docs(release): APR-RELEASE-001 spec + PMAT-1098 receipt for session 1 — the 0.67.0 train by noahgift · Pull Request #3164 · paiml/aprender

noahgift · 2026-09-12T11:58:12Z

Draft until the train reaches its terminal step. Carries the operator's spec (untracked until now), the PMAT-1098 receipt-so-far (andon: K crossed at k=312), the R-17 routing row, and the two agy lane receipts under evidence/release/0.67.0/agy/. The §7 report and the docs/build-ledger train record are added when v0.67.0 is tagged or SKIPPED.

Refs #3078, PMAT-1098.

🤖 Generated with Claude Code

…c, and the R-17 model-movement row for PMAT-1098 The spec was dropped by the operator on 2026-09-12 and is untracked until this commit; the routing row records the orchestrator model moving opus->fable-5-1 at phase 4 (phase-boundary.sh). Pmat-Ticket: PMAT-1098

…(the 0.67.0 train) + the two agy lane receipts Andon obligation: K=240 was crossed at k_measured=312 when the operator issued the spec run; WIP committed and this draft PR carries the receipt-so-far. Amended at phase 5. Pmat-Ticket: PMAT-1098

github-actions · 2026-09-12T12:02:55Z

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=3164 head=246efe069680d511ce26cf91150ecee1452bdf13 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

…s the bottleneck, gx10 is 2.3x faster and half idle One append-only JSON per (sha, host, job) from the Actions REST API: queue_wait_s (created->started), exec_s, total_s, host, conclusion. peak_rss_mb and free_disk_gb are null with an explicit "[U]" in unmeasured[]: the REST API does not expose them, and an absent measurement must never read as a zero. What the 182 records say (p50 / p95 exec_s, this repo, 2026-09-12): workspace-test intel 1790 / 4148 s gx10 790 / 790 s yoga 259 / 3167 s guard-cargo intel 1201 / 1820 s gx10 422 / 516 s yoga 830 / 1046 s guard-tree intel 604 / 879 s gx10 192 / 213 s yoga 379 / 379 s queue wait p95 intel 2995 s gx10 233 s yoga 1654 s Per-run required-check wall clock, grouped by which boxes the run touched: gx10 only n=4 mean 9.4 min gx10 + yoga n=4 mean 11.3 min gx10 + intel n=10 mean 42.6 min gx10 + intel + yoga n=11 mean 31.6 min Every run that touched intel cost 32-43 min; no run that avoided it cost more than 11.3. intel took 15 of the 22 workspace-test records because that job is pinned to X64 -- and it is the job whose p95 is 69 minutes. #3139 unpins it. §1 coupling, no longer [U]: p95 gate = 72.3 min, so max PRs per train = 72h / 72.3 min = 59.7. §8's "stop cutting trains below 10" does not fire. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

noahgift · 2026-09-12T12:32:20Z

Build-server packing + build-time kaizen — 0.67.0 train, 12:35Z

Packing right now (org runners, all repos; the aprender queue is what they are draining)

box	busy / total	operator target	verdict
gx10	4/6	80%	UNDER target (66%)
yoga	3/5	50%	at or over target (60%)
intel	15/16	80%	at or over target (93%)

The finding: the packing target is being met on the wrong box. intel is at 94%
and it is the slowest box; gx10 is at 67% and is 2.3x faster. Filling intel is
not throughput, it is queueing.

182 ledger records written this train (docs/build-ledger/2026-09-12/, one JSON
per sha+host+job, from the Actions REST API). p50 / p95 exec_s:

job	intel	gx10	yoga
workspace-test	1790 / 4148 s	790 / 790 s	259 / 3167 s
guard-cargo	1201 / 1820 s	422 / 516 s	830 / 1046 s
guard-tree	604 / 879 s	192 / 213 s	379 / 379 s
queue wait (p95)	2995 s	233 s	1654 s

Per-run required-check wall clock, grouped by which boxes a run touched:

host mix	runs	mean wall clock
gx10 only	4	9.4 min
gx10 + yoga	4	11.3 min
gx10 + intel	10	42.6 min
gx10 + intel + yoga	11	31.6 min

Every run that touched intel cost 32-43 min. No run that avoided intel cost more
than 11.3 min. intel holds 15 of the 22 workspace-test records because that job
is pinned to X64 — and it is the job with a 69-minute p95.

