Method · v2.6

A deterministic pipeline,
not a vibe check.

15 agents run in orchestrated waves. Every claim carries a link back to the artefact that proved it. Every memo is versioned. Nothing is ever believed without a source.

15
agents
3
waves
100%
source-linked
5 stages · always running

The pipeline

01
Ingest
02
Extract
03
Stress test
04
Synthesise
05
Watch
01

Ingest

Decks, filings, founder comms, web signals — every artefact parsed into structured claims.

33x · ingest
parsePitch deck v7 · 42 slides → 214 claims
parseCH filing 2024-12 → 38 obligations
parseTranscript 41m → 96 quotable passages
02

Extract

Each claim is tagged to its source position — page, line, cell. No fact without provenance.

33x · extract
bindclaim#114 ← deck.pdf p.14 §revenue
bindclaim#221 ← cohort.xlsx S1:L14-62
bindclaim#307 ← interview.md L.118-142
03

Stress test

Deterministic scenarios — CAC shocks, runway extensions, supplier concentration, dilution.

33x · stress test
runscenarios/cac-2x → payback 22mo
runscenarios/runway-6m → cash-out Q4-26
runscenarios/concentration → 3 flags
04

Synthesise

Findings compose into a structured memo — TL;DR, scorecard, risks — versioned on every change.

33x · synthesise
writememo v2.6 · +312 words, -48
scorecapital 74 · execution 81 · conf. 84
flagsupplier concentration (high)
05

Watch

Diligence doesn't end at the memo. The swarm re-runs when new filings, press, or signals land.

33x · watch
watchAtlas raised Series B → delta fired
watchFoil press release → comp map Δ
watchGM variance 2.1σ → review queued
Anatomy of a claim

Every phrase, traceable.

Hover any claim in a memo and see the exact artefact, page, and line it came from. The pipeline refuses to output claims it can’t cite.

NovaPay · memo v2.6 · excerpt

Net revenue retention has held above 120% for the last four quarters, with cohort margins improving from 41% to 58% 3 sources as the self-serve funnel scales. Payback sits at 14 months on blended CAC, though concentration with one acquirer remains a flag 1 source in the Q3 financials.

Provenance · claim #114
XLSX
Cohort_2024_Q2.xlsx
sheet: Cohorts · rows 14–62
PDF
QC_Pitch_v7.pdf
p. 14 · §Cohorts
MD
founder-call.md
L. 118–142
Validator passed · 3/3 artefacts current
The swarm · 15 agents

Orchestration fans out, findings fan in.

The Planner decomposes a deal into parallel jobs. Specialists run concurrently. The Synthesiser composes. The Validator blocks anything unsourced.

Orchestration3
Planner
Sequences the wave & budgets tokens
Synthesiser
Composes the memo from claims
Validator
Blocks claims lacking provenance
Wave 1 · specialists7
Company
Team
Market
Product
Traction
Business model
Competitive
Wave 2 · deepen5
Financials
Stress test
Cap table
SEO / traffic
Radar
Versioned

Every memo has a diff.

Each new filing, each new quarter, each Planner re-run produces a new memo version. You can always see what changed and why.

v2.3v2.6NovaPay · 14s ago · triggered by CH filing 2024-12
§ TL;DR
Cohort margins improving; payback under 16 months.
+Cohort margins improving from 41% to 58%; payback at 14 months on blended CAC.
§ Risks
+Concentration with one acquirer remains a flag in the Q3 financials.
§ Scorecard
Capital 72 · Execution 76 · Confidence 79
+Capital 74 · Execution 81 · Confidence 84
+3 sources+2 claims−1 risk (resolved)+5 scorecard
Rubric

Scoring is deterministic.

No opaque model in the critical path. Every score maps to a rubric line with a fixed weight.

CriterionWeightScoreΔ v2.3
Capital efficiency0.1874+2
Execution velocity0.1681+5
Market pull0.1469
Unit economics0.1472-1
Team coverage0.1288+3
Survivability0.1063-4
Moat depth0.0858
Governance hygiene0.0891+1

Every claim source-linked

Hover any phrase in a memo and see the original page, line, or cell it came from.

Deterministic scoring

Scores are computed from a fixed rubric. No opaque model calls in the critical path.

Versioned memos

Every memo has a hash and a diff. You can see exactly what changed between v2.3 and v2.6.

Method — 33X