feat(retrieval): a review pass judges whether injected lines related — menus_to_review, judge_menu and a judged readout (#4772)

An open rate cannot say whether a menu line related: every line carries its
matched passage (#4364), so "not opened" covers unrelated, enough as shown,
and already in context. #4772 "Injected notes are never judged".

- retrieval_judgments (0115): a reviewer verdict per line of a logged call,
  on_point / adjacent / unrelated, with its reason, rank, budget side and
  whether the agent opened it within the hour.
- menus_to_review re-runs a random sample of unjudged auto_inject calls with
  the arm's own parameters, past its budget, passage on every line.
  judge_menu records verdicts, re-deriving rank from a fresh re-run.
- retrieval_telemetry gains a judged block (by rank, within/beyond budget,
  on_point_unopened). surfaced_never_pulled stops blaming titles.
- missed-retrieval guidance names the review before a budget move.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

alembic/versions/0115_retrieval_judgments.py

@@ -0,0 +1,56 @@
+"""retrieval_judgments — a reviewer's verdict on each line of a logged menu
+(#4772)
+
+Revision ID: 0115
+Revises: 0114
+Create Date: 2026-10-03
+
+Whether an injected line RELATED, which the usage tables cannot say: since
+every line carries its matched passage, "not opened" no longer implies "not
+relevant". FK-free like the telemetry tables beside it; no CHECK on `verdict`,
+which the service validates.
+"""
+import sqlalchemy as sa
+from alembic import op
+
+revision = "0115"
+down_revision = "0114"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "retrieval_judgments",
+        sa.Column("id", sa.BigInteger(), primary_key=True),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), nullable=False,
+            server_default=sa.text("now()"),
+        ),
+        sa.Column("user_id", sa.BigInteger(), nullable=True),
+        sa.Column("retrieval_log_id", sa.BigInteger(), nullable=False),
+        sa.Column("source", sa.Text(), nullable=False),
+        sa.Column("query", sa.Text(), nullable=True),
+        sa.Column("record_id", sa.BigInteger(), nullable=False),
+        sa.Column("rank", sa.Integer(), nullable=False),
+        sa.Column("score", sa.Float(), nullable=True),
+        sa.Column("within_budget", sa.Boolean(), nullable=False),
+        sa.Column("verdict", sa.Text(), nullable=False),
+        sa.Column("reason", sa.Text(), nullable=False),
+        sa.Column("opened_after", sa.Boolean(), nullable=True),
+        sa.UniqueConstraint(
+            "retrieval_log_id", "record_id", "user_id",
+            name="uq_retrieval_judgment_line",
+        ),
+    )
+    op.create_index(
+        "ix_retrieval_judgments_source_created", "retrieval_judgments",
+        ["source", "created_at"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_index(
+        "ix_retrieval_judgments_source_created", table_name="retrieval_judgments",
+    )
+    op.drop_table("retrieval_judgments")