WHALE CACHE V1 — Leviathan Capture Matrix

P_c = (S × A) / (D² × N)

S — Signal

Objective quality of the payload. Only thing we fully control. Rick's law: can't P_c past weak S.

A — Alignment

Hitting the target's active window. A high-signal post during a SpaceX design review = 0.

D — Distance

Degrees of separation. Squared in denominator. Reduces via Remora Strategy (T2/T3 targeting).

N — Noise

How many others are shouting. Reduces via temporal precision. Never zero.

Source: ATMOS Rick · 2026-04-09 · Fleet Standard: ABR-809

ESTIMATED P_c — CURRENT FLEET STATE

—

adjust sliders

Live P_c Calculator

S — Signal Quality 0.30

A — Temporal Alignment 0.50

D — Degrees of Sep. 3.0

N — Ambient Noise 5.0

What moves the needle most:
Raise S (close boss fights, publish results) → biggest lever.
Reduce D (post in Lean4 Zulip / ARC-AGI forum first) → D=1 from 3 = 9× improvement.
Optimize A (don't post Monday 9AM EST) → 2-3× improvement.
N is structural — accept it, route around it.

HONESTCurrent S ≈ 0.30. Not enough for S-Tier direct. Path: T2/T3 first → build track record → S rises naturally.

NOTERick's law: the formula reveals the problem. You cannot P_c your way past weak S.

LEVER 1Close IRF-ARC-DSL-002 (translation) → S jumps to ~0.5. Close AxiomElimination boss fight → S jumps to ~0.7.

Signal Inventory — Honest S-Scores (0.0 → 1.0)

What Raises S (Ordered by Impact)

Action	S Δ	Est. Time	Status
Close AxiomElimination (29 sorrys → 0)	+0.35	weeks	OPEN
ARC-AGI-2 first pass results (any score)	+0.20	days	READY TO START
Close IRF-ARC-DSL-002 (translation)	+0.15	hours	OPEN
τ_γ₁ preprint (honest hypothesis paper)	+0.15	days	NOT STARTED
Post to Lean4 Zulip (open IRF question)	+0.10	30 min	READY NOW
Close IRF-ARC-DSL-003 (object enumeration)	+0.10	days	NOT STARTED
DeBrangesInnerProduct boss fight (4 sorrys)	+0.10	hours	OPEN

PHASE 1

OSINT TELEMETRY — The Radar

Map target community activity patterns. Not individual surveillance — community-level signal windows.

What we track: Lean4 Zulip post/reply times · ARC-AGI Discord peaks · arXiv submission windows · GitHub merge times
What we don't track: Private individuals · Personal communications · Anything requiring ToS violation

Temporal windows (known): Academic Twitter peaks Tue/Wed afternoon. Lean4 Zulip peaks Mon/Thu morning UTC. ARC forum peaks after paper drops.

PHASE 2

NETWORK TOPOLOGY — The Remora Strategy

Never aim at T5 first. Map who T5 actually reads. Target T2/T3 with real content.

T2 targets (D=2): Lean4/Mathlib maintainers · ARC-AGI community leads (Ryan Greenblatt, Jack Cole) · Physics arXiv readers with active Twitter
T3 targets (D=3): Math Twitter / proof assistants community · LLM benchmark trackers · Engineering newsletter writers

The move: Post a real open problem to Lean4 Zulip. If three T2 people find it interesting, D to T5 drops from 5 to 2. D² goes from 25 to 4. P_c multiplies 6×.

PHASE 3

DYNAMIC HONEYPOT — The Engagement Pages

Bespoke landing pages that hydrate based on referrer context. Not generic marketing.

?ref=lean4 → ARCGrid.lean source + proof stats + open IRF question. No marketing. Ask: "We're stuck on translation invariance. Here's the type sig."
?ref=arc-agi → Grid visualizer + current results + thermodynamic filter explanation
?ref=physics → τ_γ₁ calculation + honest hypothesis framing + Romero 2014 citation
?ref=engineer → Real stack: AKS + Go + Lean4 + Redis. Honest numbers. No claims.

