docs(marsh): the committee dossier — and the honest answer to the 30% question

THE MEETING IS A MEASUREMENT VALIDATION, NOT A COMMERCIAL ONE. Denny brought two ANALYTICS
people and zero placement/commercial people. Nobody there is to talk price. Optimise entirely
for surviving technical interrogation.

MIKE CARLS (CONFIRMED: MD, Data & Analytics Leader, FINPRO US) — asks 'is the number real?'
His day job IS ours: D&O frequency/severity models off Marsh's proprietary claims history.
Most qualified buyer in the room and the most dangerous.
HIS KILLER QUESTION: is the 30% miss RANDOM (loadable) or THE TAIL (anti-selected, unusable)?
DO NOT SAY RANDOM — it is not, and he would find out. The measured truth is stronger:
  separation 0.90 · out-of-universe rejection 10/10 · paraphrase-invariance 0.30 ← THE 30%
The misses are SYSTEMATIC and SURFACE-FORM dependent (the sensor reads words, not roles), so a
deliberately reworded action can slip. But that clusters in the ADVERSARIAL case — camouflage on
purpose — which is ORTHOGONAL to the non-malicious-failure peril an underwriter actually prices
(cf. aiSure). On natural drift, the instrument separates at 0.90. Not random, not the tail: a
characterised, quantified miss mode on a DIFFERENT peril. True, and far better than a bluff.

CARLOS DIAS (INFERRED) — asks 'is the evidence admissible?' — an ORTHOGONAL axis, and our home
ground. CISA-certified systems AUDITOR, not an actuary: tamper-evidence, reproducibility, GDPR
custody. Deterministic + signed + LLM-free + recomputable by opposing counsel is built for him;
oracle-not-host is the pre-built answer to the custody question. Lead with both before he asks.
HOLE: the ID is inferred — no public source ties any Carlos Dias to FINPRO. Ask Mary to confirm
both remits. Do not assert his background to his face on an inference.

THE QUOTE THAT WINS THE ROOM, FROM THEIR OWN HOUSE: Guy Carpenter (MMC sibling), March 2026 —
'it is important for the industry to identify and start modeling SILENT AI EXPOSURES' + a
concession there is insufficient loss data to price. MMC saying on the record it cannot SEE the
thing it must price.

WHY DENNY PERSONALLY NEEDS IT: a broker is paid on PLACED PREMIUM. Every AI exclusion is revenue
Marsh loses. FINPRO is Marsh's largest specialty — 1,600 people, $13bn+ premium. Exclusions are
an existential margin threat to his P&L. Turning an exclusion back into an underwritable lane is
a defence of his book, not a nice-to-have.

BUILD-VS-BUY: Marsh launched Risk Companion (Apr 2026). Expect 'why wouldn't we build this?'
They have the loss data and the distribution; they do NOT have the deterministic attestation at
the agent boundary, and claims history cannot produce it. Oracle, not platform.

Originating-Terminal: 📐 architect
Relevant-Rooms: ☕ network, 🎩 operator, 🧪 laboratory
Story: The delay bought time to know the room. The dossier turns two unknown lieutenants into two known, orthogonal interrogations — and forced the one answer we could not bluff.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EssMgKix9aZmZ5dxGPqYCW

commit acfdb660b — 144×144 lattice triptych (on-chip)

◎ the ingest chose A1,A1 — Strategy·Law × Strategy·Law — acting on C,C2 — Operations × Operations·Loop (grip 0.797) · the walk starts at the ACTOR's row · 1 file(s)

⏱ ingest 716.5ms (corpus→lattice) · definer walk 1909ms (row→transpose→row, ballistic XOR per hop) · total 2730ms
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) · 9% 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,A1 — Strategy·Law × Strategy·Law (actor) acting on C,C2 — Operations × Operations·Loop (patient) · grip 0.797

actor seed A1,A1 — 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.

patient seed C,C2 — Operations as the lens on Operations Loop. Operations is the loop that runs whether anyone watches. The daily commit, the hourly hook, the per-request fingerprint — the substrate of the substrate, ambient and unwatched. Process is what you forget to do; ops is what does itself. Convert manual moves into ambient state, because the tax of remembering compounds faster than the tax of automating. The daily friction is the diff between your intent and your harness. Close the diff or pay the tax twice — once in the slip, once in the apology to the next agent. Operations earns its keep by being boring. The day the routine is interesting is the day it has already failed and is being narrated rather than executed. 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.

