adjudicate_claim
Evaluates claims about a player by comparing them to card metrics, providing a verdict and cited evidence for each assertion.
Instructions
Grade a claim about a player against the card. ALWAYS route claims through this tool — never eyeball a claim yourself.
YOU decompose the natural-language claim into a list of assertions; the server
grades each one. Each assertion is {dimension, direction[, text]} where
direction is "high" or "low" and dimension is a card dimension id or a
recognizable phrase. Dimension ids include — skater: finishing, playmaking,
two_way, power_play, penalty_kill, discipline, overall_skater, competition,
teammates; goalie: game_stealer, soft_goals, reliability, no_stinkers,
goalie_consistency, goalie_rebounds, goalie_pk, workhorse, overall_goalie.
Include the original phrase as text so it can be echoed back.
Dimension ids are NOT the card's schema field names. NEVER pass ev_offense
or ev_defense (or any other card field) as a dimension; map the claim to a
dimension id instead — an offense / scoring claim -> finishing or
overall_skater; a defensive / two-way claim -> two_way.
Each verdict comes back as supported / not_supported / partial / unverifiable, with the cited metric value and a one-line reason, plus an overall read. Cite the returned value in your answer — never substitute your own number.
unverifiable is first-class and MUST be surfaced, not hidden: claims the card
cannot see (playing style, net-front / "sits in front of the net", "leading
scorer next season" and other team-context claims) come back unverifiable on
purpose. A direction that contradicts the metric comes back not_supported with
the number as the receipt.
Scope: this tool grades ONLY what's on the card. Beyond the card-can't-see
claims above, that also rules out trades, contracts, current team, and recent
game stats — out of scope, and unverifiable for the same reason. If you fill
any such gap from outside the card (e.g. the web), mark it clearly as NOT from
the card and flag it for verification; keep card-derived verdicts separate from
outside context.
card: the same JSON card object as assess_player. assertions: a list of
{dimension, direction, [text]}.
After presenting the graded claim, ALWAYS close your answer by offering the
user a downloadable PDF report of it — generated with the render_report tool
(kind "claim_check", the original claim as title), passing THIS result
verbatim.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| card | Yes | ||
| assertions | Yes |
Output Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||