🏥 Repo Health Dashboard — dotnet/machinelearning

Last updated: 2026-07-03 07:01 UTC · Run: #28644126447

Status	Count
🔴 Critical	0
🟡 Warning	7
i️ Info	2
✅ Resolved (recent)	1
📋 Baselined	3

Overall: 🟡 Warning — stable from yesterday

📋 Executive Summary

The repo is holding at a Warning state for Day 23. No new critical findings. One new PR (#7652) was opened yesterday to isolate agentic workflow authentication using a Copilot token pool. All seven persistent warnings from prior runs are unchanged — the untriaged issue backlog, blocking-clean-ci cluster, stale community PRs, and agentic noise remain the dominant concerns.

AzDO pipeline monitoring (vsts-ci, codecoverage-ci, night-build, outer-loop-build) is disabled — AZDO_PAT not configured.

🔴 Critical Findings

None.

🟡 Warning Findings

ID	Category	Finding	Status	Age
I3-untriaged	Issue	50+ untriaged open issues; agentic bot PRs adding noise	📌 EXISTING	Day 23
I2-no-milestone	Issue	821+ open issues without milestone — planning gap	📌 EXISTING	Day 23
C5-blocking-ci-6978	Pipeline	#6978 TorchSharp QA OOM (`blocking-clean-ci`) — updated again today	📌 EXISTING	Day 18
C6-blocking-ci-cluster	Pipeline	~15 open `blocking-clean-ci` issues; count stable	📌 EXISTING	Day 15
P2-community-pr-7609	PR	#7609 arm64 SymSGD community PR awaiting review — 59 days old	📌 EXISTING	Day 9
INFO-agentic-noise	Issue	`ci-scan-feedback` bot loop: fallback issues accumulating in tracker	📌 EXISTING	Day 6
PR-7625-stale	PR	#7625 SentencePieceTokenizer Copilot PR awaiting human review — 23d since last update	📌 EXISTING	Day 4

i️ Info Findings

ID	Category	Finding	Status	Age
PR-7652-new	PR	#7652 New: Isolate agentic auth with Copilot token pool (by `@kotlarmilos`)	🆕 NEW	Day 0
PR-7626-review	PR	#7626 PredictionEnginePool hot-swap fix awaiting review — 19d old, recently active	📌 EXISTING	Day 4

✅ Recently Resolved

ID	Category	Finding	Resolved
P5-stale-prs	PR	Stale bot PRs #7607 and #7570 actively maintained	2026-06-20

📋 Baselined Findings

ID	Finding
I1-P0-5805	#5805 P0: MKLImports PDB missing — known tech debt
P5-stale-long-prs	#7416, #7406 long-lived PRs (TorchSharp, 4-bit quantization)

📈 7-Day Trends

Metric	Value	Trend
Open Issues (total)	821+	→ Stable
Untriaged Issues	50+	↗ Slow growth
Open PRs	~17	↗ +1 new (#7652)
blocking-clean-ci issues	~15	→ Stable
backport.yml (last run)	✅ success	→ Stable
locker.yml (run #945)	✅ success	→ Stable
AzDO pipelines	⚠️ Not monitored	N/A

🔗 Links

Note

🔒 Integrity filter blocked 65 items

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

#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 49 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 · ● 935.4K · ◷

🔍 Investigation: `I3-untriaged`

Finding: 221+ open issues with untriaged label — triage process gap persisting for 7+ days
Severity: Medium
Category: Issue

Evidence

221+ open issues carry untriaged — stable at this level for 7+ days per dashboard trend (count unchanged between 2026-06-10 and 2026-06-17). — Source: dashboard 🏥 Repo Health Dashboard — dotnet/machinelearning #7634
untriaged label policy is milestone-driven — .github/policies/untriaged.yml adds the label on issue open/reopen when no milestone is assigned; removes it only when the issue is closed or assigned to a milestone. There is no time-based or activity-based removal. — Source: .github/policies/untriaged.yml
I3-untriaged and I2-no-milestone are the same problem — the 221+ untriaged issues are a direct subset of the 821+ issues without a milestone. Adding any issue to a milestone automatically removes untriaged. No milestone assignment = untriaged forever. — Source: policy logic + dashboard
Issue triage agent adds but never removes untriaged — .github/workflows/issue-triage.agent.md (triggered on issues: [opened]) classifies new issues and always adds untriaged alongside other labels. The prompt explicitly says "Always add untriaged alongside any other label." Removal depends entirely on a human assigning a milestone. — Source: .github/workflows/issue-triage.agent.md
Oldest visible untriaged issue: ~3 years old — Object detection: Enable loading from checkpoint #6670 created 2023-05-10 (enhancement + needs-further-triage), never assigned a milestone. — Source: list_issues query (ASC by created_at)
Known Build Error (Helix CI bot) issues inflate count — Issues like Failure to load lib_lightgbm in Microsoft.ML.Fairlearn.Tests.GridSearchTest.TestGridSearchTrialRunner2 #6841, ArgumentOutOfRangeException when running Microsoft.ML.Fairlearn.Tests.GridSearchTest.TestGridSearchTrialRunner2 #6842, Microsoft.ML.Tests.TrainerEstimators.TrainerEstimators.TestEstimatorMulticlassNaiveBayesTrainer failed with assert #7060, AutoML tests writing test output outside of a test which causes test process to fail with no failing test cases #7067 are auto-generated by Helix CI (ericstj bot), carry untriaged + blocking-clean-ci + Known Build Error, and are tracked in a separate AzDO system. They were bulk-touched today at 07:00 UTC but still carry untriaged because no milestone is assigned. — Source: list_issues (updated_at DESC)
Automated agentic workflow failure reports also get untriaged — [aw] CI Failure Scanner - Feedback (machinelearning) failed #7633 (CI Failure Scanner failure, created today, auto-expires 2026-06-24) carries untriaged. These transient operational issues go through a different lifecycle but are counted in the backlog. — Source: list_issues
No closed/resolved untriaged issues in 2026 visible — A search for label:untriaged is:closed closed:>2026-01-01 returned only agentic workflow failure reports (operational, not community issues). Zero community bug/feature untriaged issues have been triaged closed this year. — Source: search_issues
Active feature work is happening on untriaged issues — [API] Public way to construct SentencePieceTokenizer (Unigram) from tokenizer.json / in-memory pieces+scores #7624 ([API] Public way to construct SentencePieceTokenizer, created 2026-06-10) has an assignee (Copilot) and 6 comments, yet still carries untriaged simply because no milestone was assigned. This shows the label is losing signal value. — Source: search_issues

Root Cause

Confidence: High

The untriaged label is mechanically equivalent to "no milestone assigned." The policy is working as designed — but no systematic milestone assignment process exists, so the label accumulates indefinitely. Three compounding factors worsen the backlog: (1) the oldest community issues have been awaiting triage for 2–3 years, (2) Helix CI and agentic workflow bots auto-create issues that immediately receive untriaged but are managed through separate systems and never get milestones, and (3) the issue-triage.agent.md intentionally preserves untriaged because removing it requires a human milestone decision that never happens. This is a process gap, not a tooling failure.

