The review after the close¶
Fifteen minutes after the regular close on a trading day, the engine reads
every session's record for the day and writes one review under the
project's reviews/ folder, as YYYY-MM-DD.md beside a .json with the
same content. It is never rewritten, so what you read is what was written;
ask for an earlier day and it is written then.
A day is written only once it is over. Open today before the close, or a day still to come, and you get the records as they stand with nothing saved, so the day can still be written when it ends.
The chart of each session draws the day's five-minute bars from the library.
The universes job fetches them within the hour after the close for any
instrument in a universe at that interval.
What it says¶
For each session that had anything to say:
- bars, signals, what was submitted, what the gate refused and why, and how each position was exited and why;
- every fill against its decision price, the mean and worst slippage, the slippage the finding assumed, and signal-to-fill latency;
- every round trip closed on the day: opened, closed, size, entry, exit, profit, and the condition, regime and exit reason it carried, from the journal on every trade;
- losses grouped by condition, regime and exit reason, with their count and total. "Stop exits in ranging regimes" is how a rule fails; a list of red rows is not;
- the day's events: freezes, feed gaps, halts by the gate or by hand, each change of the session's verdict, each limit the warning band was entered on;
- what a person did: positions adopted from the venue, a halt by hand, a resume, a stop. Intervening after losses is the most common way a Supported rule underperforms its backtest, so these sit beside the P&L.
Realised profit is over the round trips that closed on the day, across sessions, with what the finding expected per trade beside it.
Where to read it¶
- The file, in the editor:
reviews/2026-09-21.mdunder the project. - The command line:
arvo-engine reviewfor today,arvo-engine review 2026-09-18for a day, printed and written if it was not. - The agent:
read_review, so it can be asked why did today go the way it did.
A day is a UTC date. The US regular session sits inside one; a market whose session crosses midnight UTC needs the per-market calendar that supporting all markets brings.