Compare verified U.S. daily reports

quant-research-review describes changes between two original public daily-report bundles. It reuses quant-market-brief validation for each report and compares facts, report statements and quality gaps without generating new research claims.

quant-research-review \
  --baseline /path/to/baseline/daily_report.json \
  --baseline-manifest /path/to/baseline/publication.json --baseline-date 2026-10-02 \
  --current /path/to/current/daily_report.json \
  --current-manifest /path/to/current/publication.json --current-date 2026-10-05

For news revisions provide the immutable parent through --baseline-parent or --current-parent. Dates and cutoffs must progress forward. A later version of the same report date must have a strictly later generation time. An identical snapshot is explicitly identified. Historical backfills may be generated later than their market dates; report ordering does not establish point-in-time data availability. No implicit archive lookup, source fetching or file output occurs.

The output schema is quant.research_review.v1:

Field	Meaning
report_kind, comparison_kind	U.S. daily report; different dates, same-day revision, or identical snapshot
baseline, current	Report date, cutoff, generation time, owner run ID and hashes
fact_changes	Stable fact IDs, before/after values, absolute deltas and comparability status
claim_changes	Statements grouped by cited evidence IDs, marked added/removed/revised/unchanged
data_quality	Both quality summaries/source statuses, new gaps and resolved gaps
evidence	Public fact/event records indexed separately for each snapshot, including available provenance
validation, limitations	Descriptive interpretation boundaries; no watchpoint or predictive-performance results

Fact deltas require unchanged metric, instrument, unit, source and source URL and finite numeric values. A source/basis change prevents numeric comparison. For percent-valued metrics, a move from 1% to 3% is 2 percentage points, not a 2% cumulative return. Other deltas use the original unit; no conversion is inferred. Missing observations remain missing. Equal numbers may still have different observation dates or quality; inspect the evidence and quality records.

Claim grouping describes report content, not semantic truth or hypothesis success. Matching evidence IDs do not prove the same underlying event. Added statements cite supplied evidence; removed statements are not refutations. Ordering-only changes to source/reference lists are ignored. Evidence-ID collisions between facts and events are rejected. Hash validation confirms consistency with the supplied public manifests, not publisher identity or market truth. Asia insight → watchpoint → outcome and backtest contracts remain separate.

Exit 0 returns JSON. Validation/read errors return 2 with diagnostic stderr and no JSON on stdout. Install the distributable quant-research-review Skill in your Agent environment and use its installed CLI; do not import adjacent repository source. Retained original bundles support longer comparisons without expanding the five-date public Pages window.