fix(llm): wall-clock EVERY ollama fetch — a wedged model must never hang + saturate

Diagnosed the UserPromptSubmit '30s timeout, output discarded': it was NOT gzip and NOT an LLM in the
hook. All three UserPromptSubmit hooks are LLM-free on the critical path and fast even on a 5k-char prompt
(on-prompt-timer 0.02s · refine-prompt 0.23s · prompt-lens --emit 0.41s; the long prompt still lensed to a
4394-char focused context). The 30s was TRANSIENT machine saturation: unbounded background LLM children
(commit-story.llm, region-narrative.callQwen, ollama-monologue's /api/tags ping) hung for minutes on a
wedged ollama, and the herd starved even the ~0.4s hooks past their timeout.

Fix: AbortSignal.timeout on every ollama fetch (45s for generation, 5s for the ping) → a wedged model
fails FAST, the call returns null, the story/narration degrades gracefully (it is the knock-on, never the
receipt). Same discipline as receipt-backfill's perl-alarm render bound + the lens's already-bounded
qwenGenerate. Guard: tests/pmu-simulator/llm-calls-bounded.test.mjs — an unbounded ollama fetch is now a
red build. The deterministic pre-part stays 100% on-chip; the LLM stays a bounded, later, detached injection.

Originating-Terminal: 🔨 iTerm Builder
Relevant-Rooms: 🔨 builder, 🧪 laboratory, 🎤 voice
Story: The separation the operator asked for — LLM calls never block the on-chip receipt — enforced by bounding every one so a degraded model degrades the story, never the machine.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BUf5M5p42Zw73VaGt33ZL

commit 6895dd8d9 — 144×144 lattice triptych (on-chip)

◎ the ingest chose A,A2 — Strategy × Strategy·Goal — acting on A,B3 — Strategy × Tactics·Signal (grip 0.781) · the walk starts at the ACTOR's row · 10 file(s)

⏱ ingest 90.5ms (corpus→lattice) · definer walk 1898ms (row→transpose→row, ballistic XOR per hop) · total 2090ms
ingest here = commit-scoped SENSING only (this commit's message + changed files → lit anchors) — deep seed authoring lives in the reef-self-loop, off the commit path

rows = ACTOR (lens) · cols = PATIENT (object) · diagonal = self-reference; ◎ at actor-row × patient-col — every panel, same orientation. Heat above the diagonal = actors ranking ShortLex-earlier than their patients (the walk follows definers uphill in rank).
The 12 canonical lanes (rows top→bottom, cols left→right): A·Strategy · B·Tactics · C·Operations · A1·Law · A2·Goal · A3·Fund · B1·Speed · B2·Deal · B3·Signal · C1·Grid · C2·Loop · C3·Flow

PAIR LATTICE (the map we normally see)

PRE-WALK SENSE · INTENT grid — the snapshot, no point of view

the no-point-of-view snapshot — what the sensor lit before any walk; the leaf walk below is what bridges the two grids

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

PRE-WALK SENSE · REALITY grid — the snapshot, no point of view

the no-point-of-view snapshot, reality side — same sensor law

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

PRE-WALK Δ · straight comparison of the two ingested grids (green both · cyan intent-only · amber reality-only) · 3% overlap · ◎ = where the competence pixel sits in the raw overlap

good = green overlap without being a copy — near-100% is self-similarity, not alignment

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

◎ the pixel this commit belongs to

A,A2 — Strategy × Strategy·Goal (actor) acting on A,B3 — Strategy × Tactics·Signal (patient) · grip 0.781

actor seed A,A2 — Underwriting a strategy requires a long-horizon frame to occupy a specific quarterly goal and position target. The lattice we build today is what next year's underwriter inherits. We set the substrate that downstream pricing rides on, and that choice is paid back across every quarter we keep operating. Direction is what you keep when you forget the calendar. Multi-year posture, irreversibly chosen, defines the boundary every quarter renegotiates inside. Strategy is the bearing the rotation is measured against. Capital lives where strategy is durable. Frame the ten-year shape of the business before optimizing the ten-week sprint; the substrate decisions compound while the surface decisions decay. The strategic question is which doors close on purpose. Anything that keeps every option open until next year has already chosen the option of inaction, paid in optionality cost. The target is the cell coordinate the system MUST occupy by end-of-quarter. Anything that does not drive toward that coordinate is exhaust, and exhaust does not become substrate by being labelled progress. Vision is the coordinate the lattice rotates around — the fixed bearing every iteration is measured against. Without the fixed point, motion is movement; with the fixed point, motion is convergence. Define the goal as a verifiable predicate, not a feeling. The system terminates when the predicate evaluates true, and the predicate is what the next agent inherits when this one hands off. A goal that cannot be tested is a wish. Wishes have no termination condition, so they consume budget forever; goals have a finish line and therefore a payable price.