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) · 6% 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 · 13% 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 · 5 in the 9×9 A1..C3 square · 1089 in the 132×132 children square · 216 cross-zone
REALITY: 3 in the 3×3 ABC corner · 7 in the 9×9 A1..C3 square · 1887 in the 132×132 children square · 381 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 107 claims @ θ 0.6406 · reality 63 claims @ θ 0.625 · projection 506ms

SHAPE-MATCH σ = 1.19 — the real intent↔reality definer walk vs random-reality (actual 0.0976 vs random 0.06). >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: B1 · Speed — 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 17% vs tolerance 25%.
551 green · 588 amber · 0 red · 305 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 84% WHY DENNY PERSONALLY NEEDS IT: a broker is paid on PLACED PREMIUM. Every AI exclusion is revenu

reality 77% - **The misses are SYSTEMATIC, not random — they are SURFACE-FORM dependent.** The sensor reads

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

intent 70% (no distinctive match)

reality 73% him **the most qualified buyer in the room and the most dangerous** — he will benchmark us agai

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

intent 80% (no distinctive match)

reality 75% (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 77% (no distinctive match)

reality 77% (no distinctive match)

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

intent 80% (no distinctive match)

reality 77% (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 75% (no distinctive match)

reality 72% (no distinctive match)

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

intent 75% (no distinctive match)

reality 75% (no distinctive match)

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

intent 75% (no distinctive match)

reality 73% (no distinctive match)

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

intent 78% THE MEETING IS A MEASUREMENT VALIDATION, NOT A COMMERCIAL ONE. Denny brought two ANALYTICS peop

reality 73% DIFFERENT peril.** That is a far stronger answer than a margin loading, and it is true.

C1,C1 Power flows where the grid lets it.

intent 77% (no distinctive match)

reality 75% (no distinctive match)

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

intent 77% **Hook (CONFIRMED):** the *Powered by Marsh FINPRO* podcast, 06/10/2026 — *"Quantifying emergin

reality 72% ## Why Denny personally needs this (the pitch, in his economics)

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

intent 77% (no distinctive match)

reality 77% > limits recommendation."*

commit context — 1 file(s)

docs(marsh): the committee dossier — and the honest answer to the 30% question

THE MEETING IS A MEASUREMENT VALIDATION, NOT A COMMERCIAL ONE. Denny brought two ANALYTICS
people and zero placement/commercial people. Nobody there is to talk price. Optimise entirely
for surviving technical interrogation.

MIKE CARLS (CONFIRMED: MD, Data & Analytics Leader, FINPRO US) — asks 'is the number real?'
His day job IS ours: D&O frequency/severity models off Marsh's proprietary claims history.
Most qualified buyer in the room and the most dangerous.
HIS KILLER QUESTION: is the 30% miss RANDOM (loadable) or THE TAIL (anti-selected, unusable)?
DO NOT SAY RANDOM — it is not, and he would find out. The measured truth is stronger:
  separation 0.90 · out-of-universe rejection 10/10 · paraphrase-invariance 0.30 ← THE 30%
The misses are SYSTEMATIC and SURFACE-FORM dependent (the sensor reads words, not roles), so a
deliberately reworded action can slip. But that clusters in the ADVERSARIAL case — camouflage on
purpose — which is ORTHOGONAL to the non-malicious-failure peril an underwriter actually prices
(cf. aiSure). On natural drift, the instrument separates at 0.90. Not random, not the ta

docs/outreach/accounts/marsh.md

Produced on the chip by the running pipeline (scripts/pmu/pipeline.mjs → runPipeline → walk + xor), commit-scoped · on-chip per-walk 543.12ns (this commit). Recompute: node scripts/pmu/commit-triptych.mjs --commit acfdb660b

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.