🏥 Repo Health Dashboard

[github-actions]

🏥 ML.NET Repository Health Dashboard

Last Updated: 2026-06-11 07:31 UTC
Overall Status: 🟡 Warning

Counts: 4 findings (1 critical, 3 warnings) | 0 items baselined

---

📊 Executive Summary

Repository health remains in warning state. The P0 issue (#5805) is still open (Day 2 of active tracking). A new warning was identified today: 4+ stale PRs with no activity in >30 days. Two new PRs were opened since yesterday (#7625, #7623). CI/CD workflows remain healthy. Issue #6978 (blocking-clean-ci) was updated today, potentially indicating ongoing TorchSharp CI instability.

Key Observations:

• 🔴 P0 MKLImports PDB not included with packages #5805 aged 5+ years — stalled; investigation complete, re-triage needed
• 🟡 NEW: Stale PRs — 4+ PRs (>30d inactivity), approaching warning threshold
• 🟡 Untriaged backlog — 221+ issues with untriaged label, 821+ without milestone
• ⚠️ Issue Microsoft.ML.TorchSharp.Tests.QATests.TestSimpleQA followed by process killed / return 137 #6978 (blocking-clean-ci) updated today — TorchSharp test killed (OOM)
• ✅ CI workflows stable — backport.yml and locker.yml both healthy
• i️ 2 new PRs opened since yesterday (Add public SentencePieceTokenizer factory methods for Unigram from vocab list and tokenizer.json #7625 SentencePiece tokenizer, Add report-green AzDO pipeline so Build Analysis completes on build-excluded PRs #7623 report-green pipeline)

---

🔍 Findings

🔴 Critical

Status	Finding	Details
📌 EXISTING (Day 2)	P0 issue #5805 open for 5+ years (investigated ↗)	#5805 "MKLImports PDB not included with packages" — created 2021-05-25, assigned to @michaelgsharp. Root cause identified: manual build friction + missing CI enforcement. Action needed: re-triage or close as won't-fix.

🟡 Warnings

Status	Finding	Details
🆕 NEW	Stale PRs (4+ with >30 days no activity) (investigated ↗)	#7609 (37d, arm64 SymSgd), #7416 (>15mo, torchsharp update), #7406 (>15mo, GenAI quantization), #6664 (>3yr, DataFrame math).
📌 EXISTING (Day 2)	221+ untriaged issues (investigated ↗)	View untriaged. Mix of product bugs and workflow automation issues. Recommendation: exclude agentic-workflows / repo-health issues from count.
📌 EXISTING (Day 2)	821+ issues without milestone (investigated ↗)	View no-milestone. Triage gap — ~60% are from 2024. Recommendation: milestone sweep targeting oldest 100.

i️ Informational

Metric	Value	Change
P0 issues	1 (100% assigned)	→
P1 issues	23	→
Bug issues	59+	→
Open PRs	17	↑ +3 vs Day 1
Blocking CI issues	1 (#6978, updated 2026-06-11)	⚠️
Backport workflow	✅ Latest: success	→
Locker workflow	✅ Run #923: success (2026-06-11)	→
AzDO Pipelines	⚠️ Not monitored (no AZDO_PAT)	—

---

✅ Recently Resolved

No findings resolved since last run (2026-06-10)

---

📋 Known Baseline

Baseline file not found — .github/health-baseline.md does not exist.

To baseline known long-term items (e.g., the P0 issue if accepted as low-priority), create .github/health-baseline.md with their fingerprints. Baselined findings are suppressed from critical/warning tables.

