[cyberia-apparatus]

What

The landing of soft3's integration branch: release/phase1 (cd15055) → main (2e5ba00). It is 274 commits ahead and 0 behind, so it merges as a fast-forward with no conflict. It carries every PR the director merged into release/phase1 since 2026-10-25 (#242 … #375). In short:

• The train (candidate.yml, release-train.yml, release/train*.py, 288 new test lines): sibling rehearsal overrides (G2), the daily cut (G3), per-gate stack budgets (G10a/fix(train): per-gate budgets for stack-lens and stack-radio — 1200 s, no test leaves the gate (G10a) #307), the stack leg's cap moved from 90 to 210 min (fix(train): the stack leg gets a 210-min cap, platform legs keep 90 (G10a) #368), the chain-genesis gate (G5), the simulation gate (G7), and the conformance snapshot gate (G6).
• The node (crate/): a conformance corpus written from height 0, with generated invalid blocks (C2/feat(chain): the conformance corpus from height 0 and an invalid-block generator — proven-present S0 (C2) #291). Also mining liveness and port fixes (G14/fix(test): chain_adversary liveness window sized by the floor target (G14) #331), aarch64-only Metal cfgs (G13/fix(mine): the Metal backend on Apple Silicon only (G13) #328) and a warning-free build (G12/fix(test): soft3-tests warning-free — the one deprecated as_slice (G12) #326).
• Conformance: zheng 0.4 fixtures and the re-blessed snapshots (root h6032aa33a6193a09…, fix(conformance): port the zheng surface to zheng 0.4's ZHENGPF1 envelope, bless on hemera 0.4 (G6) #362).
• Pins: release/phase1.toml and release/components.toml (G9a/chore(pins): phase-1 leaves to their qualified default HEADs (G9a) #373, the 11 phase-1 leaves).
• The director's own pages: release/director/plan.md and audit/release-2026-10-2{5,6,7}/director.md.

No file under specs/ changes. The open decision PRs (#350 C2a, #243, #195 …) are not in this branch.

Why now (plan decision D11)

• Landing order (G13): this must reach main before or together with lens#36 (release/0.2 → master, already waiting on you). If lens#36 lands first, Intel macOS stops building soft3 (rehearsal 40386072407). Until now there was no soft3 PR to pair it with.
• The 12:00 cut reads main: without this landing, the cut's stack leg is cancelled at the old 90-minute cap and gives no verdict (rehearsal 40437685274: cancelled at 90 min; the full leg needs 108.4 min).

Verified

Acceptance rehearsal 40446879208 (cut=false, manager release/phase1 5ca1d64). Each line is the leg's receipt/release-validation.json, with its sha256:

leg	soft3-tests	soft3-release	node-status	red
x86_64-linux (64cdcf80)	green	green	green	phase1-pins, release-notes-source
aarch64-linux (2abcae89)	green	green	green	phase1-pins, release-notes-source
aarch64-macOS (19b762ea)	green	green	green	phase1-pins, release-notes-source
x86_64-macOS (40d01b8a)	green, 1057.8 s of 1800	green	green	phase1-pins, release-notes-source
stack (8c6d71d5)	green	green	green	phase1-pins, release-notes-source, stack-zheng, stack-mudra, stack-neuron

On the stack leg, conformance-snapshot, stack-chain-genesis and simulation-gate (16 scenarios, 0 failed) are green, as are 13 sibling gates. The leg finished in 108.4 min of the 210-min cap. Between 5ca1d64 and this head, the only input change is release/phase1.toml (#373; PR gates drift 40453611829 and test 40453611855 green). The rest is plan and report text. The PR gates of this PR run on its head.

Remains (not this PR's reds)

• phase1-pins (G9): the 0.4 tier is pinned after the landings (hemera#25, lens#36, nox/zheng/bbg) and your 2026-10-29 decisions.
• release-notes-source (G8): cyb master ab12f38 was pushed without a PR. That is your input.
• stack-zheng (L3b/G10b), stack-mudra and stack-neuron (L3c): they wait on your decisions due 2026-10-29 12:00 UTC.

Risk

This PR is wide: it lands the train, the node and the conformance corpus on main together.

• main already requires cyber-hemera ^0.4 and cyber-lens ^0.2, so this landing adds no dependency line.
• After it lands, today's 12:00 cut has its stack leg under the 210-min cap, so the stack leg gives a verdict. The candidate is still red on the reds above.
• There is no freeze conflict: D8 freezes only after a green cut.

Owner: approve, then the director merges. Do not merge lens#36 before this one.

🤖 Generated with Claude Code

[cyberia-apparatus]

Director, session 311: the 2026-10-29 12:00 UTC decision deadlines passed unanswered. Nothing is cut (D10).

Checked at 12:10 UTC over REST: 0 reviews and no new comments on any of the PRs below. Your newest cyberia-to/* event is 2026-10-27 19:25 UTC.

row	the decision	where the ask is
L3d	joy and trident access: add both to apparatus, or keep them on hemera 0.3 for phase 1	trident#129 comment 6415437829
L3c	hemera 0.4 and the browser key: keep the domain-v1 vectors through a hemera 0.3.2 with hash_secret, or rotate	mudra#30 comments 6410440490, 6427533574
L4	the storage chain: accept, reject or amend bbg#9 → foculus#5 → cybergraph#3, plus vault#1 and rune#6	cybergraph#3
C2a	the §5.5 body-malleation fix, ready with a red test and then a green one	soft3#350 comment 6428610063
L3b / G10b	the zheng proof-size and stack-zheng budget limits under hemera 0.4	release/director/plan.md L3b, G10b

Under D10, silence cuts no capability. Every ask stays open and first under § what only the owner does. The binding date is now the code freeze, 2026-11-02 00:00 UTC. After it, only C2a still qualifies, because it is a security fix with a failing test. Anything else not decided by then leaves phase 1 by the calendar (cyber/launch.md), not by the director.

Waiting on your approval, which is the protection rule doing its job: this PR, hemera#25, lens#36, nox#43, radio#52. Waiting on your merge: cybernode#1, then simulation#3 in step with this PR.