The trap (Rick's insight): Leave one open problem visible. S-Tier targets cannot walk past an unsolved puzzle. Don't ask for a meeting. Ask the question.

PHASE 4

MATHEMATICAL RESONANCE — The P_c Dashboard

This page. Track S over time as boss fights close. Optimize A by watching community windows.

When to fire: S ≥ 0.5 AND boss fight closed within 48h AND community window open → POST
When to hold: S < 0.4 OR major news event eating N → wait

The honest rule: If we have to oversell to make it interesting, the payload isn't ready. Return to signal.

Target Tier Map (PIP-PII-001: No private individuals named)

Tier	Community	D	Best Signal	Window	P_c Multiplier
T2	Lean4 / Mathlib maintainers	2	ARCGrid.lean + open IRF	Mon/Thu morning UTC	25× vs D=5
T2	ARC-AGI community (forum, leaderboard)	2	Formal ARC type system + results	After paper drops	25×
T2	Physics preprint readers (arXiv)	2	τ_γ₁ honest paper	Tue/Wed post	25×
T3	Math Twitter / proof assistants	3	Interesting open problem + clean code	Tue/Wed afternoon	2.7×
T3	LLM benchmark trackers	3	ARC-AGI-2 result ≥ 65%	With paper	2.7×
T3	Engineering newsletters	3	Fleet architecture writeup	Monday drops	2.7×
T5+	S-Tier individuals	5+	Needs S ≥ 0.8 + T2/T3 amplification	After T2/T3 land	currently 0.04×

First Move (Ready Now)

Post to Lean4 Zulip #general or #mathlib4:

"We're formalizing ARC-AGI grid topology in Lean4. We have 4-connectivity and path reversal proved (ARCGrid.lean, ~200 lines). Stuck on translation invariance — translate {H W H' W'} (g: ARC_Grid H W) (Δr Δc: ℤ): ARC_Grid H' W' — the domain mismatch when Δ puts cells out of bounds. Is the right approach a partial function, a dependent type on valid offsets, or something else? Happy to share the file."

This is a real question. It does not oversell anything. It invites engagement from people who know. If two Lean4 people answer, D drops to 1. P_c for T2 amplification goes to max.

Bespoke Landing Page Router (?ref= parameter)

ref=	Content Shown	Open Problem	CTA
lean4	ARCGrid.lean source, proof stats, IRF list	"Translation invariance — how do you handle domain mismatch?"	Zulip link
arc-agi	Grid viz, thermodynamic filter, honest score	"We have connectivity. How do you handle occlusion ordering?"	Forum link
physics	τ_γ₁ calc, Romero 2014 ref, honest hypothesis	"Is 540/T the right simplification or are we missing a constant?"	arXiv
engineer	Real stack: AKS/Go/Lean4/Redis, honest numbers	"We have 183 sorry stubs. Which 5 would you close first?"	GitHub
default	Standard pemos.ca overview	—	—

What Makes a Good Honeypot

✅ Real open problem — not manufactured
✅ Enough context to engage immediately
✅ No marketing language in the first screen
✅ Terminal aesthetic for engineering targets
✅ One specific question, not "contact us"
❌ Any claim that isn't backed by a file we can link
❌ "Universe verified" / "coherence boundary" / inflated language
❌ Ask for anything before establishing S

Admiral Gate (Every Page)

Every honeypot page passes VERIFIER + OVERSEER before it goes live.

⬜ Is every claim backed by a real file?
⬜ Is inflated language absent?
⬜ Is the open problem genuine?
⬜ Does Rick's Law apply? (Would we describe this modestly?)
⬜ Admiral reviewed and approved?

Admiral Gate — Every Outreach Piece Before It Fires

Outreach Log

What We Are Not Building

❌ Automated spam / follow bots
❌ Scrapers that violate ToS
❌ OSINT on private individuals (PIP-PII-001)
❌ Manufactured signal — if S isn't there, we don't fire
❌ Anything that manufactures hype without substance behind it

This is amplification of genuine work. If the S-score isn't there, we return to signal. The fleet exists because the work is real. The outreach exists to let the right people find it.