---

📈 Trends (7-day)

Metric	2026-06-10	2026-06-11	Change
P0 issues	1	1	→
P1 issues	23	23	→
Open PRs	14	17	↑ +3
Untriaged	221+	221+	→
No milestone	821+	821+	→
CI health	✅	✅	→
Total findings	3	4	↑ +1

---

🔗 Configuration

Repository: dotnet/machinelearning
Tracked GitHub Actions: backport.yml, locker.yml
Azure DevOps: ⚠️ Monitoring disabled — AZDO_PAT not configured (dnceng/public: vsts-ci, codecoverage-ci, night-build, outer-loop-build)
Priority Labels: P0, P1, P2, P3, bug
Baseline: .github/health-baseline.md (not found)

Investigators Dispatched this run: 1 (stale PRs — P5-stale-prs)

---

Last Workflow Run: #27331139767

This dashboard is automatically updated daily by the repo-health-check workflow.

---

Generated by Repo Health Check — Orchestrator

Note

🔒 Integrity filter blocked 146 items

The following items were blocked because they don't meet the GitHub integrity level.

• large models loading with mlcontext.model.load ends up OutOfMemoryException #6948 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• After upgrading to .NET 7.0, I'm getting "PositiveRecall" is not supported. Why? #6472 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• DataFrame.OrderBy(string columnName) does not perform stable sorting! #6443 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• KMeans cluster analysis is non-deterministic when using KMeansYinyang initialization, even with fixed MLContext seed #6375 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Suspicious fragments found by PVS-Studio Static Analyzer #6321 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Error while using Model.Save() #6281 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• VectorType attribute with dynamic dimension while dealing with csv files #6151 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Microsoft.Data.Analysis DataFrame.Join infinite loop #6128 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Fix Recursive Parsing for PFI #6084 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• File path and/or name are too long when saving training model. #6036 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• IndexOutOfRangeException at Microsoft.ML.Data.BufferBuilder`1.AddFeature #6035 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• 'Probability' column missing in output schema of NaiveBayes MulticlassClassification trainer #6033 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• BufferBuilder.MakeDense: IndexOutOfRangeException #6032 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Memory leak when using ML.NET and TensorFlow.NET. #6005 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Cursor.MoveNext() gets stuck #5926 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• Loading time of 2MB ML model is significantly slower in .NET Core compared to .NET Framework #5913 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
• ... and 130 more items

To allow these resources, lower min-integrity in your GitHub frontmatter:

tools:
  github:
    min-integrity: approved  # merged | approved | unapproved | none

Generated by Repo Health Check — Orchestrator · ● 617.9K · ◷

Generated by Repo Health — Dashboard Groomer · ● 3M · ◷

[github-actions]

🔄 Health Check Run Complete — 2026-05-27

Run ID: 26496768080

Summary

✅ First health check run completed successfully

Findings:

• 🔴 Critical: 0
• 🟡 Warning: 3
• ℹ️ Info: 5
• 📋 Baselined: 1

Changes from previous run: N/A (first run)

Investigations Dispatched

Dispatched 3 investigation workflows for warning findings:

1. I1-P2 — 298 P2 issues open — backlog growing
2. I3-untriaged — 214 untriaged issues — triage backlog
3. I3-milestone — 813 issues without milestone

These investigations will analyze the findings in detail and provide actionable recommendations.

Key Metrics

Metric	Value
P0 Issues	1
P1 Issues	23 (23 baselined)
P2 Issues	298
P3 Issues	101
Untriaged Issues	214
Issues w/o Milestone	813
Open PRs	14 (6 baselined)

---

Next run: Scheduled daily. Trend data will be available after the next run.

Generated by Repo Health Check — Orchestrator

[github-actions]

