Ignition + 64k measurement + no-pump: probes, five-axis harness, M/O arms, hot-window design, ack-theater deletion

[AdaWorldAPI]

Relationship to #892 (MERGED — this PR is now just the three remaining arcs)

The original oversized branch (68 commits, +24,956/−2,305) was split: its first arc (bible_wave whole-book fix + corpus module + stance lift + the BLW arm) merged as #892. This branch was then rebased onto main — a pure replay of the 29 remaining commits, zero conflicts (the branch was a strict descendant) — so this PR is now exactly the three arcs below: 48 files, +14,353/−1,816, head 463f807. The original title/body (bible_wave OT truncation) lives on in #892.

---

Arc 2 — Ignition + the 64k main model (0e59bfd..505496a, 16 commits, +8,127/−25)

• PROBE-IGNITION — GREEN (tests/probe_ignition.rs, 2/2, 11 gates with can-fire + can-stay-silent halves): first driver of the built-but-undriven write path. 64 real MailboxSoA owners from the real corpus; armed by a MetaWord write, discovered by board scan alone, cast write-on-behalf through emit_bootstrap_intent → BatchWriter::cast → run_cycle. Cycle 1 = 24 casts (20 Flow + 4 Block); cycles 5–6 rest with zero casts and no seal. Two OPEN D-MBX-A6-P4: cycle loop-closure driver — sparse seal/apply + MUL-gate thought seam (control-loop contract) #879 caveats made observable.
• D-IGN-B — GREEN (tests/d_ign_b_lenses.rs, 1/1): arming z ∈ {1..4} selects which of four stance readings is recorded over byte-identical rows; same-lens bit-identical; z=5 reserved with a printed blocker.
• PROBE-IGNITION-64K — GREEN: start() at the full main-model population — 65,536 real 1:1 owners, one StyleStrategy::plan, ONE WAL write, strictly monotone positions, then fleet-wide rest.
• OPERATOR ORDER canonized in three places: the 64k 1:1-owner model is THE main model (E-64K-1TO1-OWNERS-IS-THE-MAIN-MODEL-1); the one-tenant configuration is a benchmark harness shape, never the architecture.
• D-BLW-5 doctrine banked, build PAUSED by operator: TFPN observer-effect design (observer-effect-tfpn-doctrine.md) — distribution shape × Prozentrang payload, single-measurement law, remeasure guard. Plus the PROBE-ARC-TORQUE / Rosetta / cosine>helix plan sections (§12.10–12.10b) and the external-review triage (12 fixes + 4 reasoned skips — including the all-horizon C7 churn gate Codex independently re-found on Part 1/4: bible_wave whole-book fix + corpus module + stance lift + the BLW arm #892).

Arc 3 — measure-64k-axes (e98d865..e69d0f3, 9 commits, +4,984/−0)

• Plans v1/v2/v3: the operator-specified five-axis benchmark (one axis at a time); the rolling-epoch-closure model (WriteOrderKey, chunk baton, Libet 200 ms closure window); crypto REMOVED from the seal benchmark (seal path verified crypto-free in source; encryption = separate later layer).
• Stage A0 MEASURED (examples/measure_wal_curve.rs, release, 183 CSV rows): three of four answers; the WAL knee reported NOT REPRODUCIBLE and deliberately unclaimed (6× swings across runs — the refusal is the result).
• M-arm MEASURED — Morton LOSES under this workload/host: digests matched (valid comparison), reorder 9.4 ms, downstream −25.8 ms ⇒ Δtotal +35.2 ms; the ordered-chunk fast path was slower than the generic path. Open measurement defect logged (ISS-MARM-T1-4X-A0-GAP) blocking one cross-run comparison only.
• O-arm MEASURED — DIVERGED: ordering sourced from temporal replay does NOT reproduce the seal's ordering (64565f36… ≠ 3e71c2aa…); the seal's ordering is load-bearing for this construction. Pre-registered two-sided; three harness defects caught at the gate.

Arc 4 — no-pump + the cleanup (775fa96..463f807, 4 commits, +1,243/−1,792)

• Seal-vs-temporal answered from source (seal-vs-temporal-ordering-information.md): four things the seal encodes that temporal.rs does not (cross-owner TOTAL order vs per-owner partial; arrival as durable ordering input; the per-row coalescing fold; cohort + read horizon). Standing position: temporal.rs = temporal authority, seal = ordering authority, the gap = an explicit research question with three pre-registered probes.
• Hot version window design banked, NOT built (measure-64k-axes-v4.md): publication clock decoupled from persistence clock; barrier-flush fork chosen over version multiplexing; five panel-bought invariants H-1..H-5.
• Operator ruling E-PROGRESSION-IS-EXISTENCE-NOT-COMMAND-1: think → seal → publish version → next cycle reads it. No pump, no ack, no scheduler; ack/SLA vocabulary survives only in external consumer membranes.
• The ack theater DELETED: KanbanMsg/KanbanActor/5 RPC drivers/run_to_absorbing/KanbanRouteError removed; PhaseCensus added as the message-free visibility surface; onebrc lane E migrated to the direct &mut owner; W2b probe rewritten direct + census-over-real-SoA. OGAR boundary verified (zero consumers; its ActionHandler ack surface is legitimate membrane protocol). The kanbanstep (VersionScheduler::on_version) is NOT theater and stays canonical.

CI finding (unchanged)

cycle_driver tests are behind --features cycle-driver, which the supervisor CI step does not pass — 24+ loop-closure falsifiers never execute in CI. This branch deliberately does NOT edit the workflow (CI changes are operator-approved); the exact needed step is recorded on the board. The new test files are gated centrally by the orchestrating session.

Gates (central, re-anchored at pre-rebase head 9c5fc9e; the rebase was a pure replay onto content already validated by #892's merge)

supervisor --features supervisor: clippy --no-deps -D warnings clean, 9 lib + w2b 3/3; --features cycle-driver: probe_ignition 2/2 + probe_ignition_64k 1/1 + d_ign_b 1/1; onebrc --features lane-e 20/20 + clippy clean; measure_wal_curve full release run reproduced; fmt clean. Pre-existing unattributable reds recorded on the board (ontology oxrdf lints, bindspace.rs:475 too-many-arguments, callcenter unused import).

🤖 Generated with Claude Code

https://claude.ai/code/session_01K3RyLEbuNSHxxB3NTTrGki