Recommendations

Priority	Action	Who
🔴 Immediate	Bulk-assign `Known Build Error + blocking-clean-ci` issues to a "Backlog" or "Tracking" milestone (or close as "won't fix" if stale). These are CI-bot-generated and managed in AzDO — they should not count as untriaged community issues. Approx 15–30 issues.	Team
🔴 Immediate	Exclude agentic workflow failure issues (`[aw]` title prefix or `agentic-workflows` label) from the `untriaged.yml` policy — they have `expires` markers and don't need standard triage.	Team
🟡 Short-term	Create a "Backlog" milestone and assign all enhancement/bug issues that have been reviewed (e.g., by the triage agent) but not yet scheduled. This will deflate the `untriaged` count while preserving planning flexibility.	Team
🟡 Short-term	Run a triage sprint targeting issues >90 days old with `untriaged` — close stale/duplicate ones, assign the rest to a milestone. Focus on pre-2024 issues first (~50+ visible, more filtered by integrity).	Team
i️ Long-term	Update `issue-triage.agent.md` to remove `untriaged` when it determines an issue is a `Known Build Error` (has Helix CI structure) or is an agentic workflow failure.	Team
i️ Long-term	Add a stale bot rule: issues with `untriaged` older than 180 days without activity should be auto-closed or auto-assigned to "Backlog" milestone — this prevents indefinite accumulation.	Team

Dashboard: 🏥 Repo Health Dashboard — dotnet/machinelearning #7634 (I2-no-milestone is the same underlying gap — 821+ open issues without milestone)
Triage policy: .github/policies/untriaged.yml
Triage agent: .github/workflows/issue-triage.agent.md
Sample old untriaged issues: #6670 (2023-05-10), #6719 (2023-06-01), #6735 (2023-06-14)
CI bot untriaged issues: #6841, #6842, #7060, #7067
Active work on untriaged issue: #7624 (has assignee but still untriaged)

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

Note

🔒 Integrity filter blocked 92 items

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

ResNet101 [ShapeInferenceError] Attribute pads has incorrect size #6580 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[Object Detection] Bounding box rotated due to image exif data #6592 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Improvements when specifying a ModelLoader when adding a predictionenginepool to DI. #6618 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does ML.NET support KNN? (part 2) #6639 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: Distributes learning / potential "hack" using checkpoints CSV on a network drive? #6648 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: How to get trainer used for best model #6651 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: Trying to serializing params from TrialResult fails #6652 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: Get columnInformation from binary IDV file #6653 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Need to find multiple columns for the same model for prediction #6680 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: setting temporary folder model #6687 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Visualization and analysis of LightGBM and FastTree #6688 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Support Multiple Threads on ML Related DB Access #6692 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Parallelism: trouble saturating CPU when making predictions #6702 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML 2.0: Should Seed be a tunable param? #6704 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Clarifications Regarding ForeCastBySsa #6720 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AutoML Forecasting Experiment Programaticaly #6725 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 76 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 — Investigate Finding · ● 1.1M · ◷

🔄 Health Delta — 2026-06-22 07:49 UTC

	Count
🆕 New findings	0
✅ Resolved findings	0
📌 Existing findings	4
📋 Baselined	1
🔴 Critical	0
🟡 Warning	4

Changes since 2026-06-21:

#6978 (TorchSharp QA OOM blocking-clean-ci) updated by bot at 07:00 UTC today — ongoing, Day 7
All other findings carry forward unchanged; day counters incremented

⚠️ Partial run — GitHub MCP WASM server trap mid-collection. Untriaged count, PR details, and GitHub Actions workflow data unavailable this run. Carried from 2026-06-21.

Severity: 🟡 Warning (unchanged since 2026-06-11) | Dispatches: 0 / 5 budget used

i️ Baseline file .github/health-baseline.md last reviewed 2026-03-06 — over 90 days ago, consider re-reviewing.

Note

🔒 Integrity filter blocked 58 items

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

Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Troubles with CustomMappingEstimator after save/load because InputSchemaDefinition is not being saved. #3988 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does GetFeatureWeights support categorical splits? #3766 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
How to inspect OneVersusAll models #3701 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[AutoML] Allow to use serialized IDataView as an input #3684 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OvaModelParameters is not strongly-typed #2467 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Conditions in KeyToVectorMappingEstimator.GetOutputSchema seem wrong #2185 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 42 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 · ● 1.2M · ◷

🔄 Health Check Delta — 2026-06-23

Run vs. previous: 2026-06-22 (#27937550540) → 2026-06-23 (#28008648004)

Category	Count
🆕 New findings	0
✅ Resolved	0
📌 Existing warnings	4
📋 Baselined	1
🔎 Dispatched investigators	0

Key observations:

All 7 blocking-clean-ci issues still open and unchanged — 5 were auto-touched by CI bot at 07:02 UTC today: #6978, #7393, #7417, #7418, #7419
PR #7609 (arm64 community contribution) is now 49 days old — was updated 2026-06-22, but still awaiting merge
untriaged label query timed out again (GitHub MCP WASM trap); count estimated ≥100 (consistent)
AzDO monitoring skipped — no AZDO_PAT configured
No critical escalations; no new findings require investigator dispatch

Overall status: 🟡 Warning — stable, no changes from prior run

Note

🔒 Integrity filter blocked 65 items

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

#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
AccessViolationException when running Microsoft.ML.Scenarios.ScenariosTests.TestRegressionScenario #6960 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 49 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 · ● 1.5M · ◷