🔍 Investigation: I3-untriaged

Finding: 214 untriaged issues — triage backlog
Severity: high
Category: issue

Evidence

1. Total backlog: 215 issues with untriaged label currently open — [Source: GitHub Issues API search]
2. Age distribution: Average age 1,006 days (~2.8 years), oldest is 1,187 days (3.3 years, issue ResNet101 [ShapeInferenceError] Attribute pads has incorrect size #6580), newest is 887 days (2.4 years in first 100) — [Source: created_at timestamps]
3. Staleness: Average 919 days since last update, with oldest update 1,117 days ago — [Source: updated_at timestamps]
4. Missing triage metadata:
   • 0% have area-* labels (0/100 sampled)
   • 0% have priority labels (0/100 sampled)
   • 1% have milestones (1/100 sampled)
   • 50% have zero comments (50/100 sampled)
   • [Source: labels, milestone, comments fields]
5. Label patterns: 42% have ONLY the untriaged label (42/100), 34% have one additional label (34/100), rest have 3-6 labels — [Source: label counts]
6. Inflow vs outflow imbalance:
   • 112 untriaged issues opened since Jan 2024 (~6-7 per month inflow)
   • 53 issues from 2023 were triaged (removed untriaged label)
   • Net accumulation: backlog growing faster than triage throughput
   • [Source: created date filtering, comparison searches]
7. Most common secondary labels: enhancement (31), documentation (8), Deep Learning (7), onnx (7), question (7) — [Source: label frequency analysis]
8. Duplicate triage flag: 14% also have needs-further-triage label (14/100), suggesting even within untriaged issues, some are flagged for additional review — [Source: co-occurrence analysis]
9. Top issue authors with untriaged backlog: superichmann (15), torronen (14), asmirnov82 (7) — [Source: author grouping on oldest 100 issues]

Sample oldest untriaged issues:

• ResNet101 [ShapeInferenceError] Attribute pads has incorrect size #6580 (1,187 days): ResNet101 ONNX ShapeInferenceError — technical bug, 1 comment
• [Object Detection] Bounding box rotated due to image exif data #6592 (1,180 days): Object detection bounding box rotation bug — technical bug with screenshots, 4 comments
• Improvements when specifying a ModelLoader when adding a predictionenginepool to DI. #6618 (1,145 days): PredictionEnginePool ModelLoader enhancement — feature request, 1 comment

Sample recent untriaged issues:

• Add Quantile Regression support to LightGbmRegressionTrainer #7603 (43 days): Add quantile regression to LightGBM — detailed enhancement with API proposal, 5 reactions
• SdcaLogisticRegression test fails on macOS ARM64 Release — LogLoss exceeds 0.5 threshold #7600 (67 days): SDCA test failure on macOS ARM64 — test failure bug, 0 comments
• Update Windows Server 2019 VM images — end of mainstream support #7596 (70 days): Update Windows Server 2019 VM images — infrastructure task, 0 comments

Root Cause

Confidence: High

The triage backlog is caused by insufficient triage throughput relative to issue inflow, compounded by lack of systematic triage process. Key factors:

1. No area owners assigned: 100% of untriaged issues lack area-* labels, meaning they haven't been routed to domain experts who could triage them
2. No priority assignment: 0% have priority labels, indicating no severity/urgency evaluation
3. Long dormancy: Issues sit for 2-3 years without triage, suggesting no periodic sweep or backlog grooming
4. Inflow exceeds outflow: 6-7 new issues/month accumulate faster than the 2023 triage rate (~4/month from 2023 cohort)

The oldest issues date to Feb 2023, suggesting triage process was active before then but broke down or lost capacity around that time.

Recommendations

Priority	Action	Who
🔴 Immediate	Halt new issue accumulation: Configure GitHub Actions workflow to auto-label new issues with appropriate area-* based on title/body keywords (e.g., "ONNX" → area-onnx, "DataFrame" → area-dataframe). This prevents future untriaged accumulation.	DevOps/Triage Lead
🔴 Immediate	Backlog triage sprint: Dedicate 2-4 hours/week for next 8 weeks to triage oldest 100 issues (age >3 years). Goal: assign area labels, close duplicates/obsolete issues, escalate blockers. Track progress weekly.	Area owners + triage rotation
🟡 Short-term	Monthly triage ritual: Schedule recurring 30-min session on first Friday of month to triage all untriaged issues from previous month. Aim for <30 day triage SLA.	Team standup agenda
🟡 Short-term	Close stale zero-engagement issues: For issues with 0 comments and >365 days old, add "stale-candidate" label and post auto-comment: "This issue has not received engagement. Please comment if still relevant, otherwise will close in 14 days." Auto-close after 14 days silence.	GitHub Actions workflow
ℹ️ Long-term	Triage automation with LLM: Implement GitHub Actions workflow using GPT-4o to analyze new issue body/title and suggest area labels, priority, and whether it's a duplicate. Human reviews suggestion but saves manual categorization.	AI/Automation team
ℹ️ Long-term	Expand area ownership: Current untriaged issues span onnx (7), Deep Learning (7), Microsoft.Data.Analysis (6), AutoML (4+), TorchSharp (2+). Explicitly assign area owners for each domain who are responsible for monthly triage. Document in CODEOWNERS.	Engineering management

