anchors judge

Record an AI’s verdict for a judgment gate.

Records the verdict of an AI JUDGMENT gate over a target — the same
double output as a deterministic gate (stamp on the map + issue).

  anchors judge --pending                      lists the targets awaiting judgment
  anchors judge <alvo> --gate <g> --verdict pass|fail|waived --reason "..."

The flow: 'anchors check' marks the targets of a 'measures: judgment' gate as
pending and queues them. The AI (worker) reads the gate's guide, confronts the target, and
reports here.

IMPORTANT — do not waste what you have already read. To judge, you read the whole guide
and the whole target; so --reason is NOT a verdict sentence, it is the FULL REPORT. The
issue body IS your --reason, verbatim (multi-line markdown accepted). On a 'fail',
list EVERY non-conformity with: what is wrong, WHERE (file:line), WHY
(which rule of the guide), and HOW to fix it. That way nobody has to reprocess the target
later just to find out what to fix. E.g.:

  anchors judge apps/.../AlertsScreen.tsx --gate atomic-design --verdict fail \
    --reason "$(cat <<'EOF'
  ## Report — atomic-design @ AlertsScreen
  ### 1. Inline list item should be a molecule  (l.151-166)
  - **what:** the item's JSX (View + icon + texts) is assembled inline in the .map
  - **rule:** SCREEN_GUIDE §Pages — screens inject components, they do not assemble blocks
  - **how:** extract to components/molecules/AlertRow.tsx and use <AlertRow item=.../>
  ### 2. ...
  EOF
  )"

'fail' opens the issue with that report; 'pass' resolves the issue if there was one (and the
--reason, if any, becomes a recorded observation).

'waived' is for when THE TARGET OF THE QUESTION DOES NOT EXIST. A spec that declares
'@TBD: code' states that the code has not been written yet — the normal flow, since the spec
is the anchor and is born first. Then the question "does the excerpt REALIZE what the rule
describes?" has no excerpt, and the other two verdicts lie: 'pass' states that the code realizes
the rule (and the stamp stays on the map looking like a real verification), 'fail' fails work nobody
got wrong. The --reason is mandatory and names the absence: which piece is missing, and where
it is declared.

Usage

anchors judge <target> [flags]

Flags

Flag	Default	What it does
--gate		the judgment gate being evaluated
--map		path to the map
--patch		file with the diff that FIXES the finding — opens an applicable suggestion (anchors suggest)
--pending		lists the targets awaiting judgment
--reason		the full REPORT (multi-line markdown) — becomes the issue body; mandatory on fail
--record-issues		mode: manual — write the issue (on fail) and close it (on pass); by default none is written
--root	.	project root
--verdict		pass | fail

Judgment and review — family

Commands of the same family — what tells each apart:

Command	What it does
anchors judge	Record an AI’s verdict for a judgment gate
anchors review	Record a review of a target, or list what is to review
anchors suggest	List, apply and decide the proposed fixes