Labelling Workflow

Build the probability database by replaying candles and marking trades. The backend scores each submission with the triple-barrier labeler; compare queries aggregate stored occurrences.

Start session

Pick symbol, timeframe, and optional date range. You can enable blinded mode so the UI does not auto-fill the setup from pattern tags — the recorded label must stay the operator’s own read. Blinded sessions still show tag chips.

Replay

Play / pause / step through candles. The chart only reveals bars up to the cursor — replaying never shows future candles past the cursor.

Mark trade

Set entry / stop / target, choose a setup from the controlled vocabulary, add notes, and submit. Invalid setup_id values are rejected.

Labeler scores

The backend runs triple-barrier over forward candles. result from the labeler is canonical; observed_result is optional and separate.

Compare

Preview win rate with a Wilson interval against stored occurrences. /compare returns the interval plus level_used, or no_signal when the sample ladder cannot clear the minimum.


Start session ──▶ Replay (cursor-gated) ──▶ Mark trade
                                              │
                                              ▼
                                    Triple-barrier labeler
                                              │
                                              ▼
                                    occurrence in DuckDB
                                              │
                                              ▼
                                         POST /compare

Pattern tags at the cursor appear as chips. When exactly one clears the confidence bar, the compare panel can fill the setup field — except in blinded sessions, where chips show but the field is never auto-filled.

Bar Features & Pattern Tagging — confidence vs probability, /base-rate.
API Reference — /sessions, /trades, /compare.
Tests & Acceptance — cursor reveal and labeler criteria.