Related

• Historical context: Issue alexnet fails with exception #4075 (closed) had similar ONNX ShapeInferenceError, suggesting ResNet101 [ShapeInferenceError] Attribute pads has incorrect size #6580 may be a regression
• Pattern: Many untriaged enhancement requests (31) with detailed proposals (e.g., Add Quantile Regression support to LightGbmRegressionTrainer #7603 quantile regression) that received positive reactions but no triage follow-up — indicates community engagement but no team bandwidth for feature evaluation
• Comparison: 53 issues from 2023 were triaged (removed untriaged label), showing the process worked in the past but capacity dropped significantly in 2024-2026

---

Investigated by repo-health-investigate • Finding: I3-untriaged

Generated by Repo Health — Investigate Finding

[github-actions]

📅 Daily Health Check — 2026-05-28

Run: #26560269202 | Status: 🟡 Warning

Changes Since Last Run (2026-05-27)

New Findings: 11 (first comprehensive run with detailed tracking)

• 🟡 5 warnings: High triage backlog (215 untriaged, 814 no-milestone), 298 P2 issues, 59 bugs
• ℹ️ 7 info: Stable GitHub Actions workflows, 14 open PRs, baseline tracking active

Resolved Findings: 0

Key Metrics:

• Issue velocity (24h): 0 opened, 0 closed — weekend/quiet period
• GitHub Actions: ✅ Backport & Locker workflows passing
• AzDO Pipelines: ⚠️ Monitoring disabled (AZDO_PAT not configured)

Top Priority

1. Triage bottleneck — 215 issues need milestone/priority assignment
2. P2 backlog review — 298 P2 issues may contain stale items for closure
3. Enable AzDO monitoring to track main CI pipelines

No investigation workflows dispatched (0/5 budget used).

Generated by Repo Health Check — Orchestrator

[github-actions]

🔍 Investigation: I3-untriaged

Finding: 215 untriaged issues — identify triage process bottleneck
Severity: High
Category: Issue Backlog

Evidence

1. Historical accumulation — 70 issues from 2024 (33%), 115+ from before 2024. Only 30 from 2025-2026 combined. The bulk of untriaged issues are legacy backlog, not new inflow.

2. Triage agent underutilized — Issue triage agent (.github/workflows/issue-triage.agent.md) exists but only ran once in 2026 (April 14). Not configured to run automatically on issues: opened events despite the workflow definition.

3. Zero engagement rate — 54 of 215 untriaged issues (25%) have zero comments, indicating they've never been reviewed by maintainers or community.

4. Triage velocity insufficient — Only 18 issues created since 2025 have been triaged and had "untriaged" removed, while 30+ from the same period remain untriaged. Triage throughput is roughly 40% of inflow.

5. Label removal only via milestone — Per .github/policies/untriaged.yml, the "untriaged" label is only automatically removed when an issue is added to a milestone or closed. Adding area/priority labels doesn't remove it, so even issues with enhancement, bug, or area labels remain marked "untriaged".

6. Enhancement backlog — 32+ enhancement requests are marked untriaged. Many appear valid (e.g., How to export ML.NET models for use in other programming languages? #7489: "How to export ML.NET models for other languages?" with 598 comments) but lack area labels or milestones.

