docs(marsh): agenda HTML+PDF — and HOLD the numbers until the tuned run

OPERATOR'S CHALLENGE, AND HE IS RIGHT: '0.90/0.30 were measured on ONE setup — just because we
measured it on a bad setup does not mean that is the floor.'

PROVENANCE VERIFIED (not invented): 0.90 separation / 0.30 paraphrase-invariance trace to the
documented sealed run ('the FAIR test RUN: full spec, full reef, real tolerances'). The AUC set
0.496/0.415/0.406 is the drift-vs-defects experiment that forced the prediction→provenance pivot.
Both are real repo numbers. But BOTH were measured on ONE configuration.

THE ASYMMETRY THAT DECIDES IT:
· send now, tuned number is BETTER → we have PERMANENTLY anchored an actuary to a worse figure.
  He prices against 0.30, and we can never un-ring it because WE published it.
· send now, tuned number is WORSE → catastrophic; he finds it in diligence after we wrote
  'should improve'.
· RUN FIRST → we either walk in with a stronger honest number, or we discover the 0.30 is
  structural BEFORE Marsh does.
There is no branch where running it first is wrong. The meeting is not scheduled yet — we have the
one thing we will not have later: time.

SO: the numbers section is REPLACED with a promise ('I will bring the full measured set, including
the ones that look bad') and the full section is held in marsh-numbers-HELD.md. Everything that
makes the agenda strong survives without them — liability allocation, the five facts, T.J. Hooper,
the schema homework. Tuned held-out parked to laboratory, due 07-15, BLOCKING.

QWEN GHOST-READ AS CARLS CAUGHT A REAL FAILURE FIRST: the AUC null read as PRODUCT FAILURE, not
honesty — 'the AUC values are so low they're basically random, I'd need to see something much
stronger.' The number was arriving BEFORE the frame, so his actuarial reflex fired and the
credibility payoff never landed. Our strongest credential was functioning as our own kill shot.
Reordered: the claim we are NOT making comes first, the number second AS PROOF of the honesty.
Also glossed every metric in plain English (the table read as 'another vendor's flashy metrics
table' — he did not know what 'out-of-universe rejection' meant).

Originating-Terminal: 📐 architect
Relevant-Rooms: 🧪 laboratory, ☕ network, 🎩 operator
Story: The operator asked whether the numbers were real or invented. They are real — and asking forced the better question: they were measured once, on one setup, and writing them to an actuary makes them permanent. Run the experiment first.

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

commit 2b7d155b8 — 144×144 lattice triptych (on-chip)

◎ the ingest chose A,C3 — Strategy × Operations·Flow — acting on C2,B1 — Operations·Loop × Tactics·Speed (grip 0.828) · the walk starts at the ACTOR's row · 5 file(s)

⏱ ingest 691.3ms (corpus→lattice) · definer walk 565ms (row→transpose→row, ballistic XOR per hop) · total 1436ms
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) · 27% 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,C3 — Strategy × Operations·Flow (actor) acting on C2,B1 — Operations·Loop × Tactics·Speed (patient) · grip 0.828

actor seed A,C3 — Establishing a long-horizon strategy frame allows the flow rate and delivery throughput to cross the lattice. 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. 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.

patient seed C2,B1 — Feedback loop measures and hypothesis tests are used to optimize the speed and tempo of execution. Latency is the price of a missed beat. Every hundred milliseconds of round-trip turns a parallel race into a stalemate, because the first arrival sets the rules the second arrival has to honor. Throughput compounds: doubling the rate doubles the surface area where serendipity can hit. A pipeline at twice the velocity is not twice as productive — it is exponentially more lucky. Time-to-first-byte is the only number the operator feels at the wheel. Optimize the felt latency before the raw throughput; the operator's grip on the loop is what drives the loop forward. Velocity is irreversible. A delivery sent fast cannot be unsent slow; a release deployed cannot be undeployed at the same speed it was deployed. Speed buys position, and position is harder to surrender than to occupy. 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) · 18% 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 · 16% 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: 4 in the 3×3 ABC corner · 12 in the 9×9 A1..C3 square · 1752 in the 132×132 children square · 434 cross-zone
REALITY: 0 in the 3×3 ABC corner · 7 in the 9×9 A1..C3 square · 792 in the 132×132 children square · 167 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 160 claims @ θ 0.6563 · projection 803ms

