release: 0.71.2

[noahgift]

Release bump for 0.71.2: bump-version.sh 0.71.2 across every workspace (--check green), the CHANGELOG section below, and the CRUX prompt certification carried to evidence/crux/0.71.2/ from 0.71.1 (apr_commit set to this bump's parent, 7c32203).

Cut by hand at its parent, the commit where the last 0.71.2 fix (#5239) merged, not by prepare_bump.sh: that script cuts only at a commit the T-2 preflight judged GO, and T-2 has never produced a GO, so it is not a gate for this train. This PR is opened unarmed.

A patch release on 0.71.1: five new RustSec advisories against quinn-proto are fixed by the upgrade, apr-cli builds again without the inference feature, the silicon-coverage read survives transient 5xx answers and a push run's token, a batch of cargo-free guards moves from the PR path to the nightly LAB lane, and 28 dark source files whose code already lived in a compiled file are deleted.

Security

• fix(deps): quinn-proto 0.11.15 -> 0.11.18 for RUSTSEC-2026-0338..0342 (fix(deps): quinn-proto 0.11.18 for RUSTSEC-2026-0338..0342 (main red) #5236)
• fix(security): interim ignore for RUSTSEC-2024-0449 (unsafe-libyaml unmaintained) — no patched release exists (Refs cargo-deny: unsafe-libyaml unmaintained (RUSTSEC-2024-0449) — interim ignore, then serde_yaml swap #5205) (fix(security): interim ignore for RUSTSEC-2024-0449 (unsafe-libyaml unmaintained, no patched release) #5206)

Fixed

• fix(apr-cli): build without the inference feature (LAB: apr-cli builds without inference (cargo check -p apr-cli --no-default-features --lib) #4972) (fix(apr-cli): build without the inference feature (#4972) #5167)
• fix(serve): size minimal llama fixture weights as [out, in] rows (Refs test fixture build_minimal_llama_gguf_with sizes Q4_K tensors from the first dim, so non-square FFN tensors get half their bytes #4985) (fix(serve): size minimal llama fixture weights as [out, in] rows (Refs #4985) #5186)
• fix(serve): ALiBi slope doc says 2^(-8(h+1)/n); delete the dark layers/a_li_bi.rs (Every src file must compile: widen the #3809 dark-file guard to the whole tree (138 orphan .rs files, 53,211 lines) #4700) (fix(serve): ALiBi slope doc says 2^(-8(h+1)/n); delete the dark layers/a_li_bi.rs (#4700) #5168)
• fix(silicon-coverage): retry a 5xx read (cap 3, failure stays not_measured); a 403 body is not a runner (fix(silicon-coverage): retry a 5xx read (cap 3, failure stays not_measured); a 403 body is not a runner #5220)
• fix(silicon-coverage): push runs read with the job token too; probe runners only where the listing is readable (silicon coverage: push runs on main are NO-GO because the job token cannot list runners (403) #5237) (fix(silicon-coverage): push runs use the job token too; probe runners only where readable (main red) #5239)
• test(train): the models row test pins the row fix(night): a dispatched run on main counts as the night, one rule at every site #5142 made, a dispatch is a night (test(train): the models row test pins the row #5142 made, a dispatch is a night #5166)

Changed

• LAB guard batch: guard: refuse a shallow fetch into the shared checkout's common git dir (2 incidents: 10-08, 10-10) #5025 check_roadmap_diff_additive.sh: SKIP row exits 0 and self-test prints n/n with no row floor (L25) #5047 LAB: deep-nightly red on main nightly (run 38008037263) #5010 LAB: CUDA Nightly (GPU QLoRA falsifiers) red on main nightly (run 38008296149) #5011 CI compute row passes a run of 0 tests (ci/sections.yml workspace-test-shard) #5048 fleet_cells_gate: a wildcard waiver waives any RED cell, and a non-date until never expires #5034 fleet_cells_gate: self-test rows for DEL(127), locale, day guards; run --self-test nightly #5064 run nightly, off the PR path (LAB guard batch (re-cut, supersedes #5216): #5025 #5047 #5010 #5011 #5048 #5034 #5064 #5223)
• chore(src): delete 27 dark src files whose code a compiled file holds (Every src file must compile: widen the #3809 dark-file guard to the whole tree (138 orphan .rs files, 53,211 lines) #4700 batch 1) (chore(src): delete 27 dark src files whose code a compiled file holds (#4700 batch 1) #5163)

