Summary
Status: Mixed — auto-labelling engine is healthy, correction pipeline is fully stalled.
The Label Discussions workflow ran successfully every day over the last 30 days. However, the Labelling Correction Collector has zero runs in the same window, meaning no new correction signals have been generated since mid-April 2026. All 333 open correction signals are 40–49 days old with no resolution path active.
Key Metrics
| Metric |
Value |
Notes |
| Discussions reviewed — last 7 days |
≥ 1 (conservative floor) |
Only 1 of 7 daily summary issues was available in the dataset; the single parseable summary (2026-05-27) shows 1 reviewed |
| Label changes applied — last 7 days |
≥ 1 (conservative floor) |
The same summary shows 1 change (Billing & Payment Issues added to #285) |
| Change rate — last 7 days |
~100% for the one parsed day |
Insufficient data across the full 7-day window to compute a meaningful rate |
| Correction-collector runs — last 7 days |
0 |
No Labelling Correction Collector runs found in the entire 30-day window |
| Correction Feedback runs — last 7 days |
7 (all skipped) |
Runs daily but skips; no corrections are flowing through the pipeline |
| Open correction signals |
333 |
332 are 40–49 days old; 1 is 30–39 days old |
| Signals created — last 7 days |
0 |
|
| Signals created — last 30 days |
0 |
All signals predate the 30-day window; last batch was ~April 8–18, 2026 |
⚠️ Only 1 daily summary issue was present in the dataset. The reviewed/changed counts are a floor estimate. With 7 successful Label Discussions runs in the window, actual throughput is likely higher but cannot be confirmed without the missing summary issues.
Correction Pressure
The correction backlog is concentrated, not spread. 333 open signals cluster heavily into three categories:
| Category |
Open Signals |
| Copilot Conversations |
110 (33%) |
| GitHub Education |
71 (21%) |
| Other Feature Feedback, Questions & Ideas |
52 (16%) |
| Apps, API and Webhooks |
21 (6%) |
| Mobile |
14 (4%) |
| Discussions |
13 (4%) |
| Actions |
10 (3%) |
The most-corrected labels across open signals are Copilot (56), Copilot in GitHub (42), GitHub Education (38), and bug (37). The split between labeled (197) and unlabeled (136) events indicates bidirectional uncertainty — both incorrect additions and incorrect removals are present.
Several individual discussions show very high correction counts (up to 19 per signal), suggesting persistent labelling disagreement on specific threads.
High-churn correction signals (correction_count ≥ 14)
| Signal |
Category |
Last Label |
Corrections |
#114 — Disc #147 |
Other Feature Feedback |
Code Search and Navigation |
19 |
#69 — Disc #185 |
GitHub Education |
question |
17 |
#63 — Disc #195 |
Other Feature Feedback |
Profile |
17 |
#55 — Disc #189 |
GitHub Education |
GitHub Education Benefits |
17 |
#51 — Disc #186 |
Other Feature Feedback |
Code Search and Navigation |
17 |
#109 — Disc #151 |
Copilot Conversations |
Copilot |
16 |
#75 — Disc #175 |
GitHub Education |
GitHub Education |
15 |
#118 — Disc #148 |
Apps, API and Webhooks |
Apps API and Webhooks |
14 |
#77 — Disc #170 |
Copilot Conversations |
Copilot |
14 |
#73 — Disc #184 |
Other Feature Feedback |
Copilot |
14 |
Open Instruction Debt
The correction backlog is growing stale. All 333 open signals originate from approximately April 8–18, 2026 (Batch 02). None have been closed in the last 30 days; only 2 signals are closed across all time.
- Oldest open signal:
#47 — created 2026-04-08, ≈49 days old
- Open parent intake issues: 4 (all "Batch 02", all open since April 8, 2026) —
#45, #46, #49, #50
- Closed parent intake issues: 1 (Batch 01,
#26)
The root cause appears to be the complete absence of Labelling Correction Collector runs. Without the collector firing, no new signals are generated and the feedback loop cannot engage. The Labelling Correction Feedback workflow runs daily but consistently skips, which is consistent with no new signals arriving. The backlog is not actively growing but is not shrinking either — it is effectively frozen.
Open correction signals by age bucket
| Age range |
Signal count |
| 40–49 days |
332 |
| 30–39 days |
1 |
| < 30 days |
0 |
All signals were created between 2026-04-08 and 2026-04-18.
Open parent intake issues
| Issue |
Title |
State |
Created |
#50 |
Labelling Correction Intake — Batch 02 |
open |
2026-04-08 |
#49 |
Labelling Correction Intake — Batch 02 |
open |
2026-04-08 |
#46 |
Labelling Correction Intake — Batch 02 |
open |
2026-04-08 |
#45 |
Labelling Correction Intake — Batch 02 |
open |
2026-04-08 |
#26 |
Labelling Correction Intake — Batch 01 |
closed |
2026-03-31 |
Recommendations
-
Investigate and re-enable the Labelling Correction Collector. The workflow has not run in at least 30 days. This is the single most impactful gap — without it, no new correction signals enter the system and the feedback loop is broken. Check the workflow schedule, trigger conditions, and any disabling commits.