SHAPE-MATCH σ = 4.95 — the real intent↔reality definer walk vs random-reality (actual 0.2344 vs random 0.079). >0 = the walk distinguishes real alignment from noise; aggregate over independent walks → the divergent series.
σ_drift · forming · separation is forming — the direction is real, the trust is not yet

Most out-of-lane row: C · Operations — 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 26% vs tolerance 25%.
607 green · 225 amber · 288 red · 234 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 77% > **What this is:** a technical read of a measurement instrument, and an honest scoping of what

reality 75% 2. **Carlos:** what must this join to on your side — client hierarchy, policy number, program s

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

intent 78% (no distinctive match)

reality 72% <h2>3 · Who is liable for what — the part that matters to Paul (5 min)</h2>

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

intent 83% **That is not a bug in the instrument. It is Rice's theorem showing up in a spreadsheet.** Whet

reality 78% <p><em>I am re-running one of them on a tuned configuration this week and I would rather hand y

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

intent 80% PROVENANCE VERIFIED (not invented): 0.90 separation / 0.30 paraphrase-invariance trace to the d

reality 78% We license the instrument built on the receipts, not the arithmetic that produces them.</li>

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

intent 80% (no distinctive match)

reality 78% `IN_LANE`. **That is not an AI failure. That is a negligent specification**, and the client had

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

intent 78% (no distinctive match)

reality 80% reproducible, and it is the only thing I am here to sell you. **I am publishing the null becaus

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

intent 78% (no distinctive match)

reality 80% or blast radius for that agent, **a number they already have.** It is the only severity proxy o

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

intent 81% (no distinctive match)

reality 80% (no distinctive match)

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

intent 75% (no distinctive match)

reality 75% 4. **Running the check is free, takes about 90 seconds, requires no installation, and sends us

C1,C1 Power flows where the grid lets it.

intent 75% (no distinctive match)

reality 75% current figure than a stale one.</em></p>

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

intent 81% **The one field that might break the chicken-and-egg:** `authority_at_risk` — the deployer's ow

reality 81% | **Can it tell work that belongs from work that doesn't?** | **0.90** separation, on a sealed

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

intent 75% (no distinctive match)

reality 77% installs, nothing leaves your network, no model runs anywhere in the check. **If it does not do

commit context — 5 file(s)

docs(marsh): agenda HTML+PDF — and HOLD the numbers until the tuned run

OPERATOR'S CHALLENGE, AND HE IS RIGHT: '0.90/0.30 were measured on ONE setup — just because we
measured it on a bad setup does not mean that is the floor.'

PROVENANCE VERIFIED (not invented): 0.90 separation / 0.30 paraphrase-invariance trace to the
documented sealed run ('the FAIR test RUN: full spec, full reef, real tolerances'). The AUC set
0.496/0.415/0.406 is the drift-vs-defects experiment that forced the prediction→provenance pivot.
Both are real repo numbers. But BOTH were measured on ONE configuration.

THE ASYMMETRY THAT DECIDES IT:
· send now, tuned number is BETTER → we have PERMANENTLY anchored an actuary to a worse figure.
  He prices against 0.30, and we can never un-ring it because WE published it.
· send now, tuned number is WORSE → catastrophic; he finds it in diligence after we wrote
  'should improve'.
· RUN FIRST → we either walk in with a stronger honest number, or we discover the 0.30 is
  structural BEFORE Marsh does.
There is no branch where running it first is wrong. The meeting is not scheduled yet — we have the
one thing we will not have later: time.

SO: the numbers section is REP

data/self-prompts.json
docs/outreach/accounts/marsh-agenda.html
docs/outreach/accounts/marsh-agenda.md
docs/outreach/accounts/marsh-agenda.pdf
docs/outreach/accounts/marsh-numbers-HELD.md

Produced on the chip by the running pipeline (scripts/pmu/pipeline.mjs → runPipeline → walk + xor), commit-scoped · on-chip per-walk 1230.01ns (this commit). Recompute: node scripts/pmu/commit-triptych.mjs --commit 2b7d155b8

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.