Root Cause

Confidence: High

The bottleneck is insufficient automation and manual triage capacity. The triage agent exists but isn't running on new issues. The policy auto-adds "untriaged" but removal requires milestone assignment, which is a manual step that doesn't scale. Result: issues accumulate in untriaged state even after receiving labels like enhancement or bug.

The 70+ issues from 2024 represent a one-year backlog that hasn't been processed. Current triage velocity (40% of inflow) means the gap widens over time.

Recommendations

Priority	Action	Who
🔴 Immediate	Enable automatic triage agent runs by adding workflow_dispatch trigger to .github/workflows/issue-triage.agent.lock.yml and setting up scheduled runs (weekly) to process backlog	@dotnet/mlnet-maintainers
🔴 Immediate	Update .github/policies/untriaged.yml to remove "untriaged" label when area/priority labels are added, not just milestones	@dotnet/mlnet-maintainers
🟡 Short-term	Bulk triage sprint: Run triage agent against all 215 untriaged issues to apply initial labels. Close stale issues (>2 years old, zero engagement) as "needs reproduction"	@dotnet/mlnet-maintainers
🟡 Short-term	Add triage capacity: Recruit 2-3 community members as "triage collaborators" with label permissions. Document triage process in CONTRIBUTING.md	Repo owner
ℹ️ Long-term	Implement triage SLO: 90% of new issues triaged within 7 days. Monitor via repo health dashboard	@dotnet/mlnet-maintainers

Related

• Triage policy: .github/policies/untriaged.yml
• Triage agent: .github/workflows/issue-triage.agent.md (exists but underused)
• Last triage agent run: #24379157772 (April 14, 2026)
• Example high-activity untriaged issue: How to export ML.NET models for use in other programming languages? #7489 (598 comments, still marked untriaged)
• Example zero-engagement issues: TreeEnsembleRegressor is deprecated in ONNX #7581, erfInv is very inneficient #7568, FeaturizeText: System.ArgumentOutOfRangeException: Schema mismatch for input column 'Text_CharExtractor': expected Expected known-size vector of Single, got Vector<Single> #7529, MulticlassClassificationExperiment preFeaturizer not acting on the schema #7522 (all 0 comments, months old)

Estimated Impact of Fixes

• Enabling automation + policy fix: Reduce new issue triage time from ~14 days to ~2 days
• Backlog sprint: Clear 150+ legacy untriaged issues in 1-2 weeks
• Community triage team: Increase triage throughput 3-5x

---

Investigated by repo-health-investigate • Finding: I3-untriaged

Generated by Repo Health — Investigate Finding

[github-actions]

🔍 Investigation: I1-P2-count

Finding: 298 P2 issues — review backlog for stale/closable items
Severity: medium
Category: issue

Evidence

1. Total stale backlog: All 298 P2 issues have not been updated in 5+ years (last update between May 2019 - July 2020) — Source: GitHub Search API, sorted by updated_at
2. Zero engagement: 110 issues (37%) have no comments and no reactions, indicating complete lack of community engagement — Source: Aggregated reaction/comment counts from API
3. Average staleness: Mean of 2,342 days stale (~6.4 years) across all 298 issues — Source: Calculated from updated_at timestamps
4. Triage labels present but ineffective: 7 issues tagged need info (oldest: The curious case of TrainedWrapperEstimatorBase and friends #2841, 2562 days stale), 5 tagged question (oldest: What is the CLR bug mentioned in Div64Core #3726, 2563 days stale) — no follow-up action taken
5. Issue type distribution: 173 enhancement requests (58%), 28 API changes (9%), 23 AutoML.NET (8%), 14 bugs (5%), 12 documentation (4%), 12 tests (4%) — Source: Label aggregation
6. No resolution activity: Zero closed PRs with P2 label in repository history — Source: GitHub Search for closed PRs

Sample of oldest stale, unengaged issues (0 comments, 0 reactions, unassigned, no milestone):

