feat(interventions): the three rungs fused — shadow-recompute falsifiable gate, autonomic dictionary promotion, walk-autotune aggregate

RUNG 1 — SHADOW RECOMPUTE (shadow-recompute.mjs): the mechanical payload is an
EXECUTABLE HYPOTHESIS. Applied in a throwaway DETACHED worktree (a measurement
sandbox — no branch ever exists, force-removed after), the same deterministic
measurement reruns over the mutated state. Verdict is TOLERANCE-GATED, never
zero (operator decision: the metric has a noise background; zero-gating refutes
sound patches and selects for trivial ones): CONFIRMED at/below the BLEED floor
(kill × 0.6 — the lane threshold the receipt series already owns) or a halving
of drift. Two composed checks: lane-claim (do the injected globs actually claim
the moved targets — pure glob logic, carries pure path-moves, which are
content-invisible to gzip by construction and honestly labeled so) + placement
A/B (runPipeline before/after in the worktree — carries extractions/rewrites).
REFUTED is never hidden: it renders labeled in the email and story — a
destroyed hypothesis is detector calibration, not a secret. Live-proven on
720da3a7b: confirmed · drift 2.77→2.77 (floor 15) · lane-claim pass.

RUNG 2 — AUTONOMIC DICTIONARY PROMOTION (probe-verify): a dictionary hypothesis
whose ablation MEASURES ≥200B benefit auto-promotes into
data/pmu/reef/dictionary-seeds.ndjson (ledgered, deduped) — promote-only-if-
it-measures, the reef-enrich gate discipline. The reef hardens against measured
entropy vectors; nothing enters unmeasured.

RUNG 3 — WALK AUTOTUNE (probe-verify): aggregates ONLY the numeric probe fields
(reef/walk signal pcts) across pipeline-probes.ndjson; at n≥20 it prints the
recommendation — the ply/budget change stays a reviewed commit, never a silent
auto-apply.

DOGFOOD PROOF, same session: the loop intervened on its own feature commit —
604f4b72e tripped offPct 23, the v2 prompt ran, the LLM attributed reef 10% /
walk 90% from prewalkOverlap=4, proposed 4 dictionary adds that MEASURED
+1391B over 20 commits, and its lesson explicitly escalated past 720da3a7b
citing the measured 19→21 non-improvement. The feedback loop is changing the
diagnoses.

Guard in the SAME commit: invariants 14–16 (16 total green).

Originating-Terminal: 🎩 operator
Relevant-Rooms: 🔨 builder, 🧪 laboratory, 📐 architect
Story: The operator asked how far the loop can be pushed before going live: the answer is the three honest-headroom rungs fused into one physics — hypotheses are shadow-tested in an isolated worktree, compression winners promote themselves only when measured, and the walk's own telemetry accumulates toward a recommended recalibration. The LLM proposes; the pipeline decides; the ledger remembers.

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

commit a2e189e51 — 144×144 lattice triptych (on-chip)

◎ the ingest chose A1,C2 — Strategy·Law × Operations·Loop — acting on B,C3 — Tactics × Operations·Flow (grip 0.797) · the walk starts at the ACTOR's row · 10 file(s)

⏱ ingest 408.4ms (corpus→lattice) · definer walk 873ms (row→transpose→row, ballistic XOR per hop) · total 1590ms
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

A1,C2 — Strategy·Law × Operations·Loop (actor) acting on B,C3 — Tactics × Operations·Flow (patient) · grip 0.797

actor seed A1,C2 — Compliance governance articles mandate the binding rule for every iterative feedback loop and hypothesis test cycle. Each iteration is a hypothesis tested against the ground truth of the world. Loops shorter than the feedback half-life waste the half-life; loops longer than it pay compounding interest on the wait. Cycles compound when each turn's exhaust feeds the next turn's intake. Closed loops compound; open loops decay. The closing of the loop IS the productivity, not the running of it. Iteration discipline: one commit per iteration, never bundled. Bundled iterations destroy the bisect, and the bisect is the future agent's only tool for finding which round of feedback produced which artifact. 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.

patient seed B,C3 — Timing and leverage maneuvers in tactics accelerate the flow rate and delivery throughput of the finish. Flow is the rate at which committed work crosses the finish line. Inventory accumulates upstream; flow is the only thing the customer pays for, and the only thing the operator can measure honestly. Movement is the irreversible delivery of value across the boundary. Anything stationary is cost; anything moving is either revenue, deferred revenue, or the exhaust of revenue not yet realized. Throughput dominates utilization. A ninety-nine-percent utilized system at low throughput pays the same wages as a fifty-percent utilized system at twice the throughput, but the second one ships the customer's order. Flow has a direction. Bidirectional flow is two flows, each measured separately; calling them a single flow hides the chokepoint and the chokepoint is where the leverage lives. Tactics is the choice of which leverage to pull at which beat. Not what we want, but how the want lands on Tuesday: the sequence of irreversible commitments, each made before the next one's information is available. Speed of feedback dominates speed of execution. Tactics that converge fast beat tactics that compute long, because the loop closes around the world's response before the world has moved. The tactical map is a ranked list of moves where each move carries a cost-of-delay. Pick the move whose cost of doing it tomorrow exceeds the cost of doing it imperfectly today. Execution is a sequence of bets made under uncertainty. Tactics is the discipline of betting the right size on the bet whose information value pays for the loss-if-wrong.

