MarketDeck docs

review

AI position-review history — manual or auto-loop reruns of the prompt against an open position.

A review is an AI run against a position that's already open — "what would you do now?" The runtime parses the response for one of HOLD, ADJUST_SL, ADJUST_TP, CLOSE_PARTIAL, CLOSE_FULL and either applies the action (when configured) or files it for human confirmation.

Usage#

marketdeck review <subcommand> [args]

Subcommands#

SubcommandWhat it does
list <position-id> [--limit N]Newest-first reviews for one position.

Examples#

marketdeck review list pos:dev:BTC:long:1714123456000 --limit 5 --format text

Where reviews live#

The position_reviews table holds one row per review:

  • position_id — FK to positions(id), cascades on delete.
  • ts, provider, model, output.
  • action_keyword and action_json — the parsed action.
  • applied — whether the action was applied automatically.
  • source — 'manual' for UI-driven reviews, 'auto' for the scheduled re-review loop.
  • reject_reason — populated when the parser rejected the AI's output (no **Action:** marker, unknown keyword, missing payload). Lets the UI show why an auto-review didn't translate into an action.

Generating a review#

Reviews are generated through the GNOME plugin's per-position review dialog or by the auto-review loop running in the shell. There is no standalone marketdeck review run verb — the action- applying half is wallet-aware and lives in the engine path that already knows about paper vs. real.