[Labelling Health] Labelling Health Report — 2026-04-01

[mnkiefer]

Report period: last 7 days (2026-03-25 → 2026-04-01) · generated 2026-04-01T03:32Z

---

Summary

Status: Mixed — system newly operational, no trend baseline yet

All labelling activity is concentrated on a single day (2026-03-31), indicating the system was recently launched or re-activated. Auto-labelling runs completed successfully with a healthy change rate (~60–65%), but there is no previous-period data to establish a direction of travel. Two correction signals are open with zero resolved, so the backlog is brand-new rather than stale.

---

Key Metrics

Metric	Value	Notes
Discussions reviewed — last 7 days	~50 (est.)	See detail below; one summary parsed cleanly (10), two body-parsed with overlapping discussion ranges (~40)
Label changes applied — last 7 days	~32 (est.)	6 clean + 26 body-parsed from non-overlapping run
Change rate — last 7 days	~64%	est. 32/50; clean-parse rate 60% (6/10)
Previous 7-day comparison	⚠️ No data	System has no runs in the 2026-03-18 → 2026-03-25 window
Correction-collector runs — last 7 days	3 (all success)	All on 2026-03-31
Open correction signals	2	Both created 2026-03-31; none resolved

⚠️ Two of the three daily summary issues used non-standard body formats that prevented automatic reviewed/changed field extraction. Figures above are partly body-parsed estimates. Precision will improve as the summary format stabilises.

---

Correction Pressure

Correction pressure is low in volume but immediately present from the first day of operation. Both open signals were generated on 2026-03-31 and are parented under a single intake batch (#26 "Labelling Correction Intake Batch 01"), suggesting pressure is concentrated rather than spread.

Signal breakdown:

Signal	Discussion	Category	Current Labels	Correction Count	Latest Label Event
#29	#98 — How do I debug GitHub Actions matrix builds failing only on arm64?	Other Feature Feedback, Questions & Ideas	Actions, Code Search and Navigation, question	2	Actions labeled
#27	#118 — Zero support from Github	Enterprise	bug, Enterprise	1	bug labeled

The highest-pressure item is signal #29 (correction_count=2): Discussion #98 was initially labelled with Actions by one run, then also received Code Search and Navigation — an unexpected second label for a clearly Actions-focused question. This suggests a possible over-labelling pattern or a labelling instruction gap around the distinction between Actions and navigation/search topics.

Signal #27 touches the Enterprise category, which is not well-represented in the correction instruction set; the Bug label may be appropriate but the Enterprise category label is a secondary concern warranting review.

All raw correction pressure flows through the single parent intake #26.

---

Open Instruction Debt

The correction backlog is brand-new (< 24 hours old) and contains 2 open signals with 0 resolved. It is too early to characterise it as growing or shrinking. However, the fact that 2 signals appeared within the first day of operation suggests recurring edge cases will surface quickly.

• Oldest open signal: #27, created 2026-03-31T09:52Z (~18 hours old)
• Open parent intake issues: 1 (Batch 01, #26)
• Backlog status: Actionable — both signals are fresh, clearly scoped, and linked to a single intake parent

The backlog will become stale if not triaged within the next few days. The 5 skipped Labelling Correction Feedback runs (out of 6 total) suggest that most label-event triggers did not meet the feedback criteria, which is expected for a new system but worth monitoring.

Daily summary issue breakdown

Issue	Date	Run	Reviewed	Changed	Change Rate	State	Notes
#31	2026-03-31	23818828336	10	6	60%	Open	✅ Clean parse; covers discussions #142–151
#28	2026-03-31	23793744435	40	26	65%	Closed	⚠️ Body-parsed (non-standard format); covers discussions #92–131; superseded by same-day re-run
#25	2026-03-31	23789074117	30	28	93%	Closed	⚠️ Body-parsed; covers discussions #92–130; subset overlap with #28 — likely superseded

Issues #25 and #28 cover the same discussion number range; #28 appears to be a broader re-run that supersedes #25. The 50/32 estimate above uses #28 (not #25) to avoid double-counting.

Open correction signal breakdown

Signal #29 — Discussion #98

• Title: "How do I debug GitHub Actions matrix builds failing only on arm64?"
• Category: Other Feature Feedback, Questions & Ideas
• Current labels: Actions, Code Search and Navigation, question
• Correction count: 2 (highest pressure signal)
• Latest event: Actions labeled
• Concern: Code Search and Navigation appears to have been added erroneously to a pure GitHub Actions debugging question

Signal #27 — Discussion #118

• Title: "Zero support from Github"
• Category: Enterprise
• Current labels: bug, Enterprise
• Correction count: 1
• Latest event: bug labeled
• Concern: Enterprise category may require clarification in labelling instructions

Parent: #26 — Labelling Correction Intake Batch 01 (open, created 2026-03-31)

Recent workflow run references

Workflow	Run #	Date	Conclusion	Event
Label Discussions	#23	2026-03-31T20:51Z	success	schedule
Label Discussions	#22	2026-03-31T10:56Z	success	workflow_dispatch
Label Discussions	#21	2026-03-31T08:57Z	success	workflow_dispatch
Labelling Correction Collector	#10	2026-03-31T11:35Z	success	repository_dispatch
Labelling Correction Collector	#9	2026-03-31T11:35Z	success	repository_dispatch
Labelling Correction Collector	#7	2026-03-31T09:52Z	success	repository_dispatch
Labelling Correction Feedback	#17	2026-03-31T11:37Z	success	issues
Labelling Correction Feedback	#18	2026-03-31T20:56Z	skipped	issues
Labelling Correction Feedback	#16	2026-03-31T11:35Z	skipped	issues

---

Recommendations

1. Triage the 2 open correction signals in Batch 01 (#26). Signal #29 (correction_count=2) is the most pressing: investigate whether .github/instructions/community-discussion-labeling.md needs a clarification that Code Search and Navigation should not be applied to GitHub Actions debugging questions. Close the signal once the instruction or labelling is corrected.

2. Standardise the daily summary issue body format. Two of three summary issues used divergent formats (Discussions updated: vs Total label changes applied: vs table-based counts), preventing automatic metric extraction. Aligning the Label Discussions workflow output to a single consistent format will improve future health report precision.

3. Establish a baseline by running the scheduled Label Discussions workflow consistently for the next 7 days. With only one day of data, trend analysis is not possible. Consistent daily scheduled runs will populate the previous-7-day window and enable meaningful week-over-week comparisons.

4. Review the Enterprise category coverage in the labelling instructions. Signal #27 touches the Enterprise discussion category. If the instructions do not have clear guidance for Enterprise-category discussions, add a rule to .github/instructions/community-discussion-labeling.md covering when Enterprise as a topic label should or should not be applied alongside type labels like bug.

---

References

• §23 — Label Discussions scheduled run (latest)
• §10 — Labelling Correction Collector (batch 01)
• §17 — Labelling Correction Feedback (only successful feedback run)

Generated by Labelling Health Report · ◷

• expires on May 1, 2026, 3:44 AM UTC