-
Triage the Batch 02 correction signals before they become stale data. With 333 signals across 4 open intake parents (all ~49 days old), the window for acting on relevant discussion state is narrowing. Prioritise the three dominant categories — Copilot Conversations, GitHub Education, and Other Feature Feedback — which together account for 70% of open signals.
-
Update .github/instructions/community-discussion-labeling.md for high-churn patterns. Several discussions accumulate 14–19 corrections, indicating the current instructions produce inconsistent results for specific topics (e.g., Copilot billing edge cases, GitHub Education eligibility questions, Code Search navigation issues). These are strong candidates for targeted rule clarifications.
-
Audit daily summary issue generation. Only 1 of 7 expected daily summary issues was present in the dataset. If summary issues are not being created reliably, maintainers lose visibility into auto-labelling throughput. Verify that the Label Discussions workflow is consistently publishing summaries and that they are not being silently dropped or expiring too quickly.
References
Recent daily summary issues
| Issue |
Date |
Reviewed |
Changed |
#445 |
2026-05-27 |
1 |
1 |
Only 1 summary issue was available in the dataset. 6 other Label Discussions runs in the last 7 days did not have associated summary issues in the data.
Recent workflow run references
| Workflow |
Run |
Date |
Result |
| Label Discussions |
§84 |
2026-05-27 |
success |
| Label Discussions |
§83 |
2026-05-26 |
success |
| Labelling Correction Feedback |
§422 |
2026-05-27 |
skipped |
| Labelling Correction Collector |
(none in last 30 days) |
— |
— |
References
- §84 — Label Discussions, 2026-05-27
- §83 — Label Discussions, 2026-05-26
- §422 — Labelling Correction Feedback (skipped), 2026-05-27
Generated automatically from health-data.json · workflow run 26553021547 · 2026-05-28
Generated by Labelling Health Report · ● 385.5K · ◷
Summary
Status: Mixed — auto-labelling engine is healthy, correction pipeline is fully stalled.
The
Label Discussionsworkflow ran successfully every day over the last 30 days. However, theLabelling Correction Collectorhas zero runs in the same window, meaning no new correction signals have been generated since mid-April 2026. All 333 open correction signals are 40–49 days old with no resolution path active.Key Metrics
Billing & Payment Issuesadded to#285)Correction Pressure
The correction backlog is concentrated, not spread. 333 open signals cluster heavily into three categories:
The most-corrected labels across open signals are Copilot (56), Copilot in GitHub (42), GitHub Education (38), and bug (37). The split between
labeled(197) andunlabeled(136) events indicates bidirectional uncertainty — both incorrect additions and incorrect removals are present.Several individual discussions show very high correction counts (up to 19 per signal), suggesting persistent labelling disagreement on specific threads.
High-churn correction signals (correction_count ≥ 14)
#114— Disc#147#69— Disc#185#63— Disc#195#55— Disc#189#51— Disc#186#109— Disc#151#75— Disc#175#118— Disc#148#77— Disc#170#73— Disc#184Open Instruction Debt
The correction backlog is growing stale. All 333 open signals originate from approximately April 8–18, 2026 (Batch 02). None have been closed in the last 30 days; only 2 signals are closed across all time.
#47— created 2026-04-08, ≈49 days old#45,#46,#49,#50#26)The root cause appears to be the complete absence of
Labelling Correction Collectorruns. Without the collector firing, no new signals are generated and the feedback loop cannot engage. TheLabelling Correction Feedbackworkflow runs daily but consistently skips, which is consistent with no new signals arriving. The backlog is not actively growing but is not shrinking either — it is effectively frozen.Open correction signals by age bucket
All signals were created between 2026-04-08 and 2026-04-18.
Open parent intake issues
#50#49#46#45#26Recommendations
Investigate and re-enable the Labelling Correction Collector. The workflow has not run in at least 30 days. This is the single most impactful gap — without it, no new correction signals enter the system and the feedback loop is broken. Check the workflow schedule, trigger conditions, and any disabling commits.
Triage the Batch 02 correction signals before they become stale data. With 333 signals across 4 open intake parents (all ~49 days old), the window for acting on relevant discussion state is narrowing. Prioritise the three dominant categories — Copilot Conversations, GitHub Education, and Other Feature Feedback — which together account for 70% of open signals.
Update
.github/instructions/community-discussion-labeling.mdfor high-churn patterns. Several discussions accumulate 14–19 corrections, indicating the current instructions produce inconsistent results for specific topics (e.g., Copilot billing edge cases, GitHub Education eligibility questions, Code Search navigation issues). These are strong candidates for targeted rule clarifications.Audit daily summary issue generation. Only 1 of 7 expected daily summary issues was present in the dataset. If summary issues are not being created reliably, maintainers lose visibility into auto-labelling throughput. Verify that the Label Discussions workflow is consistently publishing summaries and that they are not being silently dropped or expiring too quickly.
References
Recent daily summary issues
#445Recent workflow run references
References
Generated automatically from
health-data.json· workflow run 26553021547 · 2026-05-28