• SaveAsText documentation doesn't address scope of separator argument #2512: SaveAsText documentation (2563 days stale)
• Perf: Optimization for DoubleParser (VTune) #3007: Perf optimization for DoubleParser (2563 days stale)
• What is the CLR bug mentioned in Div64Core #3726: Question about CLR bug in Div64Core (2563 days stale)
• The curious case of TrainedWrapperEstimatorBase and friends #2841: Code sanitation question about TransformWrapper (2562 days stale)
• Avoids pointer overflow in bound checking of SSE/AVX intrinsics #980: Pointer overflow in SSE/AVX intrinsics (2524 days stale)

Root Cause

Confidence: High

The P2 backlog represents abandoned triage state, not active work planning. P2 was defined as "Needs to be fixed at some point" (per label description), but with 100% of issues stale 5+ years and zero progress, this priority level has effectively become a holding pen for deferred/postponed work with no mechanism for:

1. Periodic review — No evidence of backlog grooming or re-triage
2. Promotion criteria — Issues never promoted to P0/P1 or closed as wontfix
3. Community engagement — 37% have zero interaction, indicating low discoverability or interest

The issue types (58% enhancements, 9% API changes) suggest these were aspirational features from 2018-2019 era that never gained traction post-initial filing.

Recommendations

Priority	Action	Who
🔴 Immediate	Audit closure sweep: Sample 20-30 issues across categories (docs, questions, "need info") for closure. Close issues where: (a) question answered elsewhere, (b) docs now exist, (c) blocked on unavailable info, (d) superseded by newer APIs	Triage team
🟡 Short-term	Batch re-triage: Review remaining P2 issues by category. For each: promote to P1 if still relevant, close as wontfix if no longer aligned with roadmap, convert enhancements to discussions for community input	Maintainers + area owners
🟡 Short-term	Deprecate P2 label: Replace with backlog or future label to set accurate expectations. P2 implies "will fix soon" but reality is "unlikely to be addressed"	Triage team
ℹ️ Long-term	Quarterly backlog review: Implement automation to flag issues stale >1 year for re-triage. Add policy: issues stale >2 years with no engagement auto-close with "wontfix" + explanation	DevOps + maintainers

Related

• Dashboard issue: 🏥 Repo Health Dashboard #7612
• Comparable patterns: Check if P1 and P3 backlogs exhibit similar staleness
• Community impact: High noise-to-signal ratio makes it harder for contributors to find actionable issues (current up-for-grabs count in P2: 9 issues, all 4+ years stale)

Immediate Next Steps (Actionable Example)

Quick wins for closure (high confidence):

1. Questions answered: What is the CLR bug mentioned in Div64Core #3726 (CLR bug question), Shall we add option to OneHotEncoding to treat each slot separately? #3111 (OneHotEncoding API question), Question in the Text Encoding when doing mlContext.Data.ReadFromTextFile #2736 (text encoding question) — likely resolved by docs or newer API versions
2. Need info with no response: The curious case of TrainedWrapperEstimatorBase and friends #2841 (code internals question, 0 comments, internal-only concern)
3. Superseded by AutoML/newer APIs: Many 2019-era enhancement requests pre-date AutoML.NET v2 and GenAI packages

---

Investigated by repo-health-investigate • Finding: I1-P2-count • Run: #26560566647

Generated by Repo Health — Investigate Finding

[github-actions]

📊 Daily Health Delta — 2026-05-29

Changes Since Last Run (2026-05-28):

New Findings

• P5-stale-prs 🆕 — 5 stale PRs identified (no activity > 30 days): update torchsharp to 0.105.0 #7416, [GenAI] Use BitsAndBytes for 4bit quantization. #7406, Update Projects to .NET 8 in MLNET 4.0 Branch #6749, (WIP) Generic DataFrame Math #6664, Add DataViewSchema overloads to ConvertToOnnx #6449
• I2-velocity-24h 🆕 — Issue velocity flat: 0 opened, 0 closed in last 24h

Existing Findings (Day 2)

• All warning findings persist: 100+ P2 issues, 59 bugs, 100+ untriaged, 814 without milestones, 23 P1 issues

