[integrity] DIFC Integrity-Filtered Events Report — 2026-03-21 #22114
Closed
Replies: 1 comment
-
|
This discussion has been marked as outdated by Daily DIFC Integrity-Filtered Events Analyzer. A newer discussion is available at Discussion #22253. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Executive Summary
In the last 7 days, 1,537 DIFC integrity-filtered events were detected across 95 workflow runs. The most frequently filtered tool was
list_issues(1,134 events, 73.8% of all events), and the dominant filter reason was integrity (1,527 events, 99.3%). The AI Moderator and Auto-Triage Issues workflows together account for over 70% of all filtering activity.The filter rate shows a sharp peak on 2026-03-20 (984 events) driven primarily by high-volume AI Moderator and Auto-Triage runs that batch-enumerate issue lists. The underlying pattern is consistent: scheduled and event-triggered workflows attempt to read issue/PR lists or individual items authored by external contributors, all of which carry
integrity_tags: ["none:all"]. Only 10 events (0.65%) were secrecy-filtered, indicating the system is overwhelmingly blocking on integrity rather than confidentiality.Key Metrics
none:all📈 Events Over Time
The 7-day window shows a highly non-uniform distribution: the vast majority of events (984, 64%) occurred on 2026-03-20, with two lighter activity clusters on 2026-03-18/19 and 2026-03-25/26/27. The 2026-03-20 spike correlates with multiple concurrent high-volume runs of AI Moderator and Auto-Triage Issues, each enumerating 20+ issues per run. The relative quiet on 2026-03-21–24 suggests a scheduling gap or maintenance window.
🔧 Top Filtered Tools
list_issuesdominates at 1,134 events — nearly 3× more thansearch_issues(213). This reflects that AI Moderator and Auto-Triage workflows fetch full issue lists and each item with a contributor'snone:allintegrity tag generates a separate filtered event.issue_read(143 events) comes from Issue Monster's targeted reads of specific known-low-integrity issues.pull_request_read(24) andsearch_pull_requests(7) are minority cases from AI Moderator and PR Triage Agent.🏷️ Filter Reasons and Workflows
99.3% of filtering is integrity-based (tag
none:all), meaning content from external contributors that has not yet been approved. Only 10 events (0.65%) involve secrecy filtering (privatetag) onsearch_issuesandsearch_pull_requests, occurring in Workflow Health Manager, Contribution Check, Architecture Diagram Generator, and Auto-Triage Issues.📋 Per-Workflow Breakdown
📋 Per-Server Breakdown
All filtered events originate from the
githubMCP server — no other server contributed any integrity-filtered events during this period.💡 Tuning Recommendations
AI Moderator — reduce list_issues blast radius: AI Moderator generates 640 filtered events (42% of total) by calling
list_issueswhich returns 20+ items per page, each triggering a separate filter event. Consider adding a pre-filter step that checks contributor association before callinglist_issues, or configuring the workflow to only list issues from approved actors. This single change could eliminate ~40% of all filtering noise.Auto-Triage Issues — same pattern: Auto-Triage Issues contributes 436 events (28%) with identical behavior. Both AI Moderator and Auto-Triage should be reviewed together. Adding an integrity promotion step (e.g., a human-approved label or a triage gate) for new issues before they enter the batch list would substantially reduce filtering.
Issue Monster — persistent low-integrity issues: Issue Monster runs every 20 minutes and repeatedly encounters the same low-integrity issues (e.g.,
#21935,#22058,#22092). These issues appear to never be approved. Consider either (a) raising their integrity level after initial human review, or (b) configuring Issue Monster to skip resources withnone:allintegrity rather than retrying them every cycle.secrecy-filtered
search_issues/search_pull_requests(10 events): Four workflows attempt queries that returnprivate-tagged results. Review whether these workflows genuinely need access to private-tagged search results or can operate with public-only data. If private access is required, configure the relevant workflows to explicitly request it via DIFC policy rather than triggering ad-hoc filtering.Investigate the 2026-03-20 spike: The 984-event day warrants closer examination. Multiple high-volume runs coincided. If there was a change to issue volume, contributor activity, or workflow trigger frequency on that date, it should be documented as a known spike rather than ongoing baseline.
Issue Triage Agent and Sub-Issue Closer (211 combined events): These two workflows (from large temp files) generated significant filtering volume with deeply nested list operations. Review whether their list enumeration depth can be bounded or short-circuited when encountering
none:alltagged resources.Generated by the Daily Integrity Analysis workflow
Analysis window: Last 7 days (2026-03-18 to 2026-03-27) | Repository: github/gh-aw
Run: §23375775703
Beta Was this translation helpful? Give feedback.
All reactions