All use cases
Forensic verdict engineLive in portalsign-in

Forensic Checklist

Ticker in, verdict out. Scores ~150 deterministic forensic rules plus ~44 LLM judgments against a company's filings; produces a composite score, conviction, and cited decision memo.

The Forensic Checklist is a ticker-in, verdict-out engine. Give it any US-listed ticker and it fetches the filing trail, scores ~150 deterministic rules (accounting quality, tax abnormalities, working capital games, insider signaling, disclosure red flags), runs ~44 LLM qualitative judgments against the narrative sections, and synthesizes a composite score with a conviction level and a decision memo. Every fired rule ships with cited evidence — no fabricated signals, no unsupported claims.

What makes it different: 209-rule library. Every fired rule is cited with evidence. Composite score is a scalar from −5 (bearish) to +5 (bullish), with per-family severity tiers.

What it does

Capabilities

209-rule library

Deterministic rules cover forensic categories: revenue recognition oddities, expense timing, tax abnormalities, cash-flow-vs-earnings divergences, insider transactions, disclosure changes. Each rule is a testable predicate over XBRL and text.

Composite score with constellation warnings

Score is a scalar from −5 (bearish) to +5 (bullish). Direction and conviction land as separate signals. When many rules cluster in the same family, a systemic-warning banner surfaces the pattern.

Per-family severity tiers

Rule firings roll up into family composites (forensic-tax, working-capital, insider-signal, etc.) with a red / yellow / none severity tier. Renders as a colored matrix so a committee can scan risk at a glance.

Time-bucketed forward view

Watch items split into near / mid / long buckets, each tied to a specific rule that fired. Turns the verdict into a monitoring plan, not just a snapshot.

How it works

Input → output

Input
MSFT · Microsoft Corp
  1. 1Fetch full filing trail from EDGAR (10-K, 10-Q, 8-K) — XBRL-parsed
  2. 2Score ~150 deterministic rules against parsed financials + text
  3. 3Run ~44 LLM qualitative judgments (DeepSeek) against narrative sections
  4. 4Synthesize composite score, direction, conviction — with constellation-warning trigger if clustered
  5. 5Produce cited decision memo (Bull / Bear / What-to-watch) + fired-rules list with evidence per rule
Output

Composite score card + Bull/Bear/What-to-watch memo + family-severity matrix + near/mid/long watch buckets + fired-rules list with cited evidence per rule.

Output preview
portal · live
(your ticker)
Strategy brief in active build
as of 2026-06-25
Coming soon

We're shipping the consultant-grade strategy brief format for select pilot tickers. Each brief will combine the archetype classifier, FKB citations, catalyst assessor, and quarterly-tone diff into one structured document.

Brief structure (preview)
  • Archetype classification (5 sentences + supporting financials)
  • Bull / bear / base case thesis with FKB citations
  • Catalyst assessor: next 12-month moves with expected magnitude
  • Quarterly tone diff: management language drift
Why this isn't a wrapper

Why this isn't 'AI reads a 10-K'

Most AI forensic tools generate plausible-sounding paragraphs. This one grades them against 200+ testable rules first.

  • Rules are the primary signal. LLM judgments are additive; they can't override a deterministic firing.
  • Every fired rule carries evidence — a quote from the filing or a specific XBRL value.
  • Constellation warnings surface clustered patterns (many rules in the same family) — the way real forensic accounts think.
  • Per-family severity tiers let a committee triage risk without reading the whole memo — the matrix tells you where to look.

Ready to see it?

Sign in to the portal and pull a dossier on any ticker in the universe.