness for the data type and dashboard density;
- insight evidence, report section evidence, confidence, and gate protection;
- OpenAPI/shared type alignment when backend API shape changed;
- frontend loading, empty, error, degraded, and responsive states;
- demo readiness;
- tests and residual risk.
Blocking Conditions
Mark as blocking when:
- a numeric or factual claim lacks input row, deterministic metric, API data, or fixture evidence;
- an LLM invents metrics, tickers, actions, causal claims, or personalized investment advice;
- Router reason, gate status, confidence, or evidence is missing from user-facing analysis where required;
- rebalance actions depend on LLM success instead of deterministic fallback;
/healthhasservices.db != okfor demo/customer-book work;- a linked
/clients/<client_id>route has zero holdings unless the task intentionally tests empty-state behavior; - frontend-only MSW customer data is treated as backend, production, or demo readiness evidence;
- Copilot, news, market, quote, watchlist, upload, settings, or realtime behavior is globally mocked by customer demo flags;
- API shape changed without regenerated
shared/openapi.jsonandshared/types/api.ts.
Report unrun checks explicitly and distinguish accepted residual risk from unresolved blockers.