🔁 Health Check Delta — 2026-06-24T07:04Z

Findings summary:

🆕 New: 1 (P5-stale-long-prs)
📌 Existing: 4 (C5, C6, I3, I2 — unchanged)
✅ Resolved: 0 (P5-stale-prs still closed since 2026-06-20)
📋 Baselined: 1

Key changes since 2026-06-23:

Issue #6978 (blocking-clean-ci — TorchSharp QA OOM) received CI bot activity at 07:00 UTC today — still open and blocking CI (Day 9)
🆕 New warning P5-stale-long-prs: PRs #7416 (TorchSharp 0.105.0 update, assigned @LittleLittleCloud, 450 days stale) and #7406 (GenAI 4-bit quantization, assigned @LittleLittleCloud, 470 days stale) — both targeting main, no activity since March 2025
7 blocking-clean-ci issues remain open (Day 7 for cluster finding)
2 investigators dispatched: P5-stale-long-prs (new) + C5-blocking-ci-6978 (active today)

Severity: ⚠️ Warning (no change from previous run)

Run #28081292211 · AzDO monitoring disabled (no AZDO_PAT) · ⚠️ Partial data: GitHub API WASM timeout on some queries

Note

🔒 Integrity filter blocked 64 items

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

Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Troubles with CustomMappingEstimator after save/load because InputSchemaDefinition is not being saved. #3988 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does GetFeatureWeights support categorical splits? #3766 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
How to inspect OneVersusAll models #3701 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[AutoML] Allow to use serialized IDataView as an input #3684 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OvaModelParameters is not strongly-typed #2467 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Conditions in KeyToVectorMappingEstimator.GetOutputSchema seem wrong #2185 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 48 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 · ● 846.8K · ◷

🔍 Investigation: `P5-stale-long-prs`

Finding: PRs #7416 and #7406 on main have been stale for 450–470 days with no reviewer activity — evaluate closure or reassignment
Severity: Medium
Category: PR

Evidence

PR update torchsharp to 0.105.0 #7416 ("update torchsharp to 0.105.0") — opened 2025-03-12, last updated 2025-03-31 (~450 days ago). Author: @LittleLittleCloud. State: open, mergeable_state: dirty (merge conflict with main). No CI check runs recorded.
PR [GenAI] Use BitsAndBytes for 4bit quantization. #7406 ("[GenAI] Use BitsAndBytes for 4bit quantization") — opened 2025-03-02, last updated 2025-03-11 (~470 days ago). Author: @LittleLittleCloud. State: open, mergeable_state: dirty. No CI check runs recorded.
Both PRs have the same author and assignee (@LittleLittleCloud), with no author activity after March 2025.
update torchsharp to 0.105.0 #7416 technical blocker — On 2025-03-31, @michaelgsharp ran CI and identified that TorchSharp ≥0.105.0 dropped OSX-x64 support. Additional work was needed to gate TorchSharp builds/tests away from OSX-x64. No follow-up from the author occurred after this was raised. Additionally, the current repo TorchSharpVersion is 0.102.7 — the proposed 0.105.0 is now 15+ months old and almost certainly obsolete.
[GenAI] Use BitsAndBytes for 4bit quantization. #7406 policy blocker — @michaelgsharp raised concerns on 2025-03-07/10 about introducing a new runtime dependency (LittleLittleCloud.TorchSharp.BitsAndBytes) owned by the PR author personally (not Microsoft/dotnet). The author confirmed personal ownership (MIT license). Reviewer also asked whether this could go into TorchSharp upstream and whether the version should be centrally pinned. Two review threads remain unresolved.
Neither PR has a linked issue in update torchsharp to 0.105.0 #7416. [GenAI] Use BitsAndBytes for 4bit quantization. #7406 has linked issues, but those issues have not been updated.
No human review approvals on either PR — only automated Copilot-bot review comments.

Root Cause

Confidence: High

Both PRs stalled due to concrete blockers raised by reviewers that the author did not address. For #7416, a last-minute CI run revealed a platform incompatibility (OSX-x64 dropped in TorchSharp 0.105.0) requiring non-trivial extra work. For #7406, the reviewer raised a valid governance concern about adding a personally-maintained CUDA library as a repo dependency. In both cases the author went silent and the PRs aged into merge-conflict state with no CI coverage.

Recommendations

Priority	Action	Who
🔴 Immediate	Close #7416: The proposed TorchSharp 0.105.0 is now ~15 months stale. If a TorchSharp upgrade is still desired, open a fresh PR targeting the current latest version and include the OSX-x64 platform-exclusion work.	Maintainer (`@michaelgsharp`)
🔴 Immediate	Close #7406: The `LittleLittleCloud.TorchSharp.BitsAndBytes` dependency governance question was never resolved, and the PR has diverged from main. Close with a comment inviting the author to reopen after (a) resolving the dependency ownership concern or (b) proposing an alternative approach that avoids a personally-owned CUDA package.	Maintainer (`@michaelgsharp`)
🟡 Short-term	Track TorchSharp upgrade separately: Create a tracking issue for upgrading TorchSharp to a current supported version, including the OSX-x64 platform exclusion work. Assign to a team member rather than relying on external contributions for core dependency bumps.	Team
i️ Long-term	Add stale-PR automation: Add a bot (e.g., `actions/stale`) to warn and auto-close PRs with no activity for 90–120 days. This prevents accumulation of conflicted, CI-dark PRs.	Team

PR update torchsharp to 0.105.0 #7416: update torchsharp to 0.105.0 #7416
PR [GenAI] Use BitsAndBytes for 4bit quantization. #7406: [GenAI] Use BitsAndBytes for 4bit quantization. #7406
TorchSharp OSX-x64 drop comment: update torchsharp to 0.105.0 #7416 (comment)
Dependency ownership concern: [GenAI] Use BitsAndBytes for 4bit quantization. #7406 (comment)

Investigated by repo-health-investigate • Finding: P5-stale-long-prs

Note

🔒 Integrity filter blocked 1 item

The following item was blocked because it doesn't meet the GitHub integrity level.

#7295 search_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".