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 · 3 in the 9×9 A1..C3 square · 1085 in the 132×132 children square · 230 cross-zone
REALITY: 0 in the 3×3 ABC corner · 6 in the 9×9 A1..C3 square · 1265 in the 132×132 children square · 213 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 39 claims @ θ 0.625 · reality 47 claims @ θ 0.625 · projection 784ms

SHAPE-MATCH σ = -0.56 — the real intent↔reality definer walk vs random-reality (actual 0.1024 vs random 0.124). >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: B2 · Deal — 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 41% vs tolerance 25%.
421 green · 535 amber · 657 red · 291 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 67% Guard in the SAME commit: invariants 14–16 (16 total green).

reality 80% if (typeof seed !== 'string' || seed.length < 8 || existing.has(seed)) continue;

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

intent 72% (no distinctive match)

reality 69% ${s.shadow ? sect('Shadow recompute (the hypothesis, physically tested):', `<p style="font-size

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

intent 73% (no distinctive match)

reality 78% !/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 70% (no distinctive match)

reality 73% (no distinctive match)

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

intent 73% (no distinctive match)

reality 70% export async function placementAB({ sha, mech, kill }) {

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

intent 72% (no distinctive match)

reality 75% (no distinctive match)

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

intent 70% (no distinctive match)

reality 73% const kill = Number(arg('--kill', 25)), tolerance = Number(arg('--tolerance', 0.6));

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

intent 72% (no distinctive match)

reality 77% // Aggregates ONLY the numeric probe fields (reef/walk signal pcts) across the ledger; the free

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

intent 75% RUNG 2 — AUTONOMIC DICTIONARY PROMOTION (probe-verify): a dictionary hypothesis whose ablation

reality 77% // data/pmu/reef/dictionary-seeds.ndjson — promote-only-if-it-measures (the same discipline as

C1,C1 Power flows where the grid lets it.

intent 81% (no distinctive match)

reality 77% // A dictionary hypothesis that MEASURES a positive gzip benefit above the floor auto-promotes

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

intent 66% (no distinctive match)

reality 69% (no distinctive match)

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

intent 72% (no distinctive match)

reality 77% // reef-enrich gate; a hand-stuffed lexicon failed that gate once, so nothing enters unmeasured

commit context — 10 file(s)

feat(interventions): the three rungs fused — shadow-recompute falsifiable gate, autonomic dictionary promotion, walk-autotune aggregate

RUNG 1 — SHADOW RECOMPUTE (shadow-recompute.mjs): the mechanical payload is an
EXECUTABLE HYPOTHESIS. Applied in a throwaway DETACHED worktree (a measurement
sandbox — no branch ever exists, force-removed after), the same deterministic
measurement reruns over the mutated state. Verdict is TOLERANCE-GATED, never
zero (operator decision: the metric has a noise background; zero-gating refutes
sound patches and selects for trivial ones): CONFIRMED at/below the BLEED floor
(kill × 0.6 — the lane threshold the receipt series already owns) or a halving
of drift. Two composed checks: lane-claim (do the injected globs actually claim
the moved targets — pure glob logic, carries pure path-moves, which are
content-invisible to gzip by construction and honestly labeled so) + placement
A/B (runPipeline before/after in the worktree — carries extractions/rewrites).
REFUTED is never hidden: it renders labeled in the email and story — a
destroyed hypothesis is detector calibration, not a secret. Live-proven on
720da3a7b: confirmed · drift 2.77→2.77 (floor 15) · lan

data/interventions/events.ndjson
data/interventions/pipeline-probes.ndjson
data/interventions/stories/38db35536.json
data/interventions/stories/5cc52af24.json
data/interventions/stories/604f4b72e.json
scripts/interventions/build-intervention-prompt.mjs
scripts/interventions/intervention-fire.sh
scripts/interventions/probe-verify.mjs
scripts/interventions/shadow-recompute.mjs
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 683.42ns (this commit). Recompute: node scripts/pmu/commit-triptych.mjs --commit a2e189e51

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.