Resolved Findings

• None

Key Metrics

• No milestone issues: 814 (unchanged)
• Workflows: Both backport & locker remain stable (100% success rate)
• PR count: 14 open (unchanged)

Recommended Actions

1. Review stale PRs — determine if they should be closed or refreshed
2. Continue monitoring triage backlog

---

Dashboard updated automatically by repo-health-check workflow

Generated by Repo Health Check — Orchestrator

[github-actions]

📅 Daily Health Check — May 30, 2026

Status: 🟡 WARNING (no change)
Run: #26677390163

Summary

Comprehensive health check completed. Dashboard fully updated with enhanced baseline tracking.

Changes Since Last Run

• ✅ Enhanced Dashboard: Complete rewrite with detailed baseline tracking
• 📊 Baseline Alignment: All 24 P0/P1 issues and 6 PRs verified against baseline file
• 📈 Metrics: Added detailed breakdowns (17.5% with milestones vs 82.5% without)
• 🎯 Recommendations: Added actionable next steps for triage and monitoring

Key Findings

• 0 Critical (non-baselined) — All P0/P1 items properly documented
• 1 Warning — 814 issues (82.5%) without milestones (unchanged)
• Active Workflows — Backport and locker automation running successfully
• High PR Activity — ~200 open PRs, normal for repository scale

Investigators Dispatched

None — All critical/high priority items are baselined or informational. No automated investigation required.

Next Actions

1. ⚠️ Triage Priority: Schedule sessions to address 814 unmilestoned issues
2. 📊 Trend Monitoring: Next run will establish velocity baselines
3. 🔧 AzDO Config: Add AZDO_PAT secret to enable main CI pipeline monitoring

---

Trend: Stable baseline | Velocity: N/A (first enhanced run) | Dispatches: 0/5

Generated by Repo Health Check — Orchestrator

[github-actions]

📅 Daily Health Summary — 2026-05-31

Overall Status: 🟡 WARNING

Changes Since Last Run

• First comprehensive baseline run completed
• No previous findings to compare against

Current State

• 🟡 2 warnings identified
• ℹ️ 11 informational metrics tracked
• 🔴 0 critical issues

Key Metrics

• Untriaged Issues: 216 (requires triage process attention)
• P1 Unassigned: 16 issues (need owner assignment)
• Open PRs: 14 active pull requests
• CI Health: ✅ All workflows stable (100% success)

Actions Taken

• Dashboard updated with current health snapshot
• Baseline findings saved for next run comparison
• No investigation workflows dispatched (first run establishes baseline)

---

Next scheduled run: 2026-06-01 (daily)

Generated by Repo Health Check — Orchestrator

[github-actions]

📊 Daily Health Update — 2026-06-01

Summary: Repository health remains at 🟡 WARNING level. 1 new finding identified, 0 findings resolved.

Changes Since Last Run (2026-05-31)

🆕 New Findings:

• 815 issues without milestone — Large number of issues lack milestone planning

📌 Existing Findings (unchanged):

• 216 untriaged issues — Backlog continues to require triage attention

✅ Resolved: None

Key Metrics

Metric	Change
Untriaged issues	216 → 216 (→ stable)
No milestone	— → 815 (🆕 new metric)
Open PRs	14 (→ stable)
CI Health	✅ 100% (→ stable)

Activity (24h): 0 issues opened, 0 closed, 0 PRs merged — typical low-activity period.