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

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

Generated by Repo Health — Investigate Finding · ● 418.6K · ◷

📊 Daily Delta — 2026-06-26

Run: #28223027704 · Compared to: 2026-06-25

Metric	Change
🆕 New findings	0
✅ Resolved findings	0
📌 Existing warnings	4 (unchanged)
🔴 Critical	0
🟡 Warning	4
Dispatches	0 (no new/escalated findings)

No changes since yesterday. All four warning findings remain at existing status:

C6 (Day 8): blocking-clean-ci cluster holds at 15 open issues — no change.
C5 (Day 11): #6978 TorchSharp QA OOM — automated activity at 07:00 UTC today, no triage progress.
I3 (Day 16): Untriaged issue backlog — 28+ sampled (total likely 100+).
I2 (Day 16): 821+ open issues without milestone.

GitHub Actions CI: backport.yml ✅ · locker.yml ✅ (all recent runs successful)

AzDO pipeline monitoring: disabled — AZDO_PAT not configured.

No investigator dispatches: all findings are EXISTING with no escalation this run.

Note

🔒 Integrity filter blocked 118 items

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

#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 102 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 · ● 922.6K · ◷

🔄 Health Check Delta — 2026-06-27

Metric	Count
🆕 New findings	0
✅ Resolved findings	0
📌 Existing findings	4 (+1 day each)
🕵️ Dispatches	0

Changes since 2026-06-26

No severity changes — still 🟡 Warning
All warning findings aged by 1 day: C6 Day 8→9, C5 Day 11→12, I3 Day 16→17, I2 Day 16→17
New automated issue #7645 filed by ci-scan-feedback bot at 04:16Z — agentic-workflows noise, no human action needed
Issues #7418 and #7419 received automated known-issues bot updates at 06:02 UTC (routine heartbeat)
Issue #6978 (TorchSharp QA OOM) updated at 06:00 UTC — still open, no resolution
No new human-filed issues or PRs since last run
GitHub Actions CI: 100% ✅ · AzDO: monitoring disabled (no AZDO_PAT)

Workflow run #28281845444

Note

🔒 Integrity filter blocked 126 items

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

#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 110 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 · ● 2.8M · ◷

📊 Daily Delta — 2026-06-28T07:15 UTC

	Count
🆕 New findings	0
✅ Resolved	0
📌 Persisting warnings	4
i️ Persisting info	2
📋 Baselined	3

No change from yesterday's scan. Repo remains at 🟡 Warning.

Notable updates

#6978 (blocking-clean-ci TorchSharp QA OOM) received another activity update at 07:00 UTC today — Day 13 tracked, issue remains open and unresolved
blocking-clean-ci cluster holds at exactly 15 issues for Day 10 — no new closures overnight
INFO-issue-7645 (bot ci-scan-feedback noise) transitions: 🆕 NEW → 📌 EXISTING (Day 1)

No dispatches triggered

No new 🔴 Critical or 🆕 NEW 🟡 Warning findings — investigator dispatch budget unused (0/5).

Workflow Run #28314810056

Note

🔒 Integrity filter blocked 142 items

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

Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Troubles with CustomMappingEstimator after save/load because InputSchemaDefinition is not being saved. #3988 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does GetFeatureWeights support categorical splits? #3766 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
How to inspect OneVersusAll models #3701 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[AutoML] Allow to use serialized IDataView as an input #3684 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OvaModelParameters is not strongly-typed #2467 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Conditions in KeyToVectorMappingEstimator.GetOutputSchema seem wrong #2185 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 126 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 · ● 1.1M · ◷

📊 Health Check Delta — 2026-06-29 07:31 UTC

Run: #28355875362 | Status: 🟡 Warning (stable)

Changes since 2026-06-28

	Count
🆕 New findings	2
✅ Resolved	0
📌 Existing (carried forward)	7
📋 Baselined	unchanged

🆕 New Findings

🟡 Warning — PR-7625-stale: PR #7625 — Copilot-generated SentencePieceTokenizer API PR has been open 19 days without review or merge. Copilot PRs require human review; this is blocking a tokenizer API improvement.
i️ Info — PR-7626-review: PR #7626 — Community contribution fixing memory leak + stale-model issue in PredictionEnginePool hot-swap, 15 days old, assigned to kotlarmilos.

📌 Notable Existing Findings

#6978 — TorchSharp QA OOM (blocking-clean-ci) updated again at 07:00 UTC today (Day 14 of tracking)
blocking-clean-ci cluster — Stable at 15 issues (Day 11); no new issues added or resolved
ci-scan-feedback noise — Issue #7647 filed today, continuing the daily pattern from #7645–#7646

Key Metrics

Metric	Yesterday	Today	Δ
🟡 Warning findings	4	5	+1
i️ Info findings	2	3	+1
`blocking-clean-ci` count	15	15	0
Open PRs (non-baselined, active)	~4	~5	+1
Untriaged issues	50+	50+	~+1
AzDO pipelines	⚠️ unmonitored	⚠️ unmonitored	—

🔍 Dispatched Investigations (1 of 5)

PR-7625-stale (🟡 Warning, NEW) — Copilot PR awaiting review 19+ days

Note

🔒 Integrity filter blocked 85 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 69 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 · ● 851.5K · ◷

🔍 Investigation: `PR-7625-stale`

Finding: Copilot PR #7625 (SentencePieceTokenizer APIs) awaiting review 19 days without merge
Severity: High
Category: PR

Evidence

PR created 2026-06-10, last commit push 2026-06-18 — 11 days without any new activity as of 2026-06-29 — PR timeline
mergeable_state: "blocked" — PR cannot be merged in its current state — GitHub API response
CI failing — MachineLearning-CI aggregate: ❌ FAILURE; specifically MachineLearning-CI (MacOS_cross_arm64 Release_Build) ❌ FAILURE; Windows cross-arm64 builds ❌ CANCELLED — check runs on commit e1ff3c5c
No human approval — All 25+ review events are state: "COMMENTED"; no APPROVED or CHANGES_REQUESTED from any maintainer — reviews API
Active review from ericstj — Identified two correctness bugs in initial review (2026-06-11): (a) BOS/EOS positional fallback corrupts real vocab pieces when <s>/</s> are absent, (b) ExtractPrecompiledCharsMap silently drops sibling normalizer steps instead of throwing — review comment #4472717485
"@copilot please address feedback" — ericstj re-triggered Copilot iteration on 2026-06-18 (comment #4676096355); Copilot pushed multiple commits that day
Multiple iteration rounds — 15 commits, several rounds of Copilot push → ericstj inline comments → repeat; some threads now resolved/outdated, but CI still red after last push

Root Cause

Confidence: High

The PR is dually blocked: the CI is failing on MacOS cross-arm64 (the only non-cancelled platform build that failed, suggesting a test or platform-specific issue introduced by or not covered by the new tokenizer test fixtures), and no human reviewer has issued a formal approval. ericstj is actively engaged but has not approved — likely waiting for CI to go green and/or for the remaining review thread items to be fully addressed. The Copilot author cannot self-merge without human sign-off, and the last push (2026-06-18) left CI broken, stalling the loop.

