feat(interventions): context-maximized sensemaking + measured closed loop — the fire ships prevention, probes the detector, and gets graded by the pipeline

The v1 intervention was a story; this makes it a self-improving loop:

CONTEXT MAXIMIZER (build-intervention-prompt.mjs): the sensemaking LLM now
receives every deterministic ingredient the pipeline already produced — the
FULL commit message with trailers, a bounded semantic diff, the complete
measure-history receipt row, the ENCIRCLED region narratives (coords + lanes +
per-region stories), the room's owned file surface, prior lessons WITH their
measured outcomes, and DEFINER CHAINS from the real recursive on-chip walk
seeded at the out-of-lane cells (never analytic, never guessed).

FULL QUESTION SET: semantic pull (exact diff terms → coords) · boundary-failure
class (intentional|structural|mistagged) · intent-reality gap · checklist ·
lesson (must escalate past prior ones) · MISSING GUARD (the CODE-DoD
deliverable) · detector probe (reef_signal_pct vs walk_signal_pct attribution,
walk utilization, reef-tightening experiments, gzip dictionary candidates) ·
machine-applyable mechanical payload (file_movements + configuration_updates) ·
insurability. Strict JSON, no fence.

CLOSED LOOP, MEASURED NOT ASSERTED (probe-verify.mjs): the LLM is explicitly
forbidden from inventing improvement numbers. Each new fire verifies its
predecessors deterministically — same-room mean offPct before→after (the
lattice-product delta, computed from measure-history + Originating-Terminal
trailers) and gzip dictionary ABLATION (benefit = gz(text) − (gz(dict+text) −
gz(dict)) over recent commits) — writes `outcome` back into the story, and the
NEXT prompt reads it. Live-proven: 720da3a7b measured offPct 19→21 (n=15) —
an honest 'needle unmoved' the next intervention must escalate past.

LEGS: pipeline probe → pipeline-probes.ndjson (the accumulating experiment on
the detector itself) · missing guard → ONE self-prompt routed to the
originating room. EMAIL: the encircled panel rides CID-INLINE (dogfood rule 6);
telemetry, lesson, guard, probe, and mechanical payload all render. /intervene
publishes everything including the measured outcomes.

Guard extended in the SAME commit: tests/ops/intervention-fire.test.js 8→13
invariants (maximal context · full question set · CID panel · deterministic
legs · measured-not-self-graded loop).

Originating-Terminal: 🎩 operator
Relevant-Rooms: 🔨 builder, 📐 architect, 🧪 laboratory
Story: The operator asked the intervention to stop being a drift story and become the repo's immune system — sense the breach with every ingredient the receipt already computed, propose falsifiable fixes, and let the pipeline (never the model) grade whether they worked. This commit closes that loop end to end and ratchets it with 5 new guard invariants.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

commit 604f4b72e — 144×144 lattice triptych (on-chip)

◎ the ingest chose A1,A3 — Strategy·Law × Strategy·Fund — acting on B3,A1 — Tactics·Signal × Strategy·Law (grip 0.766) · the walk starts at the ACTOR's row · 7 file(s)

⏱ ingest 656.4ms (corpus→lattice) · definer walk 249ms (row→transpose→row, ballistic XOR per hop) · total 1014ms
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) · 4% 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

A1,A3 — Strategy·Law × Strategy·Fund (actor) acting on B3,A1 — Tactics·Signal × Strategy·Law (patient) · grip 0.766

actor seed A1,A3 — Compliance articles provide the binding mandate that oversees how capital and the dollar floor fund finance. EU AI Act Article 14 requires human-on-the-loop oversight, with the supervisor able to interrupt the system at any moment. Compliance attaches to the artifact the artifact cannot escape — provenance is a regulatory primitive, not a feature. Constraints precede objectives. The cap-table, the license text, the consent envelope — each draws the inviolate line the action must respect, and the line is enforced before the optimization runs. A rule names what the system MAY NOT do. If the agent can do X, then 'forbid X' is not the constraint — it is wishful thinking dressed as policy. Constraints live in the gate, not in the prompt. Regulation is not advisory. The fine is the price tag stapled to the omitted control; insurance is the option pricing on the failure-mode the regulation already named. Every dollar builds either the floor (substrate that lifts future bets) or the ceiling (cap on the next round). Allocation is which one this dollar chooses, and the choice is irreversible by the time the wire clears. Burn rate divides into substrate-building and exhaust. Substrate compounds; exhaust evaporates by quarter-end. The investor who pays for substrate is buying a different asset than the investor who pays for exhaust. Capital allocation is the irreversible vote — the bet that closes options to keep the others open. Each dollar deployed buys both the asset and the foreclosure of every alternative the dollar could have funded. The funding question is the runway question. Eighteen months at one burn rate is a different company than twelve months at a higher burn rate, even when the bank balance matches.

patient seed B3,A1 — Signal reach and message broadcast channels are governed by the binding mandate and governance rule of law. EU AI Act Article 14 requires human-on-the-loop oversight, with the supervisor able to interrupt the system at any moment. Compliance attaches to the artifact the artifact cannot escape — provenance is a regulatory primitive, not a feature. Constraints precede objectives. The cap-table, the license text, the consent envelope — each draws the inviolate line the action must respect, and the line is enforced before the optimization runs. A rule names what the system MAY NOT do. If the agent can do X, then 'forbid X' is not the constraint — it is wishful thinking dressed as policy. Constraints live in the gate, not in the prompt. 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. Repetition is the cost of being heard. Three sends to one recipient beats one send to three; the first send announces, the second send is heard, the third send is remembered. Every message carries provenance. The sender's history is half the message's weight; an unknown sender pays in attention what a known sender pays in seconds saved.