§1 coupling, no longer [U]: p95 gate = 72.3 min ⟹ max PRs per train =
72 h / 72.3 min = 59.7. §8's "stop cutting trains below 10 PRs/train" does not fire.

Acted on this train

ci(fleet): workspace-test runs on any clean-room box — gx10 measured 3–4× faster than intel; four aarch64-only reds fixed; tree-reader step builds 20 packages, not 686 binaries (PMAT-3138) #3139 (workspace-test on any clean-room box) is in the merge queue. It is the
single change that moves the 69-minute job off the 94%-full box. It lands in 0.67.0.
Cancelled my own two 0.68.0 PRs' CI (dogfood(examples): G3.EX learns needs-data — 173 examples that need a model or tokenizer the repo does not ship were reading as defects #3163, docs(release): APR-RELEASE-001 spec + PMAT-1098 receipt for session 1 — the 0.67.0 train #3164) — six aprender workspace-tests
were running on intel at once against §3.4's one-PR-in-CI rule, and two of them
were mine and not on the train.
Dequeued perf(ci): pr-review-receipt is dispatch-only — 96 h of fleet runner-time for a job that gates nothing (PMAT-1078) #3046 (guard-tree RED on its group: check_roadmap_diff_additive.sh).
It was holding a merge-queue build slot it could not use. Moved to 0.68.0 and cut
from the 0.67.0 CHANGELOG.
Local sweep work (981-example dogfood) runs on lambda-vector, never on a CI
runner, so it takes nothing from the queue.

Still owed, and why it is not being done in this session: the P2 infra PR that
declares runner units for yoga and gx10 in forjar.yaml. §3.5 forbids the ad-hoc
SSH that would fix it faster, and §0's selector allows exactly one thing per
session — this session is the train.

… ledger answered §1 Three things the receipt did not yet carry: 1. orch_model. The harness moved this session from claude-fable-5-1 to claude-opus-5 in the middle of phase 4. A model change mid-session is a recorded event, never a silent one: route.sh record-event wrote the row to docs/audits/impl-routing.jsonl and the frontmatter now names the live model. fable_binding drops to false with the reason, rather than asserting a binding that no longer holds. 2. The build ledger, 182 records, and what they say about which box to fill. 3. §1's max-PRs-per-train, measured at 59.7 from 29 runs, so it stops being [U] and §8's stop condition can be evaluated instead of guessed. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

…real defects, 38 classifier gaps The gate #3121/#3122 added during this cycle had never been executed against a tag. Its first run says 23 fail + 21 timeout, and that number is wrong in the interesting direction: re-running all 44 with each target's own required-features and keeping the logs leaves SIX genuine defects. 9 rows pass once the feature they name at runtime is enabled (cuda, compression, embeddings, tensor, shell-autocomplete). They declare no required-features and gate themselves in code, so the sweep ran them wrong. 20 rows are servers, TUIs and unbounded benchmarks. Killing them at 120 s and calling it `timeout` -- a class the skill defines as a defect -- is wrong on all twenty. 6 rows are a missing model or a missing argument worded outside the regexes. 1 is a TTY, 1 is this box's glibc against a prebuilt ort artifact, 1 is a wall-clock perf assertion failing under load. Verdict GO. Every one of the six defects is byte-identical at v0.66.0: the expect() on weight.rs:356, the five PinnedBuffer imports, the three cwd-relative config literals in llama2/train.rs. §3.1 skips a train whose CUT cannot go green; this cut did not cause any of them, the previous tag shipped them all, and a new instrument finding a backlog is not the same event as a regression. They are #3178 #3179 #3180 #3181, and the classifier gaps are #3182. Not claimed: that the 20 timeouts are healthy. They were not individually verified and the gate cannot yet tell a serving server from a wedged one. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

…, #3068/#3139/#3175 dequeues, the T-0 correction table, G3.EX GO, and the model return (opus -> fable) recorded Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