Recommendations

Priority	Action	Who
🔴 Immediate	Investigate the MacOS_cross_arm64 CI failure — check Azure DevOps build `1471203` for the specific test failure log; determine if it is a flaky infra issue or a real test failure caused by PR changes	`ericstj` / any maintainer
🔴 Immediate	If CI failure is a real bug in the PR, re-invoke `@copilot` with specific instructions to fix the failing test/code	`ericstj`
🟡 Short-term	Once CI is green, `ericstj` (or another maintainer) should issue a formal review decision (APPROVED or CHANGES_REQUESTED) to unblock merge	`ericstj`
🟡 Short-term	Verify that the two correctness bugs called out in the initial review (BOS/EOS fallback corrupts pieces; silent normalizer drop) are fully addressed before approving — both affect real HF Unigram models	`ericstj`
i️ Long-term	Add CI test fixtures for HF Unigram tokenizers with non-conventional special-token positions and `Sequence` normalizers to prevent regressions	Tokenizer team

PR: Add public SentencePieceTokenizer factory methods for Unigram from vocab list and tokenizer.json #7625
Azure DevOps CI build: (dev.azure.com/redacted)
Failed check run: MachineLearning-CI (MacOS_cross_arm64 Release_Build) — run ID 82204245413
Reviewer: @ericstj (MEMBER, active)
PR author: Copilot (copilot-swe-agent)

Investigated by repo-health-investigate • Finding: PR-7625-stale

Generated by Repo Health — Investigate Finding · ● 444.9K · ◷

📊 Health Check Delta — 2026-06-30T07:11 UTC

vs previous run (2026-06-29)

Change	Count
🆕 New findings	0
✅ Resolved findings	0
📌 Existing findings promoted/changed	1
📌 Existing findings unchanged	6
📋 Baselined	unchanged

Key changes:

[ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 filed today (04:18 UTC): ci-scan-feedback bot created its 13th consecutive fallback issue — up from 3 noted in yesterday's run. The root cause and fix are clearly identified in the issue body: ci-scan.agent.md needs Hard Rule 10 to enforce early-exit on skip-only runs. Action needed: apply the patch from [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 or close these issues as noise once fixed.
INFO-agentic-noise severity promoted: info → warning (13 issues, escalating pattern).
PR Add public SentencePieceTokenizer factory methods for Unigram from vocab list and tokenizer.json #7625 (Copilot SentencePieceTokenizer): NEW → EXISTING (Day 1). Now 20 days old, assigned to @ericstj.
PR Fix memory leak and stale-model results when hot-swapping PredictionEnginePool #7626 (community PredictionEnginePool fix): NEW → EXISTING (Day 1). Now 16 days old.
blocking-clean-ci cluster: Unchanged at 15 issues (Day 12). All issues touched by health-check automation at 07:00–07:02 UTC.

Overall status: 🟡 Warning — no change from previous run. No new critical findings.

Run #28426815194 · AzDO monitoring disabled (no AZDO_PAT)

Note

🔒 Integrity filter blocked 53 items

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

Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Troubles with CustomMappingEstimator after save/load because InputSchemaDefinition is not being saved. #3988 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does GetFeatureWeights support categorical splits? #3766 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
How to inspect OneVersusAll models #3701 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[AutoML] Allow to use serialized IDataView as an input #3684 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OvaModelParameters is not strongly-typed #2467 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Conditions in KeyToVectorMappingEstimator.GetOutputSchema seem wrong #2185 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 37 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 · ● 1.1M · ◷

🔍 Investigation: `INFO-agentic-noise`

Finding: ci-scan-feedback bot filing 13 consecutive fallback issues (#7627–#7648) — failed PR-to-issue fallback loop
Severity: High
Category: Issue

Evidence

87+ consecutive ci-scan runs have all hit stale build window (>14d) since 2026-06-08 — the most recent failed AzDO build (#1453653) is now 22+ days old, so every run correctly skips. — Source: [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 body
Step 1 "and stop" is prose, not a Hard Rule — ci-scan.agent.md lines 80–92 have Hard Rules 1–9, but the early-exit instruction only appears inside Step 1 prose as "and stop". On some runs agents continue past it to Steps 2–7, fetching AzDO timelines, task logs, and Helix data that serve no purpose. — Source: ci-scan.agent.md SHA 0937c5f
Token waste confirmed: correct early-exit runs consume ~250K effective tokens; misbehaving runs consume 2.2–2.4M ET (10× overhead) for identical outcome. — Source: [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 bodies
Tally misreporting: skip-only runs emit | 0 | 0 | 0 | 0 | instead of | 0 | 0 | 0 | 1 | — the skipped count is not reflected in structured output because Step 7 tally is never reached. — Source: [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 body
13 identical fallback issues ([ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7630, [ci-scan-feedback] ci-scan: add Hard Rule 10 to force early exit on no scannable build #7636–[ci-scan-feedback] ci-scan: add Hard Rule 10 to force early exit on no scannable build #7637, [ci-scan-feedback] ci-scan: add Hard Rule 10 to force early exit on no scannable build #7639–[ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7646, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648), all filed between 2026-06-15 and 2026-06-30, all proposing the same one-line Hard Rule 10 addition. — Source: issue list, confirmed by reading [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7646, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648
Git push failure is the immediate trigger for the issue flood: each ci-scan-feedback run produces a patch, fails to push the branch, and falls back to creating a GitHub Issue. The fallback mechanism keeps firing because the underlying push failure is never fixed. — Source: every fallback issue's "This was originally intended as a pull request" footer
Fix is known, small, and identical across all 13 issues: add Hard Rule 10 after Rule 9 in .github/workflows/ci-scan.agent.md, elevating the early-exit to the same level as the issue-cap and label invariants. The full ready-to-apply patch (57 lines) is in [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648. — Source: [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648 body, patch preview

Root Cause

Confidence: High

There are two layered bugs, both with clear evidence:

Bug 1 (Content) — ci-scan.agent.md is missing Hard Rule 10. The early-exit on "no scannable build" lives only in Step 1 prose ("and stop"), which some model runs interpret permissively, causing them to proceed through Steps 2–7 and waste 10× tokens. The tally is also wrong on skip-only runs.