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) · 3% 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)

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 · 12 in the 9×9 A1..C3 square · 1669 in the 132×132 children square · 299 cross-zone
REALITY: 0 in the 3×3 ABC corner · 0 in the 9×9 A1..C3 square · 858 in the 132×132 children square · 113 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 27 claims @ θ 0.6094 · reality 90 claims @ θ 0.6406 · projection 563ms

SHAPE-MATCH σ = 1.16 — the real intent↔reality definer walk vs random-reality (actual 0.0979 vs random 0.067). >0 = the walk distinguishes real alignment from noise; aggregate over independent walks → the divergent series.
σ_drift · weak · panel story is not yet evidence

Most out-of-lane row: C2 · Loop — 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=armed — orthogonal out-of-lane 23% vs tolerance 25%.
473 green · 552 amber · 0 red · 41 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 72% (no distinctive match)

reality 75% if (lines.length > maxLines) diff = lines.slice(0, maxLines).join('\n') + `\n… [diff truncated

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

intent 70% (no distinctive match)

reality 73% 9. MECHANICAL INTERVENTION — the machine-applyable payload (empty arrays when not warranted; pr

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

intent 75% (no distinctive match)

reality 80% !/usr/bin/env bash. scripts/interventions/intervention-fire.sh <sha> — THE INTERVENTION KNOCK-O

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

intent 72% (no distinctive match)

reality 83% == DEFINER CHAINS (the REAL recursive on-chip walk, seeded at the encircled out-of-lane cells —

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

intent 75% (no distinctive match)

reality 77% function repeatContext(roomLabel) {

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

intent 73% (no distinctive match)

reality 78% } else if (cmd === 'legs') {

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

intent 70% (no distinctive match)

reality 75% # CID-INLINE (PMU dogfood rule 6: Gmail strips data: URIs — cid images render).

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

intent 73% (no distinctive match)

reality 78% a. reef_signal_pct / walk_signal_pct: integers 0-100 summing to 100 — your estimate of how much

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

intent 73% (no distinctive match)

reality 78% if (!sect) return [];

C1,C1 Power flows where the grid lets it.

intent 73% (no distinctive match)

reality 77% !/usr/bin/env node. scripts/interventions/probe-verify.mjs — THE CLOSED LOOP: measure, don't as

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

intent 75% (no distinctive match)

reality 73% const dir = resolve(REPO, 'data/interventions/stories');

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

intent 75% (no distinctive match)

reality 75% const row = measureRow(shaShort);

commit context — 7 file(s)

feat(interventions): context-maximized sensemaking + measured closed loop — the fire ships prevention, probes the detector, and gets graded by the pipeline

The v1 intervention was a story; this makes it a self-improving loop:

CONTEXT MAXIMIZER (build-intervention-prompt.mjs): the sensemaking LLM now
receives every deterministic ingredient the pipeline already produced — the
FULL commit message with trailers, a bounded semantic diff, the complete
measure-history receipt row, the ENCIRCLED region narratives (coords + lanes +
per-region stories), the room's owned file surface, prior lessons WITH their
measured outcomes, and DEFINER CHAINS from the real recursive on-chip walk
seeded at the out-of-lane cells (never analytic, never guessed).

FULL QUESTION SET: semantic pull (exact diff terms → coords) · boundary-failure
class (intentional|structural|mistagged) · intent-reality gap · checklist ·
lesson (must escalate past prior ones) · MISSING GUARD (the CODE-DoD
deliverable) · detector probe (reef_signal_pct vs walk_signal_pct attribution,
walk utilization, reef-tightening experiments, gzip dictionary candidates) ·
machine-applyable mechanical payload (file_movements + configuration_u

data/interventions/pipeline-probes.ndjson
data/interventions/stories/720da3a7b.json
scripts/interventions/build-intervention-prompt.mjs
scripts/interventions/intervention-fire.sh
scripts/interventions/probe-verify.mjs
src/app/intervene/page.tsx
tests/ops/intervention-fire.test.js

Produced on the chip by the running pipeline (scripts/pmu/pipeline.mjs → runPipeline → walk + xor), commit-scoped · on-chip per-walk 1097.07ns (this commit). Recompute: node scripts/pmu/commit-triptych.mjs --commit 604f4b72e

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.