Repository navigation
EPIC: release train 0.68.0 — 06x-release-schedule.md §3 (0.68.0) #3079
Description
Release train 0.68.0 — NVIDIA Rust runtime by default; yoga is a CUDA host; signed assets
Spec of record: docs/specifications/06x-release-schedule.md (PMAT-1097; this epic is §3 "0.68.0" of that document, and §2 holds every row's acceptance command). Contract: contracts/release-schedule-06x-v1.yaml.
Window (§1.1): opens 2026-09-14T18:00Z · must leave by 2026-09-15T18:00Z · planned tag 2026-09-15T06:00Z. T is the previous train's actual publishedAt; the train leaves on time — scope slips forward to the next epic with slipped_from: 0.68.0, dates do not slip. A red release gate holds the train at most 12 h.
Topic epic riding this train: #2873 (PP-066 rows R-0..R-8, C0-, G-, KEY moved to this milestone 2026-09-09) and #2869 (the installer half; the four-binary half closes with 67-A1).
Decisions apply at their §6 defaults unless overridden in a comment here (D-1 FULL tier only nightly + pre-publish dogfood · D-2 NVIDIA Rust default per layer · D-3 yoga advisory→required behind the runner-group restriction · D-4 gx10 takes arch-neutral guards · D-5 trains leave on time · D-6 parity hosts yoga + gx10 · D-7 recipe format · D-8 GH-480 rewriter fate · D-9 non-Linux binaries out). An override comment is copied into the spec's §11 with decided_by and the comment link.
Scope (spec §3, as of origin/main @ 2c584a168)
| Row | Item | Source | Host |
|---|---|---|---|
| 68-B1 | T1 loader-differential oracle (O1a/O1b/O1c) + contract | #3062 §5 | gx10 |
| 68-B2 | cuda-core default host runtime behind the registry; 3-way differential nightly | §2 B | gx10, yoga |
| 68-B3 | T2 oxide sanitizers + ptx-schedule nightly | #3062 §6 | gx10 |
| 68-D1 | CUDA 13.3 toolkit on yoga (alternatives, side-by-side) | §2 D | yoga |
| 68-D2 | cuda-unit required after 5 green nightlies |
§2 D | yoga |
| 68-C1 | guard-tree + mutants may run on gx10 (width 2, nice 19) | §2 C | gx10 |
| 68-C2 | fleet_utilization instrument + dogfood row (≥ 480 min/day gx10) | §2 C | — |
| 68-C3 | gpu-quick required after 5 green nightlies |
§2 C | gx10 |
| 68-A1 | R-5 signed manifest + KEY, R-6 installer, R-7 README; pv lane and nightly.yml off hosted runners |
milestone (#2908 #3045 #2909 #2910) | intel |
| 68-E1 | PR budget guard (dogfood row) | §2 E | — |
| 68-E2 | mutants in parallel with workspace-test | §2 E | intel |
| 68-F1 | W-A + W-B; Target c=1 decode ratio ≥ 0.85 recorded per host | §2 F | yoga, gx10 |
| 68-F2 | W-H; PP-26 witness PASS c=4/8/16 ×3 nights ×2 hosts | §2 F | yoga, gx10 |
| 68-G1 | recipe schema contract (D-7) | §2 G | — |
| 68-Q1 | Qwen 3.5 CPU Gated DeltaNet floor — 0.68 does not slip this scope (operator ruling 2026-09-15, comment); gate is gated-delta-net-v1 + qwen35-hybrid-forward-v1 + qwen35-e2e-verification-v1 contracts discharged + CPU parity reference (#3303), landing PR is #3114 (draft). Andon: 72h from 2026-09-15T12:19:39Z → 2026-09-18T12:19:39Z. Missing from this table until now — added because the underlying bug was still reproducing on main as of 2026-09-16T08:03Z and the issue had been auto-closed twice (2026-09-15T17:49Z, 2026-09-16T06:50Z) by PRs that discharge PMAT-3091 sub-work without merging the fix; see investigation comment on #3091. |
#3091 (P1) | lambda |
| milestone | R-0/R-0b registry (#2904 #3002; R-0a #3004 open), R-2 cuda default features (#2905), R-3 training banner (#2906), C0-1/C0-2/C0-4 gates (#2890 #2891 #2893), G-10b/G-11b (#3013 #3018), R-8 nightly-latest-dogfood (#3019), #2869 (closed by 67-A1's asset set, kept for the installer half) | milestone 0.68.0 (15 open) | — |
| 68-R | release | §1.3 | — |
The milestone already holds 15 issues; with the 15 rows above the train is over the §1.2 bound. Order of boarding: 68-B2/68-D1/68-A1/68-E1 first (they gate later trains), then the R-0 registry rows (68-B2 depends on them), then F, then C0/G rows; whatever is left rides 0.69 with a slipped_from: 0.68.0 note on the epic. 68-Q1 is the exception to that slip rule — per the 2026-09-15 operator ruling it does NOT slip; the train holds for it until the 2026-09-18T12:19:39Z andon deadline.