Bug 2 (Infrastructure) — ci-scan-feedback's git push is failing on every run. The bot correctly identifies the fix and generates a valid patch, but cannot push a branch to create a PR. The PR-to-issue fallback fires instead, creating a new duplicate issue on every daily run. The bot does not deduplicate against its own prior fallback issues, so the flood continues indefinitely.

Recommendations

Priority	Action	Who
🔴 Immediate	Apply the patch from #7648 manually: add Hard Rule 10 to `.github/workflows/ci-scan.agent.md` after Rule 9 — a 2-line change. This fixes the content bug and stops future ci-scan over-runs.	Maintainer (any write-access)
🔴 Immediate	Close duplicate issues #7627, #7630, #7636, #7637, #7639–#7646 as duplicates of #7648 to stop the noise accumulation.	Maintainer
🟡 Short-term	Investigate why ci-scan-feedback's `git push` is failing. Likely cause: branch protection requiring PRs, or the workflow token lacks `contents: write`. Check the workflow run logs for runs 27523314204 → 28419478972 for the push error.	Maintainer
🟡 Short-term	Add deduplication logic to ci-scan-feedback's fallback path: before filing a new issue, search for an existing open issue with the same title prefix to avoid repeated filings.	Bot prompt edit
i️ Long-term	Review ci-scan-feedback's permissions model so it can push branches and open PRs autonomously. Consider adding `contents: write` scoped to the agent branch namespace only.	Repo admin

Proposed Hard Rule 10 (ready to apply)

Add this line to .github/workflows/ci-scan.agent.md after Rule 9:

10. **Exit at Step 1 on no scannable build.** If Step 1 yields any skip reason (`stale build window (>14d)`, `no follow-up build yet, defer to next run`, or `no failed build in 7d`): append the reason to `/tmp/gh-aw/agent/coverage/MachineLearning-CI.txt`, print the Step 7 tally row as `| 0 | 0 | 0 | 1 |`, call `noop`, and stop immediately — do not fetch any AzDO timeline, download any task log, or query any Helix work item.

Also update the Step 1 trailing sentence to read: "If selection yields no scannable build, apply Hard Rule 10 immediately." Full patch with both changes is in #7648.

🔄 Daily Delta — 2026-07-01T07:22 UTC

Run: #28500685308 · compared to run 2026-06-30

	Count
🆕 New findings	0
✅ Resolved findings	0
📌 Existing findings	10 (all aging +1 day)
🔴 Critical	0
🟡 Warning	6
i️ Info	2
📋 Baselined	4

Notable changes:

🔔 Issue #6978 (blocking-clean-ci TorchSharp QA OOM) was updated at 07:00 UTC today — renewed activity on this known CI blocker
✅ Epic #6984 (Tokenizers improvement track) was closed 2026-06-29 — positive team progress
⏳ Copilot PR #7625 (SentencePieceTokenizer) now 21 days without review — assigned @ericstj
⏳ Community PR #7626 (PredictionEnginePool fix) now 17 days without review
⏳ Community PR #7609 (arm64 SymSGD) now 57 days without review

No new dispatches — all warning findings have existing investigation issues.

Note

🔒 Integrity filter blocked 160 items

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

Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Troubles with CustomMappingEstimator after save/load because InputSchemaDefinition is not being saved. #3988 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Does GetFeatureWeights support categorical splits? #3766 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
How to inspect OneVersusAll models #3701 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
[AutoML] Allow to use serialized IDataView as an input #3684 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OvaModelParameters is not strongly-typed #2467 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Conditions in KeyToVectorMappingEstimator.GetOutputSchema seem wrong #2185 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 144 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 · ● 626.9K · ◷

🏥 Repo Health Delta — 2026-07-02

Run: #28571700375 · vs. previous run 2026-07-01

Category	Count	Change
🆕 New findings	0	—
✅ Resolved	0	—
📌 Existing warnings	7	±0
i️ Info	1	−1 (PR-7609 ⬆ escalated to Warning)
📋 Baselined	3	±0

Key Changes

⬆ INFO-agentic-noise escalating (Day 5): ci-scan-feedback bot filed fallback issue #7651 today. Issue body cites 95 consecutive skipped runs and 14 prior failed push attempts since Jun 8 — bot is stuck in a loop creating ~1 noise issue/day. Investigator dispatched.
⬆ P2-community-pr-7609 → Warning: Community PR #7609 (arm64 SymSGD, fixes #5798) is now 58 days without a review — escalated from Info. Investigator dispatched.
➡ PR-7625-stale (Day 3): Copilot PR #7625 SentencePieceTokenizer APIs, 22 days old with no activity since Jun 18, still needs a human reviewer assigned.
➡ C6-blocking-ci-cluster (Day 14): ~15 blocking-clean-ci issues unchanged; #6978, #7418, #7419 all updated today by Arcade health checks.

Dispatched Investigators

Finding	Severity	Summary
INFO-agentic-noise	🟡 High	ci-scan-feedback bot loop — 95 consecutive failures, 16+ fallback issues
C6-blocking-ci-cluster	🟡 High	~15 blocking-CI issues open 14 days

AzDO monitoring: Disabled — configure AZDO_PAT secret to enable pipeline health data.
GitHub Actions CI: ✅ All healthy (backport + locker).

