Every trade gets a certificate. Most of them fail.

The problem

Entries are a solved problem. Exits are not.

A dozen tools help you find a trade. Nothing helps you leave one. Every trader knows the specific feeling of holding past their own stop and inventing the reason on the way down.

The usual diagnosis is discipline, and the usual fix is willpower. Both are wrong. The exit goes badly because nothing was ever named at the entry to hold it against — so at the moment of the decision there is no plan to deviate from, only a price and a feeling.

You cannot fail a test you never wrote. That is the whole thesis.

Running today

Not a prototype.

This runs on a real, live personal book — not a demo account and not backtested data. Every number it produces is measured off that book, including the unflattering ones.

The book

Around a thousand closed trades — stocks, options, crypto spot and perps, in one ledger.

Ingestion

Live brokerage sync, CSV import for anything else, and on-chain wallets read directly.

Grading

Each closed trade is tested against the exit rules you declared, and returns ON PLAN, ON PLAN · FLAGGED, or OFF PLAN.

Custody

A signed certificate is final. Its hash is recorded on Solana — devnet today — so a later edit cannot pass as the original.

The part that is not obvious

The machine states facts. You sign the verdict.

The obvious build is to hand your trades to a model and let it score you. That version is worthless, and it is worthless for a specific reason: a model asked to judge will always produce a judgement, and you have no way to tell the real ones from the fluent ones.

So the grader never owns the verdict. It reads your own exit prose and the evidence — did the level actually print, did you dismiss the alert, when was the plan written — and it proposes. You ratify it or you overrule it. Your grade beats the machine's, permanently, and the disagreement is kept rather than resolved.

One consequence is worth stating out loud, because it is what makes the grade mean anything: rules only count forward, from the day you declared them. A rulebook written after the exit is not a commitment, it is a retrofit, and the grader refuses it.

Check the work

Read the method instead of taking my word for it.

Two guides, written for someone who has never opened the app. They document the actual shipped grading, not an intention.