Known issues

The known-issues entries carried from the previous release (#3715, #4590) are unchanged; see the [0.71.2] section of CHANGELOG.md in this PR.

Closes #5243

keep-open: #3998 #3715 #4590 #5205 #4985 #4700 -- cited by the CHANGELOG for context; each closes via its own PR; the release EPIC closes at T-4

ont-delta: none -- version bump, CHANGELOG section and a carried certification; no ontology term changes

Refs: #3998

🤖 Generated with Claude Code

[noahgift]

seat 2, aprender-wmainred APPROVE @114cfa7cf6cde6f0adcf0801f587144f1ac31a03

Checked at this head (one commit on main 7c32203, which is current main):

• Bump only. Every changed line in the 37 Cargo.toml files, Cargo.lock and crates/facades/Cargo.lock is a 0.71.1 → 0.71.2 swap: 209 removed, 209 added, and no other line. All 84 packages at 0.71.2 in Cargo.lock are local workspace members, with no registry source and no dependency moved.
• CHANGELOG. A [0.71.2] - 2026-10-12 section. Its PR refs cover every PR merged since release: 0.71.1 (#5184): #5163 #5166 #5167 #5168 #5186 #5206 #5220 #5223 #5236 #5239. It also names the issue refs #3715 #4590 #4700 #4972 #5237.
• evidence/crux/0.71.2. prompt-certification.json is the 0.71.1 file with one line changed, apr_commit → 7c32203. prompt-certification-inventory.json pins the two Qwen3.5 sources, the bf16 GGUFs and the quant sha256s.

Note (non-blocking): the certification body (del(.apr_commit)) is byte-identical across 0.70.2, 0.70.3, 0.71.0, 0.71.1 and 0.71.2. Only apr_commit moves each release, so this follows the established pattern. Its rows grade the reference engines (llama.cpp / vLLM / HF legs), not apr output, so carrying it forward is sound. Still, a reader may take apr_commit to mean "measured at", when it means "re-admitted at". A field that names the run that produced the body (or the date it was produced) would remove that ambiguity in a later release.

[noahgift]

seat 1, aprender-la-0.71 APPROVE @114cfa7cf6cde6f0adcf0801f587144f1ac31a03

Reviewed one commit, release: 0.71.2, on base main 7c3220302b. It has 41 files.

• Manifests and lockfiles: the only changes are the version string, 209 lines from 0.71.1 to 0.71.2. No other line changes once CHANGELOG.md and evidence/ are set aside. Root version = "0.71.2".
• CHANGELOG [0.71.2]: its 10 PR numbers match exactly the 10 squash commits in v0.71.1..7c3220302b, with nothing missing and nothing extra.
• evidence/crux/0.71.2/: same shape as the 0.71.1 release (#5184).
  • apr_commit is the base 7c3220302b.
  • inventory_sha256 equals the sha256 of the committed inventory.
  • prompts_sha256 equals the sha256 of scripts/crux_inference_prompts.v2.json at the base.
  • The inventory is byte-identical to 0.71.1's.

Note, not blocking: the certification differs from 0.71.1's only in apr_commit. That is expected because it is derived from the same fixed manifests (evidence/crux/0.69.1/prompt-certification/*). It is a re-stamp, not a fresh run of the model. The release's run evidence lives elsewhere, not in this file.

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=5244 head=114cfa7cf6cde6f0adcf0801f587144f1ac31a03 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.