patient seed A,B3 — Strategic reach is framed by a long-horizon lattice that filters signal bandwidth from broadcast noise. The lattice we build today is what next year's underwriter inherits. We set the substrate that downstream pricing rides on, and that choice is paid back across every quarter we keep operating. Direction is what you keep when you forget the calendar. Multi-year posture, irreversibly chosen, defines the boundary every quarter renegotiates inside. Strategy is the bearing the rotation is measured against. Signal-to-noise is the inverse cube of bandwidth — narrow the channel, the message arrives; broaden it, recipients tune out. The right channel is the one whose floor matches the message's importance. A broadcast that addresses everyone addresses no one. Precision over reach until the address is calibrated; only then does scale convert prospects into recipients rather than into unsubscribes.

INTENT · leaf-walk clouds · BLUE→VIOLET by ply (◎ = the competence pixel: actor∩patient)

good = the clouds concentrate where the commit SAYS it works

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

REALITY · leaf-walk clouds · ORANGE→RED by ply (◎ = the same pixel — the shared perspective)

good = the same clouds as intent — shipped where declared

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

DELTA · graded cloud divergence (green agree · magenta declared-not-done · amber done-not-declared) · 2% agree

good = mostly green; magenta = said-not-done (chase it in the code), amber = done-not-said (chase it in the docs)

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

TOLERANCE · cloud topology (dim green in-shape · hot amber bleed · red orthogonal concentration) · ⚠ TOO MANY

red only matters above the flip — a few amber is normal bleed

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

SHORTLEX-3 PROJECTION (ABC · A1..C3 · the 132 children — candidate, pre-ratchet)

PROJECTED INTENT · the commit’s claims, head→axis_i tail→axis_j (cyan)

the three-length view — good = claims land inside their own zone

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

PROJECTED REALITY · same projection law (amber)

same projection law — good = reality occupies the zones intent does

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

STRAIGHT COMPARISON · green both · cyan intent-only · amber reality-only · 3% overlap

good = the zones agree; cross-zone scatter is the drift to read

	ABC A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N A1–C3
ABC A1–C3 A1A–O A2A–O A3A–O B1A–O B2A–O B3A–O C1A–N C2A–N C3A–N

shortlex-3 zone occupancy (lit cells per diagonal square)
INTENT: 0 in the 3×3 ABC corner · 8 in the 9×9 A1..C3 square · 1721 in the 132×132 children square · 362 cross-zone
REALITY: 0 in the 3×3 ABC corner · 2 in the 9×9 A1..C3 square · 743 in the 132×132 children square · 162 cross-zone
axes: identical 144 ShortLex names both sides (symmetric); weights asymmetric
zone 3 = candidate children, pre-ratchet (repo-derived dumps not yet past the perturbation-probe gate) · intent 22 claims @ θ 0.6094 · reality 8 claims @ θ 0.6094 · projection 1008ms

SHAPE-MATCH σ = -1.05 — the real intent↔reality definer walk vs random-reality (actual 0.0503 vs random 0.09). >0 = the walk distinguishes real alignment from noise; aggregate over independent walks → the divergent series.
σ_drift · noise · below zero — this read is noise, not a measurement

Most out-of-lane row: A3 · Fund — that lane is drifting most (read the red band in that row).

Three-colour tolerance (out-of-lane = reality fires where intent is weak, in orthogonal blocks): green=in-lane agreement · amber=a few out-of-lane (tolerated) · red=FIRED — orthogonal out-of-lane 66% vs tolerance 25%.
232 green · 215 amber · 865 red · 86 XOR-friction cells.