---

View full dashboard above ⬆️ | Workflow run

Generated by Repo Health Check — Orchestrator

[github-actions]

🔍 Investigation: I4-7b5d3a9e

Finding: 815 issues without milestone assignment
Severity: high
Category: issue

Evidence

1. Confirmed count: 815 open issues lack milestones — Source: GitHub Search API (is:open no:milestone)
2. 94% are labeled "untriaged" — 94 of 100 sampled issues have the "untriaged" label, indicating they have not received human review
3. Recent creation dates dominate — In a 50-issue sample: 18% from 2026, 50% from 2025, 32% from 2024. All are recent issues (post-2024)
4. Automated triage adds labels but NOT milestones — .github/workflows/issue-triage.agent.md (active workflow) only assigns labels: bug, enhancement, question, documentation, perf, test, Build, untriaged, needs-further-triage, need info. Milestone assignment is not part of the workflow.
5. Even manually triaged issues lack milestones — Example: Support 16 KB page sizes #7517 (assigned to Copilot, labeled enhancement,need info, 6 comments, 8 months old) and Add docs (tutorial) about transfer image from RAM to pipeline (model) #7449 (labeled enhancement,up-for-grabs,documentation, 2 comments, 13 months old) still have no milestone
6. No automated milestone assignment exists — Investigation found no workflows or automation that assign milestones at any stage of the issue lifecycle

Root Cause

Confidence: High

The repository has no automated milestone assignment process. The active issue-triage workflow (triggered on new issues) only assigns labels and the "untriaged" marker. Milestone assignment is a manual step that requires human intervention, but there is no defined process or workflow to ensure this happens.

The backlog of 815 issues represents the entire open issue inventory accumulating since automated triage began, with human milestone assignment not keeping pace with issue creation velocity. Issues may be labeled, discussed, even assigned (like #7517), but still remain without a milestone indefinitely.

Recommendations

Priority	Action	Who
🔴 Immediate	Create a milestone assignment workflow using GitHub Agentic Workflows. Define default milestones (e.g., "Triage", "Backlog", "Future") and auto-assign based on label combinations (e.g., bug → "Triage", enhancement+need info → "Future", up-for-grabs → "Backlog").	Repo maintainers
🟡 Short-term	Bulk-assign milestones to the backlog. Run a one-time script or workflow to assign all 815 existing issues to a "Backlog" or "Needs Triage" milestone, so the baseline is clean.	Repo maintainers
🟡 Short-term	Enhance issue-triage.agent.md to include milestone assignment logic in the initial triage step. Modify safe-outputs to allow milestone assignment and add instructions for AI to suggest or assign milestones.	Repo maintainers
ℹ️ Long-term	Document the milestone workflow in CONTRIBUTING.md or a new TRIAGING.md file. Define what each milestone means, when issues should be promoted/demoted, and who is responsible for milestone grooming.	Repo maintainers

Related

• 🏥 Repo Health Dashboard #7612 — Repo Health Dashboard (where this investigation was requested)
• Active workflow: .github/workflows/issue-triage.agent.lock.yml (labels but no milestones)
• Example triaged issues without milestones: Support 16 KB page sizes #7517, Add docs (tutorial) about transfer image from RAM to pipeline (model) #7449, Docs: HuggingFace (NLP) Migration Guide #7384, Microsoft.ML.Tests.OnnxConversionTest.NaiveCalibratorOnnxConversionTest Failed #7057, ML.NET packages contain .dbg and .dwarf files now #7049

---

Investigated by repo-health-investigate • Finding: I4-7b5d3a9e

Generated by Repo Health — Investigate Finding

[github-actions]

📅 Daily Update — 2026-06-02

Status: 🟢 HEALTHY → 🟢 HEALTHY (no change)

Changes Since Last Run

• New findings: 0
• Resolved findings: 0
• Updated findings: 0

Current State

• 🔴 Critical: 0
• 🟡 Warning: 1 (216 untriaged issues - stable)
• ℹ️ Info: 0

Activity (24h)

• Issues: 0 opened, 0 closed
• PRs: 0 merged
• CI workflows: All passing ✅

Analysis

Repository remains in healthy state with no significant changes. The 216 untriaged issues remain stable from previous scans. Low 24-hour activity is typical for weekend/off-peak periods. All CI workflows are operational.

Next steps: Continue monitoring. Consider triage sprint to address the 216 untriaged issues during next team sync.

---

Run: #26805172988

Generated by Repo Health Check — Orchestrator