Acceptance (the train is done when)
- every row above is MERGED or moved to the next epic with
slipped_from: - pre-publish dogfood FULL (
scripts/dogfood.sh --phase pre-publish) GO - tag
v0.68.0published inside the window (or §11 amendment says why) - release carries 4
aprbinaries ({cuda,cpu} × {x86_64,aarch64}) +.sha256+ 8pvassets — checked by command, never by eye - post-publish dogfood receipts on gx10 and yoga under
evidence/dogfood/0.68.0/ - closing comment here: planned vs actual instant, receipt paths, next train's
T
🤖 Generated with Claude Code
Activity
noahgift commentedon Sep 10, 2026
Spec PR: #3087 (branch
PMAT-1097-06x-release-schedule, auto-merge armed). The §3 table in this epic's body is the state atorigin/main@2c584a168; §10 of the spec records theagy /teamwork-previewreview.noahgift commentedon Sep 12, 2026
Release estimate (orchestrator, 2026-09-12 12:20Z): eligible Mon 2026-09-14 ~19:00 UTC (48 h after the 0.67 tag), tag Tue 2026-09-15 ~16:00 UTC (window Sep 14 19:00 → Sep 15 19:00 UTC, the 72 h ceiling). Scope is whatever has merged before the cut: the 16 conflicted 0.68 PRs are resolved after the 0.67 tag, and the P0 build-ledger row lands between the trains. Slides one takt if 0.67 is SKIPPED.
Basis.
docs/specifications/APR-RELEASE-001-train-and-build-kaizen.md(PR #3164): a train becomes eligible 48 h after the previous tag onmainand must be runnable by 72 h; any RED step skips the train at that HEAD and the next eligibility is 48 h later. Train duration measured on 0.67 today: queue drain of the open PRs (~1 h per FULL-tier entry on intel), pre-publish dogfood (~1 h, 0.66 precedent), the G3.EX examples sweep (981 targets, ~5 h on a loaded box — the long pole, shrinks when P0/P1/P2 land), assets + preflight (~1 h). The crates.io cascade is attended (spec §3.7): the tag and GitHub release are autonomous, the publish waits for the operator, and its minutes are recorded from 0.67 on.release(0.67): train A — 17 row PRs of the 0.67.0 train in one build …
noahgift commentedon Sep 17, 2026
0.68.1 SHIPPED — 74/74 crates live on crates.io (cascade 2026-09-17T22:00:38Z → 22:41:36Z, 40 m 58 s, unattended under standing authorization, 429-only retries). Clean-room on the tag: B2-cpu paiml/infra run 35255889257 (A0–B5, 78,794 tests) + B2-gpu run 35268563585 (2604 tests, yoga). Post-publish
cargo install apr-cli --version 0.68.1: intel x86_64 and gx10 aarch64 →apr 0.68.1. Receipt:docs/audits/release/v0.68.1/; evidence on the release. v0.68.0 stays GitHub-only. Six gate defects found by the first complete A0–B5 chain on a tag, all fixed gate-side; the tag never moved.docs(release): v0.68.1 SHIPPED — cascade timestamps, post-publish ins…
docs(release): v0.68.1 cascade dry-run receipt — committed before the…
docs(spec): APR-RELEASE-001 revision 2026-09-17 — fan-out by construc…