Both axes = the 144 ShortLex anchors, cell=row⊕col, on the chip. INTENT=docs+rules · REALITY=code · DELTA=friction.

tile dump — what the SimHash matched on the diagonal (inspect the sensor; tap to expand)

A,A The lattice we build today is what next year's underwriter inherits.

intent 80% (no distinctive match)

reality 73% (no distinctive match)

B,B Tactics is the choice of which leverage to pull at which beat.

intent 73% (no distinctive match)

reality 70% (no distinctive match)

C,C Operations is the loop that runs whether anyone watches.

intent 78% (no distinctive match)

reality 69% (no distinctive match)

A1,A1 EU AI Act Article 14 requires human-on-the-loop oversight, with the supervisor able to interrupt the system at

intent 78% (no distinctive match)

reality 67% // 30s timeout. AbortSignal.timeout kills the call at STORY_LLM_TIMEOUT_MS → llm() returns null

A2,A2 The target is the cell coordinate the system MUST occupy by end-of-quarter.

intent 78% (no distinctive match)

reality 72% (no distinctive match)

A3,A3 Every dollar builds either the floor (substrate that lifts future bets) or the ceiling (cap on the next round)

intent 77% (no distinctive match)

reality 70% (no distinctive match)

B1,B1 Latency is the price of a missed beat.

intent 73% (no distinctive match)

reality 61% (no distinctive match)

B2,B2 The deal sets the exchange rate between two value substrates.

intent 80% (no distinctive match)

reality 70% (no distinctive match)

B3,B3 Signal-to-noise is the inverse cube of bandwidth — narrow the channel, the message arrives; broaden it, recipi

intent 69% (no distinctive match)

reality 72% (no distinctive match)

C1,C1 Power flows where the grid lets it.

intent 75% (no distinctive match)

reality 70% (no distinctive match)

C2,C2 Each iteration is a hypothesis tested against the ground truth of the world.

intent 75% (no distinctive match)

reality 63% (no distinctive match)

C3,C3 Flow is the rate at which committed work crosses the finish line.

intent 78% (no distinctive match)

reality 73% (no distinctive match)

commit context — 10 file(s)

fix(llm): wall-clock EVERY ollama fetch — a wedged model must never hang + saturate

Diagnosed the UserPromptSubmit '30s timeout, output discarded': it was NOT gzip and NOT an LLM in the
hook. All three UserPromptSubmit hooks are LLM-free on the critical path and fast even on a 5k-char prompt
(on-prompt-timer 0.02s · refine-prompt 0.23s · prompt-lens --emit 0.41s; the long prompt still lensed to a
4394-char focused context). The 30s was TRANSIENT machine saturation: unbounded background LLM children
(commit-story.llm, region-narrative.callQwen, ollama-monologue's /api/tags ping) hung for minutes on a
wedged ollama, and the herd starved even the ~0.4s hooks past their timeout.

Fix: AbortSignal.timeout on every ollama fetch (45s for generation, 5s for the ping) → a wedged model
fails FAST, the call returns null, the story/narration degrades gracefully (it is the knock-on, never the
receipt). Same discipline as receipt-backfill's perl-alarm render bound + the lens's already-bounded
qwenGenerate. Guard: tests/pmu-simulator/llm-calls-bounded.test.mjs — an unbounded ollama fetch is now a
red build. The deterministic pre-part stays 100% on-chip; the LLM stays a bounded, later, detached i

docs/comms/self/approved/2026-07-05-bplan-wire-speed-to-trust.grade.json
docs/comms/self/approved/2026-07-05-bplan-wire-speed-to-trust.md
docs/comms/self/approved/2026-07-05-cbiz-dish-group-reply.grade.json
docs/comms/self/approved/2026-07-05-cbiz-dish-group-reply.md
docs/comms/self/approved/2026-07-05-mandate-1-capital-lock-ask.grade.json
docs/comms/self/approved/2026-07-05-mandate-1-capital-lock-ask.md
docs/comms/self/approved/2026-07-05-mandate-2-enterprise-lock-ask.grade.json
docs/comms/self/approved/2026-07-05-mandate-2-enterprise-lock-ask.md
docs/comms/self/sent/2026-07-05-bplan-wire-speed-to-trust.grade.json
docs/comms/self/sent/2026-07-05-bplan-wire-speed-to-trust.md

Produced on the chip by the running pipeline (scripts/pmu/pipeline.mjs → runPipeline → walk + xor), commit-scoped · on-chip per-walk 608.29ns (this commit). Recompute: node scripts/pmu/commit-triptych.mjs --commit 6895dd8d9

Commit artifacts (open on a computer — Gmail mobile won't follow file:// links; the inline images above are the phone-viewable form): reef grid · this triptych · the HDL — all three are also attached to this email.