… 34693750243..34700216664 incl. the discarded #3145 workspace-test (4958 s) and the two cancelled docs-PR runs [skip ci] Pushes to this docs branch were starting full CI runs on intel next to the train (34694025825, 34700216664, 34701097504 — all cancelled by hand); until v0.67.0 is tagged every push here carries [skip ci]. The final push before merge does not. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…e checkout paths — the shipped-tier machine-specific-path ratchet went RED on #3164 (+2 vs 81da9bc) [skip ci] check_hardcoded_paths.sh --full-if-capable, run 34700216664 step 43: evidence/release/0.67.0/agy/ph4-teamwork/delegate-receipt.json|/home/noah/src/aprender and |/home/noah/src/aprender/.git/config. Replaced with <repo_root>; the receipt is this session's artifact, not a tool output anyone re-reads by path. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

… GPU hosts, cuda-nightly standing-RED since the v0.66.0 sha, why gx10 idles (X64 pin, #3139 795 s, ENOSPC), §6 triage to untriaged=0, one-at-a-time drain [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…nder PR work under any intel queue pressure is P0 (selector row 0, §1, §3.4, §3.5 SSH, §5 P0·Pack + P0·Reap, §7 pack: line, §8 stops) [skip ci] Operator 2026-09-12, verbatim in §1. §3.4 no longer says one PR at a time: the queue builds 3 by design, dequeue only a KNOWN-RED group. §3.5 distinguishes durable config (forjar) from measurement/unclogging over SSH. Ground truth (§2) gains the measured X64 pin and the gx10/yoga disk layout. Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…changed (SSH measurement, 146 GB reclaimed, forjar P0·Reap live on gx10+yoga via infra#549, packing rule, spec e874769), the stacked-merge-group resolver defect #3186 → #3187, 16 ledger records incl. fleet-pack samples [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

… on #3186, the GPU-host nightlies on the cut, fleet-pack samples [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…n-room gate that discriminates nothing (#3189), the steward's unreachable live path (#3190), and the packing denominator [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

…or ruling 2026-09-13; the train publishes itself after dogfood GO + assets + preflight [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

…08 open / 424:124 arrival:closure / 95 fixed-but-open / 847 PR-less branches; five predicates, receipt in the ledger, no DONE without it (operator 2026-09-13, kaizen) [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

… (776 branches archived) / R4 1 / closure:arrival 22:61; 2 issues closed, 3 dirty PRs resolved, 1 closed, 10 verdicts pending [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

… — 3 bashrs NO-GO(s), then GO + tag at e45eaab Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

noahgift · 2026-09-13T09:12:07Z

APR-RELEASE-001 §7 — session report

APR-RELEASE-001 | did=TRAIN | train=v0.67.0 | verdict=PUBLISHED
train:   step reached T-4 | skip reason none | attended min [U] — T-4 is Noah's, unmeasured until he runs it
build:   row P0·Pack + P0·Instrument (records only) | PR https://github.com/paiml/aprender/pull/3164 (docs+ledger) | records added 1086 gate-job + 5 fleet-pack
gate:    p95 ci/gate 107.5 min | max PRs/train 40.2 | queue p95 intel 1036s yoga 832s gx10 1399s
pack:    occupancy(ledger) 1h intel 32% gx10 22% yoga 23% | 4h intel 38% gx10 28% yoga 25% | CAP 16/6/5 | intel-pressure no | verdict P0-UNDERUTILIZED
triage:  arrival 77 | closure 22 | open PRs 30 (age p95 15d) | untriaged 7
next:    0.68.0 train eligible 48 h after the v0.67.0 crates.io publish — [U] until the operator runs the attended drain (§3.2 spacing)

Tag: https://github.com/paiml/aprender/releases/tag/v0.67.0 · release commit e45eaab4796d02c6140c6cc89e146c07c0ed4c48 · 16 asset(s) on the release · preflight_pass=1 · T-4 automated: published=1, cascade check: 0 crate(s) behind (operator ruling 2026-09-13: the train publishes itself).

