01 · Capture
Start with the exact fact.
Record the figure, publisher, source URL, and time. Material facts get a second check. A source saying something and that thing being true are kept separate; a partial check stays labeled partial.
Operating method · Public record
From a market fact to a checkable record.
The system is designed to separate four jobs that are easy to blur: observing what happened, judging what it means, testing what might happen next, and checking the answer later. Each job has its own public surface and its own limits.
Layer 01 · Observe
What happened?
Observer files verified intraday facts without trying to turn them into a signal.
Layer 02 · Judge
What matters?
The daily brief weighs the facts, the market response, the uncertainty, and the counter-case.
Layer 03 · Test
What could happen?
Material probabilistic claims enter a dated ledger with odds, a deadline, and a grading rule.
The working loop
Four steps, with a receipt at each one.
The system does not become trustworthy because an AI sounds careful. It becomes inspectable because every stage leaves an artifact a reader can compare with the next one.
02 · Context
Ask whether the market confirms it.
Compare the story with broad credit stress, volatility, rates, breadth, and the relevant asset. The brief names both what is moving and what is not breaking, so one loud corner cannot masquerade as system-wide stress.
03 · Commit
Publish the judgment before the answer.
The daily brief states the bottom line, uncertainty, counterargument, and what would change the read. A genuinely probabilistic claim also receives odds, an objective success rule, and a deadline. Analysis that is not a forecast is not forced into one.
04 · Revisit
Grade the record, not the prose.
After the deadline, named evidence resolves the call and produces a calibration score. Misses remain visible. Corrections are appended rather than silently replacing the earlier record, and related forecasts are not pretended to be independent trials.
Channel boundaries
Each page is allowed to say something different.
These separations prevent an intraday fact from quietly becoming a recommendation, or a mock portfolio number from being mistaken for a live trade.
Observer
Verified facts
Zero to three material observations per scheduled check, each with a figure, source, time, and corroboration.
Cannot contain: forecasts, probabilities, Calm/Watch/Stress signals, or recommendations.
Daily brief
Interpretation and uncertainty
A plain-English market read, explicit source quality, the strongest competing explanation, and what would make the conclusion weaker.
Cannot contain: invented figures, personal allocation advice, or certainty the evidence does not support.
Prediction record
Falsifiable claims
Confidence, deadline, exact grading rule, source brief, resolution, and per-call score remain together.
Cannot contain: hindsight rules or quiet deletion of a miss.
Practice book
Computed mock results
Five fixed rules begin from the same mock $100K and produce their value paths and mock decisions from code.
Cannot contain: real orders, broker activity, real money, or a claim that the current leader has proven skill.
Cadence
Different clocks for different questions.
More output is not automatically more information. Each cadence exists because the underlying question changes at that speed.
Intraday · Beta
Observer checks
Scheduled during the US session. A completed check may honestly file nothing; a missed check must not look quiet.
Morning
Daily brief
Overnight evidence, the current risk read, the counter-case, and the next things worth watching.
After close
Mock dashboard
Computed strategy values, recent mock moves, and a comparison against the common starting point.
Review cycle
Grades and lessons
Due forecasts are resolved and the accumulated calibration record is reviewed without rewriting history.
Integrity rules
The guardrails are part of the product.
They matter most when the story is exciting, the data are incomplete, or a previous call looks embarrassing.
Portfolio figures come from code. Interpretation does not get to invent a cleaner number.
Factual figures name their sources. Thin or conflicting evidence stays visibly qualified.
Uncertainty is written before resolution. Odds and rules cannot be improved after the answer.
Quiet and broken are different states. Missing data never masquerade as “nothing happened.”
Mock means mock. No broker, no live orders, no real capital, and nothing to copy.
Research is not advice. The system explains risks and trade-offs, not what a person should buy or sell.
The idea behind the method
Why gray rhinos, not crash calls?
The operating method above explains the machinery. The welcome note explains the premise: unknowable shocks cannot be forecast on schedule, but visible, high-impact risks can be watched without pretending to know the day they will matter.