Note

🔒 Integrity filter blocked 45 items

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

#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add Quantile Regression support to LightGbmRegressionTrainer #7603 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
7.5 CVSS in Microsoft.Bcl.Memory 9.0.4 #7589 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TreeEnsembleRegressor is deprecated in ONNX #7581 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
erfInv is very inneficient #7568 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Updated roadmap? #7547 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Cross-Platform ML.NET Model Builder Extension for Visual Studio Code #7545 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
FeaturizeText: System.ArgumentOutOfRangeException: Schema mismatch for input column 'Text_CharExtractor': expected Expected known-size vector of Single, got Vector<Single> #7529 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
ML.NET LightGBM linear-tree #7527 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
text tokenizer EncodeToTokens repeat #7523 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
MulticlassClassificationExperiment preFeaturizer not acting on the schema #7522 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 29 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 · ● 869.3K · ◷

🔍 Investigation: `INFO-agentic-noise`

Finding: ci-scan-feedback bot loop — 95 consecutive skipped ci-scan runs producing fallback issues at ~1/day since Jun 8 (latest: #7651)
Severity: High
Category: Issue

Evidence

95 consecutive ci-scan skips (Jun 8 – Jul 2) — Every 6h run of ci-scan exits at Step 1 with stale build window (>14d): AzDO MachineLearning-CI definition 167 on main has had no builds within the 14-day scannable window since the workflow launched. Source: [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7651 body, [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627 body, dashboard 🏥 Repo Health Dashboard — dotnet/machinelearning #7634.
14+ fallback issues filed, zero PRs ever created — ci-scan-feedback correctly detects the skip-path token waste, proposes adding Hard Rule 10 to ci-scan.agent.md, attempts create_pull_request, but every push fails with "the git push operation failed" and falls back to filing an issue instead. Confirmed: 0 open [ci-scan-feedback] PRs exist. Source: issues [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7630, [ci-scan-feedback] ci-scan: add Hard Rule 10 to force early exit on no scannable build #7636–[ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7648, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7650, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7651; GitHub PR search.
All 14+ fallback issues propose the same 2-line patch — Hard Rule 10 adds an explicit early-exit invariant to ci-scan.agent.md (append skip reason, print | 0 | 0 | 0 | 1 |, emit noop, stop — without fetching AzDO timeline/logs/Helix). The patch is functionally identical in every fallback issue. Source: comparing [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7630, [ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7651 diffs.
ci-scan.agent.md on main still unpatched — SHA 0937c5f, Hard Rule 10 absent, file unchanged since initial merge. The fix proposed daily since Jun 15 has never landed. Source: github-get_file_contents.
PR [ci-scan-feedback] Fix push to existing feedback PR on scheduled runs #7620 (merged Jun 9) fixed the wrong code path — It fixed push-to-pull-request-branch (updating an existing PR) but left create-pull-request (creating the first PR) broken. Since no PR has ever been successfully created, every daily run sees "no existing PR" → tries create_pull_request → push fails → new fallback issue. Source: [ci-scan-feedback] Fix push to existing feedback PR on scheduled runs #7620 body.
Workflow runs succeed despite push failure — All 5 jobs in run 28564527554 (Jul 2) concluded success; the gh-aw runtime handles push failures gracefully by creating a fallback issue. The noise is a design artifact, not a workflow crash. Source: list_workflow_jobs.
Exact push error not surfaced in available logs — The gh-aw safe-outputs server emits a generic fallback message. The workflow declares permissions: read-all on the agent job; if the safeoutputs runtime relies on the GITHUB_TOKEN rather than the Copilot App token for branch creation, the missing contents: write scope would explain the consistent failure from day one of proposals.

Root Cause

Confidence: High

Two co-occurring issues form a feedback loop. (A) MachineLearning-CI (AzDO definition 167) has had no builds within the 14-day scan window since June 8, so every ci-scan run skips at Step 1; the existing "and stop" prose instruction is not a hard constraint and some model versions continue to fetch timelines/logs before concluding the skip (250K–5.8M effective-token variance per run). (B) ci-scan-feedback correctly diagnoses this and proposes Hard Rule 10 each day, but create-pull-request fails every time — most likely permissions: read-all on the agent job prevents branch push if the safeoutputs server uses the GITHUB_TOKEN, or the Copilot App's installation token lacks write access to this repo. Because no PR is ever created, every daily run re-proposes the same fix and creates a new fallback issue, flooding the tracker at ~1/day.

Recommendations

Priority	Action	Who
🔴 Immediate	Apply the patch from any fallback issue manually (all are identical): `gh run download <run-id> -n agent`, then `git am --3way`, push, PR. This stops the daily fallback flood once Hard Rule 10 lands.	Maintainer
🔴 Immediate	Close issues #7627, #7630, #7636–#7650 as duplicates of #7651 (most recent / most complete).	Maintainer
🟡 Short-term	Fix `create-pull-request` push failure in `ci-scan-feedback`. Add `contents: write` (and `workflows: write` if needed) to the agent job's `permissions` block and re-test via manual workflow dispatch. Compare with `dotnet/runtime`'s equivalent workflow if that one is creating PRs successfully.	Maintainer
🟡 Short-term	Investigate why AzDO `MachineLearning-CI` definition 167 has had no builds accessible in 24+ days — pipeline disabled, frequency drop, or scan window needs recalibration.	Maintainer
i️ Long-term	Add idempotency to `ci-scan-feedback` fallback: skip filing a new fallback issue if an open issue with the same title stem was filed within the last 7 days with no action taken.	Workflow owner

Fallback issues: [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7627, [ci-scan-feedback] ci-scan: add Hard Rule 10 for early exit on no scannable build #7630, [ci-scan-feedback] ci-scan: add Hard Rule 10 to force early exit on no scannable build #7636–[ci-scan-feedback] ci-scan: add Hard Rule 10 for explicit early exit on no scannable build #7651 (16 open, all duplicates)
Introducing PR: Add ci-scan and ci-scan-feedback agentic workflows #7614 (merged Jun 8)
Partial fix (wrong code path): [ci-scan-feedback] Fix push to existing feedback PR on scheduled runs #7620 (merged Jun 9)
Unpatched file: .github/workflows/ci-scan.agent.md (SHA 0937c5f)