🤖 Generated with Claude Code

noahgift · 2026-09-13T10:38:13Z

T-5 R-4 — verdicts applied (2026-09-13T10:40Z)

PR	verdict	state
#3006	REBASE	merged onto main; guard runs bare under guard_tree; fixture builder allowlisted; armed
#3004	REBASE	merged; dogfood coverage literals re-derived (144/837); tree-reader registry re-derived; armed
#3060	REBASE	merged; ci.yml keeps both step blocks; armed
#2720	REBASE	merged; perf-solo selector on the labels runner 16 actually carries; armed
#2838	REBASE	merged; ci.yml = main (beat executes in cuda-nightly); undrafted; armed
#3134	SPLIT	R2a artifacts + coverage-nightly kept, backward.rs dropped; armed
#3056	SPLIT	review + receipts kept, cuda-nightly.yml + runner-label edits dropped; armed
#3041	SPLIT	deferred until #3004 lands (R-0b sits on R-0); comment on the PR
#3021	BEHIND→fixed	DAG block re-rendered (G-4); cuda-unit `cta64_vs_cta32_vs_cublas_fp16` failed on a GpuBuffer expect (yoga, resource) — rerun in flight

Also: mini-m4 re-registered and taking macos-arm64 (#3205); refs/pull/N/head lag fixed in the packer/union (#3200 pushed); main red 34749798339 was an env death (dep-info file vanished) and is green on rerun.

The reconcile receipt file update rides the next docs PR (this one is in the merge queue; a push would dequeue it).

noahgift added 2 commits September 12, 2026 13:13

noahgift added this to the 0.68.0 milestone Sep 12, 2026

noahgift and others added 12 commits September 12, 2026 14:33

docs(audits): PMAT-1098 receipt — four missing CHANGELOG rows (aa61219)…

a882141

…, #3068/#3139/#3175 dequeues, the T-0 correction table, G3.EX GO, and the model return (opus -> fable) recorded Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs(ledger): 15 more §3.6 records — the two stacked merge groups RED…

42cf9be

… on #3186, the GPU-host nightlies on the cut, fleet-pack samples [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs(spec): APR-RELEASE-001 T-4 is automated, never attended — operat…

0ae7bce

…or ruling 2026-09-13; the train publishes itself after dogfood GO + assets + preflight [skip ci] Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

noahgift mentioned this pull request Sep 13, 2026

feat(release): T-5 reconcile guards — check_pr_closes_issue.sh + check_reconcile.sh (APR-RELEASE-001 §6, kaizen) #3200

Merged

noahgift mentioned this pull request Sep 13, 2026

Train T-4 install/hosts must write evidence/dogfood/<version>/<host>.json so the post-publish check_multiplatform_dogfood row can pass (0.67.0: RED by construction) #3203

Open

docs(ledger): the v0.67.0 train record and 818 §3.6 collector records…

246efe0

… — 3 bashrs NO-GO(s), then GO + tag at e45eaab Pmat-Ticket: PMAT-1098 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

noahgift marked this pull request as ready for review September 13, 2026 09:17

noahgift enabled auto-merge September 13, 2026 09:17

noahgift added this pull request to the merge queue Sep 13, 2026

Merged via the queue into main with commit 36a4db5 Sep 13, 2026
18 of 20 checks passed

noahgift deleted the PMAT-1098-apr-release-001 branch September 13, 2026 11:39

noahgift mentioned this pull request Sep 13, 2026

docs(ledger): the 0.67.0 T-5 reconcile receipt, measured — and three findings about the predicates themselves #3212

Open

Navigation Menu

Uh oh!

docs(release): APR-RELEASE-001 spec + PMAT-1098 receipt for session 1 — the 0.67.0 train - #3164

docs(release): APR-RELEASE-001 spec + PMAT-1098 receipt for session 1 — the 0.67.0 train#3164
noahgift merged 17 commits into
mainfrom
PMAT-1098-apr-release-001

noahgift commented Sep 12, 2026

Uh oh!

github-actions Bot commented Sep 12, 2026 •

edited

Loading

Uh oh!

noahgift commented Sep 12, 2026

Uh oh!

noahgift commented Sep 13, 2026

Uh oh!

noahgift commented Sep 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

noahgift commented Sep 12, 2026

Uh oh!

github-actions Bot commented Sep 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

noahgift commented Sep 12, 2026

Build-server packing + build-time kaizen — 0.67.0 train, 12:35Z

Uh oh!

noahgift commented Sep 13, 2026

APR-RELEASE-001 §7 — session report

Uh oh!

noahgift commented Sep 13, 2026

T-5 R-4 — verdicts applied (2026-09-13T10:40Z)

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

github-actions Bot commented Sep 12, 2026 •

edited

Loading