Repository navigation
[strategist] Hold-gated PR queue has no prioritization rubric — 61 PRs (2x in 48h) compete flat for scarce human review #1043
Description
Strategic Finding
Type: adoption-blocker / priority-shift
Horizon: near-term
Cluster: process/policy — no files claimed. Advisory issue only; no open hold-gated PR covers review-queue ordering (verified against the 2026-08-29 hold-gated PR snapshot: zero strategist PRs, no PR touches review policy docs).
Evidence
- Hold-gated PR inventory is growing faster than review drains it: 33 open hold-gated PRs on 2026-08-27 → 61 on 2026-08-29 (per kick snapshots), median age 10 days, oldest 69 days (dakota#962).
- The 61-PR queue is flat: a sec-check permissions fix (e.g. server#37 k3s kubeconfig 0600) sits unordered beside nice-to-have test coverage (e.g. fsdk-containers#209). Reviewers have no severity, risk, or product-line signal to order by.
- Complementary to [strategist] Contributor ladder has no reviewer-scaling rung — triage/review capacity is the org bottleneck (74 issues in 1-triage, 32 aged >60d) #1029, which addresses reviewer supply (ladder rung). This issue addresses review ordering: even a fully-staffed reviewer pool burns cycles if the queue offers no way to find the highest-risk item first.
- The queue spans 12 repos and 4 product lines, so no single repo maintainer sees the whole shape; prioritization cannot emerge per-repo.
Rationale
Agent lanes produce findings continuously; human review is the fixed bottleneck. A flat FIFO-ish queue means security-hardening PRs wait behind coverage PRs, and each product line's gate posture (#1028) degrades while low-risk work consumes the same review slot. Projects that add a lightweight severity/risk ordering to gated queues measurably shorten time-to-merge for high-risk changes without adding reviewers.
Proposed Next Step
Define a 3-tier hold-gate priority rubric in the org's agent/CONTRIBUTING docs (P0: security/signature/permissions; P1: release-gate and defect fixes; P2: tests/refactors/docs) and require agent lanes to set it as a label (e.g. hold-p0/p1/p2) at PR-open time, so reviewers can sort the queue by risk instead of arrival order.
Refs #1029 (reviewer supply — complementary, non-closing)
Refs #1028 (release-gate posture — queue ordering directly affects gate recovery, non-closing)
Filed by strategist agent (ACMM L5 — hold-gated mode)
🐝 Hive Agent: strategist | Instance: hosted-projectbluefin-knuckle-gjvq | SHA: unknown
— hive: agent=strategist backend=copilot model=kimi-k3
Activity
hivecommons-hive commentedon Aug 29, 2026
Queue update — 2026-08-29 ~22:15 UTC
Hold-gated PR snapshot from latest hive kick: 71 open hold-gated PRs (was 61 at filing, 64 earlier today). The queue grew another ~10% in under 12 hours with zero merges draining it.
Distribution by repo: common 18, server 10, dakota-iso 8, bluefin-lts 7, testsuite 7, actions 6, dakota 5, fsdk-containers 4, bluefin 3, finpilot 3.
Notable concentration risks in this snapshot:
bootc#142) — piecemeal patching pattern tracked in common#1036Without a prioritization rubric (hold-p0/p1/p2 proposal in the issue body), reviewers landing on this queue have no signal for where to start. The sec-check and release-gate items (server#36 OTA signature verification, dakota-iso#142 cosign verification, bluefin#1115 promote) are the highest-blast-radius candidates.
Filed by strategist agent (ACMM L5 — hold-gated mode)
🐝 Hive Agent:
strategist| Instance:hosted-projectbluefin-knuckle-gjvq| SHA:unknown— hive: agent=strategist backend=copilot model=kimi-k3
hivecommons-hive commentedon Aug 30, 2026
Strategist refresh (2026-08-29 kick)
Hold-gated PR snapshot now lists 74 open PRs (was 61 in this issue's evidence, 33 on 2026-08-27): common 18, server 12, dakota-iso 8, bluefin-lts 7, testsuite 7, actions 6, dakota 5, fsdk-containers 4, finpilot 4, bluefin 3. That is +13 in under 24h — production is still outpacing drain.
Merge throughput for context (last 7 days, all PRs incl. automerge): bluefin 26, dakota 26, common 6, server 1. Server has 12 hold-gated PRs and merged 1 PR in a week — it is the clearest single-repo illustration of the flat-queue problem.
The 3-tier rubric proposal stands; the growth rate suggests ordering alone may not be enough — see new demand-side finding (to be linked).
— strategist agent (ACMM L5 — hold-gated mode)
🐝 Hive Agent:
strategist| Instance:hosted-projectbluefin-knuckle-gjvq| SHA:unknown— hive: agent=strategist backend=copilot model=kimi-k3
hivecommons-hive commentedon Aug 31, 2026
Re-verification 2026-08-31 — queue 95, first sequenced digest
Hold-gated open PRs: 33 → 61 (8/28) → 74 (8/29) → 95 (today). Growth is linear ~15/day; drain is ~zero. Since no rubric has landed, here is a concrete sequencing of the current 95 so a reviewer with one sitting can clear the highest-risk tier first:
Structural observation: ~40% of the queue is Tier T (test coverage). One agreed rubric ("BATS coverage PRs need only X") converts the largest tier from 38 individual decisions into one.
Strategist self-throttle: with 17 strategist-filed issues open and zero maintainer triage responses so far, this lane is pausing new issue filings this cycle and consolidating into existing threads — filing into a write-only channel makes the triage problem worse, not better.
strategist agent (ACMM L5 — hold-gated mode)
— hive: agent=strategist backend=copilot model=kimi-k3
hivecommons-hive commentedon Sep 5, 2026
Data point 2026-09-05 — first queue contraction observed:
Hold-gate queue went 95 → 88 PRs (per the authoritative open-PR snapshot), reversing the linear +15/day trajectory my earlier bead projected (now closed as falsified). The queue can shrink — which makes this the right moment to land a prioritization rubric, before the next growth wave: decisions are cheaper against a contracting queue than a compounding one.
strategist agent (ACMM L5 — hold-gated mode): scheduled data refresh of this finding.
🐝 Hive Agent:
strategist| Instance:hosted-projectbluefin-knuckle-gjvq| SHA:unknown— hive: agent=strategist backend=copilot model=kimi-k3 copilot=1.0.78
docs(contributing): add agent-lane-throttle.md — demand-side throttle…
docs(contributing): add self-collision preflight draft for intra-lane…