Investigated by repo-health-investigate • Finding: INFO-agentic-noise

Generated by Repo Health — Investigate Finding · ● 1.3M · ◷

🔍 Investigation: `P2-community-pr-7609`

Finding: Community PR #7609 — Enable SymSgdNative on arm64 via system BLAS (fixes #5798)
Severity: Medium
Category: PR

Evidence

PR open 58 days with no maintainer review — Created 2026-05-05 by @anicka-net (community contributor); zero review comments; no assigned reviewers, no milestone — PR #7609
CI: 8 of 25 checks failing — All failing checks are cross-compilation targets: Ubuntu_x64_cross_arm Debug/Release, Ubuntu_x64_cross_arm64 Debug/Release, Windows_cross_arm64 Debug/Release, MacOS_cross_arm64 Debug/Release. All native x64 builds pass.
Failing builds finish in ~6 minutes (config-time failure, not compile/test) — consistent with CMake find_package(BLAS REQUIRED) halting the build immediately when BLAS cannot be found in the cross-compilation sysroot.
New MklImportsArm/CMakeLists.txt uses find_package(BLAS REQUIRED) — In CI cross-compilation environments (x64 host compiling for ARM target), ARM BLAS libraries (OpenBLAS) are not installed in the x86 sysroot and CMake's find_package does not search the target sysroot unless explicitly configured.
Author tested on native ARM64 only — PR test plan notes "Tested on NVIDIA DGX Spark (Grace ARM CPU, Ubuntu 24.04, .NET 10.0.107)". Native ARM64 machines have OpenBLAS available directly; cross-compilation environments do not.
Code coverage and CLA pass — Codecov reports no regression; CLA is signed. x64 path is unaffected and all x64 builds green.
No review threads, no blockers noted by maintainers — The PR has stalled purely from lack of triage.

Root Cause

Confidence: High

The PR correctly fixes arm64 native support by replacing Intel MKL (x86-only) with an OpenBLAS-backed shim and sparse BLAS implementations in C. However, the new MklImportsArm/CMakeLists.txt calls find_package(BLAS REQUIRED), which fails in all CI cross-compilation jobs (x64 host → ARM target) because the cross-compilation sysroot does not have ARM-architecture BLAS libraries installed, and CMake does not automatically search the target sysroot. The PR was validated only on a native ARM64 machine where OpenBLAS is available natively. The CI has no native ARM64 runners, so the cross-compilation failure was not caught pre-submission.

Recommendations

Priority	Action	Who
🔴 Immediate	Assign a maintainer to triage and review PR #7609 — the logic and approach are sound	ML.NET team
🟡 Short-term	Fix the cross-compilation failure: change `find_package(BLAS REQUIRED)` → `find_package(BLAS)` with a graceful skip/stub fallback or install `libopenblas-dev:arm64` in the CI cross-compilation Docker image	`@anicka-net` / ML.NET team
🟡 Short-term	Add comment on PR explaining the cross-compilation issue so the author can fix it and doesn't wait indefinitely	ML.NET team
i️ Long-term	Add a native ARM64 CI job (e.g., an `Ubuntu_arm64` runner) to validate arm64-specific changes without relying on cross-compilation	ML.NET team

PR: #7609 — Enable SymSgdNative on arm64 via system BLAS
Fixes: Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 (SymSGD disabled on arm64 due to MKL dependency)
Related: Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588, Unable to load shared library 'MklImports' on MacOS with CLI #7229
Failing CI jobs: Ubuntu_x64_cross_arm*, Windows_cross_arm64, MacOS_cross_arm64

Investigated by repo-health-investigate • Finding: P2-community-pr-7609

Note

🔒 Integrity filter blocked 2 items

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

#5798 issue_read: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6588 issue_read: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".

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

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

Generated by Repo Health — Investigate Finding · ● 321.4K · ◷

📊 Daily Health Delta — 2026-07-03

Run: #28644126447 · Overall: 🟡 Warning (unchanged)

Changes since 2026-07-02

	Count
🆕 New findings	1
📌 Existing (carried)	10
✅ Resolved	0 (P5-stale-prs remains resolved from 2026-06-20)
🔄 Severity changes	0

🆕 New Finding

PR-7652-new (i️ Info) — #7652 opened by @kotlarmilos: Isolate agentic workflow authentication with a Copilot token pool. New PR targeting the copilot-pat-pool environment setup.

Key Metric Changes

Metric	Yesterday	Today	Δ
Open PRs	~16	~17	+1
blocking-clean-ci issues	~15	~15	0
Warnings	7	7	0
CI (locker) runs	#944 success	#945 success	✅

Notable Activity

Microsoft.ML.TorchSharp.Tests.QATests.TestSimpleQA followed by process killed / return 137 #6978 (TorchSharp QA OOM) updated again at 07:00 UTC today — automated CI health scan touching it daily
Enable SymSgdNative on arm64 via system BLAS #7609 (arm64 SymSGD community PR, 59 days) still awaiting review — community contribution at risk of going stale
Add public SentencePieceTokenizer factory methods for Unigram from vocab list and tokenizer.json #7625 (SentencePiece Copilot PR) last reviewed 2026-06-18; no activity for 15 days; needs human reviewer assignment

No investigator dispatches — no new critical findings or severity escalations.

Note

🔒 Integrity filter blocked 65 items

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

#7610 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7606 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7605 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7569 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#7094 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
#6449 list_pull_requests: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Error: "Microsoft.ML currently supports 'x64' and 'x86' processor" when target runtime is *linux-arm* or *linux-arm64* #6588 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Exposing the tree for multiclass classification #6370 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
CreateEnumerable code fails in Visual Studio FSI but works in .NET Interactive notebook? #6353 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Add substitutes for 4 IntelMKL methods so SymSgd can be added to non x86/x64 builds #5798 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Memory leak #5744 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
Migrate to VSTest for all Unit Tests #5587 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
OMP: Error #15: Initializing libiomp5.dylib in Image Classification with AutoML #5569 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
SVMLightLoader Fails above 128 dense rows #5566 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
TensorFlow scoring sample should pass an empty dataview #4210 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
mlnet's generated projects don't include the cs files #4145 list_issues: has lower integrity than agent requires. The agent cannot read data with integrity below "approved".
... and 49 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 · ● 935.4K · ◷

🏥 Repo Health Dashboard — dotnet/machinelearning #7634

Description