Skip to content

[deep-report] Fix open_issues/total_prs glossary-scope violations causing false Daily Regulatory Report alarms #63531

Description

@github-actions

Description

The Daily Regulatory Report (#63513, 2026-09-25) flagged two "critical" cross-report discrepancies — open_issues varying 152→179 (15.1%) and total_prs varying 1000→1900 (47.4%) — and its own workflow prompt (daily-regulatory.md:238) hardcodes open_issues as "SAME SCOPE, should match within 5-10%" across all reports. Live verification shows these aren't real regressions: two source workflows compute a windowed/capped value but label it with a metric name that scratchpad/metrics-glossary.md defines as absolute, all-time, no filter:

  • .github/workflows/daily-issues-report.md:164 — open_issues = len(df[df['state'] == 'OPEN']), where df is capped to "up to 1000 issues" (line 82). The report's own body ([daily issues] Daily Issues Report - 2026-09-25 #63454) admits this only spans ~5 days when volume is high ("the underlying dataset does not extend back a full 7 or 30 days"). The glossary (scratchpad/metrics-glossary.md:41-53) defines open_issues as "All open issues in repository, no time filter" and explicitly lists Daily Issues Report as a user of that absolute definition.
  • .github/workflows/daily-performance-summary.md:209 (rendered as "Total PRs" at line 473) — 'total': len(created_prs), where created_prs is filtered to createdAt >= window_start (a rolling 90-day window, lines 90-102). The glossary (scratchpad/metrics-glossary.md:203-213) defines total_prs as "Total count of pull requests in the repository across all states... no filters applied" and lists Daily Performance Summary as a user of that absolute definition.

Because both workflows silently repurpose an "absolute" glossary metric name for a windowed/capped computation, the Daily Regulatory Report's naive same-name comparison produces false "critical discrepancy" alarms, eroding trust in a monitor that has a good track record of catching real bugs (see prior fixes from regulatory-report findings, e.g. #55519).

Expected Impact

Either fix the two source workflows to compute what the glossary promises (absolute, unfiltered counts) for these specific field names, or rename their windowed variables (e.g. open_issues_in_sample, prs_created_90d) and update scratchpad/metrics-glossary.md to document the intentional scope split — either way, this removes a recurring false-positive class from the Daily Regulatory Report and restores open_issues/total_prs as genuinely comparable across reports.

Suggested Agent

Daily Regulatory Report workflow or a general code-quality/refactor agent — the fix is confined to two .md workflow prompt files plus a glossary doc update.

Estimated Effort

Medium (1-4 hours) — requires deciding rename vs. re-scope for each field, updating both workflow prompts, and syncing scratchpad/metrics-glossary.md.

Data Source

DeepReport Intelligence Briefing analysis run, 2026-09-26 (incremental cycle since prior briefing #63485), cross-referencing Daily Regulatory Report #63513, Daily Issues Report #63454, and Daily Performance Summary #63512.

Generated by 🔬 Deep Report · claude · agent · 226 AIC · ⌖ 9.45 AIC · ⊞ 13K · ◷

  • expires on Sep 27, 2026, 5:07 PM UTC-08:00

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions