Hypothesis in,
proof out.
What CRUX actually is: a local, paper-only decision cockpit I run myself. AI can surface ideas — they land as unverified hypotheses. A claim only enters the journal after it clears a Verify Gate against primary filings (e.g. SEC EDGAR). Day-to-day is checkpoint-driven (grade when a pre-registered event is due); Cascade runs research on demand. Marks are honest. Calibration stays empty until something actually closes. Nothing places a real broker order. Ever.
An AI speaks with total confidence.
Confident isn’t true.
The gate’s job is simple: don’t believe the storyteller — check the filing. Below is an illustrative demo of that rule (anonymized cases, not live quotes, not tips). In the real cockpit the same pattern applies: pull primary source next to the claim, stamp a verdict on the claim’s truth, only then allow a journal line.
—
Illustrative mechanism demos · not investment advice · labels are demo cases, not recommendations
It remembers how lies look.
Not a stock-picking model. A standing catalog of fabrication signatures seen in the wild (and in adversarial self-tests) — patterns that only ever accelerate suspicion, never trust. Fast path to “look harder or kill it.” No fast path to “believe it.” Never a bypass of the full gate.
Four steps. No shortcuts.
This is the public shape of the process — not a full ops manual. Day-to-day, the system is checkpoint-driven (grade when a pre-registered event is due), not a promise of 24/7 watching.
A claim arrives
An AI (or other research pass) surfaces a confident thesis. It lands as a hypothesis — believed by no one yet.
Check the source
Load-bearing claims are checked against primary filings / checkable data (e.g. SEC EDGAR) — beside the claim, never instead of it.
Render a verdict
Confirmed, pending, or killed — on the claim’s truth, not on whether the story is exciting.
Journal · mark · grade
Survivors get a journal line with a kill condition. Paper positions mark to market honestly. Checkpoints grade against rules written before the outcome. Calibration stays — until something actually closes.
Even a real signal dies
when the rest of the case fails.
Catching fabrications is half of it. The harder half is refusing a convenient yes — a claim that is genuinely true at the source, but still fails a sober risk read, category fit, or pre-registered kill rule.
In a genre full of fake backtests,
showing nothing is the flex.
Calibration is empty until a real decision has a real closed outcome. This page does not invent a track record. As of 2026-08-09 it publishes the paper book and marks it live — but never as proof: the mark is drawn inside the band where a result is indistinguishable from luck, realized P&L is $0.00, and graded outcomes are 0. The mechanism and the book are public. The playbook is not — no screen, no threshold, no ranking, no selection criteria appears on this page or in anything it serves.
Scaled ×10 from a real $1,000 paper book — true value $1,050.18. The percentage is identical at either size. A bigger number here is a bigger font, not a better result.
connecting to the quote feed…
- loading…
Holdings are facts, so they are published. The index sleeve is where uninvested cash sits, so the comparison measures selection rather than the luck of holding cash.
What this is
- A personal, local decision-support system for market research
- A Verify Gate: AI ideas start as hypotheses until primary sources clear them
- A paper-only journal with integrity checks and pre-registered kill conditions
- Honest marks and checkpoint grading — calibration stays “—” until outcomes exist
What this is not
- Not financial advice, signals, or a stock-picking product for strangers
- Not a performance claim — the marks are paper, unrealized, and shown against the noise band that says they have not cleared luck yet
- Not a dump of the private playbook or the research dossiers — holdings are facts and are published; how they were chosen is not
- Not a path that can place real broker orders — that rail does not exist
Signals decay.
Proof compounds.
Not financial advice · personal paper cockpit · this page is mechanism only
No real-money path · no public track record · no playbook · no tickers as tips