docs(6205): shape_ledger comments state the reason, not the instance that taught it

Record ids, snippet numbers, dated measurements and the proposer's version
history are removed; each block now states the design reason in the present
tense. Calibration figures that justify a floor stay, without the instance
they were measured on. One error string drops a note id. The density
threshold comment moves above the constants it describes.

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

src/scribe/services/shape_ledger.py

@@ -1,4 +1,4 @@
-"""The shape ledger's write side — sync and mechanical marking (#2788).
+"""The shape ledger's write side — sync and mechanical marking.
 
 The coverage walk (services/coverage.py) is the only feed that sees every
 shape, so it is the ledger's sync point: each refresh upserts one repo's
@@ -68,7 +68,7 @@ def location_covers(loc_path: str, loc_symbol: str, path: str, name: str) -> boo
 
 
 # The rows the machine may still speak about: nobody's judgment stands on
-# them. `scoped` (#2869) is the sync's own by-construction stamp — the
+# them. `scoped` is the sync's own by-construction stamp — the
 # proposer, derive grouping, divergence, hook evidence and sweeps treat it
 # like the todo; only the human todo (`unclassified`) excludes it.
 _MECHANICAL_TODO = ("unclassified", "scoped")
@@ -88,8 +88,8 @@ async def sync_repo_shapes(
     """Upsert one repo's extracted shapes into the ledger.
 
     ``shapes`` are (path, kind, name) triples, or the richer ArchiveShape
-    records (#2792) whose 4th/5th fields — signature, body_sha — refresh the
-    row's content fingerprint, and whose 7th (#2869) says the shape is a
+    records whose 4th/5th fields — signature, body_sha — refresh the
+    row's content fingerprint, and whose 7th says the shape is a
     one-off by construction: such rows are stamped `scoped` (mechanical)
     while unjudged, and un-stamped if a later tree makes them reachable. ``seen_marker`` is the commit the archive was
     read at when the forge can say, else the ref name — provenance sugar;
@@ -147,7 +147,7 @@ async def sync_repo_shapes(
                     row.classified_sha = body_sha
                 elif judged and row.classified_sha != body_sha and row.recheck_at is None:
                     # The body moved under a standing judgment: the judgment
-                    # stands, but asks to be confirmed again (#2793).
+                    # stands, but asks to be confirmed again.
                     row.recheck_at = now
                     session.add(_event(row, "drifted", now, commit=seen_marker))
                 row.body_sha = body_sha
@@ -214,7 +214,7 @@ async def _judge(
 async def record_uses(
     session, row: CodeShape, snippet_ids, *, basis: str, evidence: str | None = None,
 ) -> int:
-    """Upsert consumption edges shape → snippet (#2870). A judgment-grade
+    """Upsert consumption edges shape → snippet. A judgment-grade
     basis (agent/audit/import) overwrites a mechanical one (reference/hook)
     on the same edge; mechanical never overwrites a judgment. Returns the
     number of edges written or refreshed. The row must be persisted (flushed)
@@ -264,7 +264,7 @@ async def uses_of(shape_ids) -> dict[int, list[CodeShapeUse]]:
     return out
 
 
-# --- the CSS consumer map (milestone 302) ------------------------------------
+# --- the CSS consumer map ---------------------------------------------------
 
 
 def resolve_consumers(
@@ -275,13 +275,13 @@ def resolve_consumers(
     consumes. ``css_rows`` are (id, path, symbol) of the repo's live css rows;
     ``references`` is scan_archive's path → class token → count.
 
-    Resolution (note 2917): a class named in file F resolves to F's OWN row
+    Resolution: a class named in file F resolves to F's OWN row
     of that name when F defines it (a scoped rule is consumed by its own
     template); otherwise to every other file's row of that name — a shared
     sheet, or, when several files define it, all of them: the map says
     "ambiguous" by fanning out rather than guessing one.
 
-    A token ending in ``PREFIX_MARK`` is a PREFIX reference (#2970) — the
+    A token ending in ``PREFIX_MARK`` is a PREFIX reference — the
     static head of a name the template concatenates, `status-*` from
     `` `status-${s}` ``. It stands for every row whose symbol starts with
     that head, each resolved by the same own-file-else-fan-out rule. The
@@ -410,7 +410,7 @@ async def mark_canonicals(
     usable = [(nid, p, s) for nid, p, s in recorded if (s or "").strip()]
     # A location naming a FILE and no symbol is a whole-file record: it makes
     # no claim about a definition inside (`location_covers`), but it is
-    # exactly the claim a `file` row asks about (milestone 439).
+    # exactly the claim a `file` row asks about.
     whole_files = {(p or "").strip(): nid for nid, p, s in recorded
                    if (p or "").strip() and not (s or "").strip()}
     now = datetime.now(timezone.utc)
@@ -477,7 +477,7 @@ async def live_rows(project_id: int) -> list[CodeShape]:
         )
 
 
-# --- classification (#2789): the judgment write path --------------------------
+# --- classification: the judgment write path --------------------------------
 
 # Statuses an explicit classification may set. All five: setting a row back to
 # `unclassified` is how a judgment is deliberately withdrawn.
@@ -493,8 +493,8 @@ def validate_classifications(items: list[dict]) -> str | None:
     """The structural error a classification batch would earn, or None.
 
     Pure and checked BEFORE anything is touched: a batch either applies or
-    errors whole — the StrictArgs lesson (#2709), a caller must never learn
-    later that half a batch silently happened.
+    errors whole, so a caller never learns later that half a batch silently
+    happened.
     """
     if not items:
         return "classifications is empty — nothing to apply"
@@ -520,7 +520,7 @@ def validate_classifications(items: list[dict]) -> str | None:
         if status in ("variant", "exempt") and not (item.get("reason") or "").strip():
             return (
                 f"classifications[{i}]: status {status!r} needs a reason — "
-                "the WHY is the record (note 2786)"
+                "the WHY is the record"
             )
         code = (item.get("reason_code") or "").strip()
         if code and code not in REASON_CODES:
@@ -554,7 +554,7 @@ async def classify_shapes(
     caller listed. Idempotent by construction.
 
     ``repo`` — a repo bound to the project — lets a judgment land on a shape
-    the ledger has not synced yet (milestone 439): the shape the agent wrote
+    the ledger has not synced yet: the shape the agent wrote
     this turn. Such an item becomes a PROVISIONAL row under that repo, the
     same row the write-path hook has always created, with no seen commit;
     the next sync confirms it, or marks it vanished until the code reaches
@@ -578,7 +578,7 @@ async def classify_shapes(
         raise ValueError(f"project {project_id} not found or no write access")
 
     # Snippet targets resolve through the caller's own read access — a
-    # family-canon snippet in another project counts (note 2786), a snippet
+    # family-canon snippet in another project counts, a snippet
     # the caller cannot read does not exist for them.
     target_ids = {
         int(item["snippet_id"])
@@ -663,10 +663,10 @@ async def classify_shapes(
 
 
 async def _resolve_ideas(user_id: int, items: list[dict]) -> list[dict]:
-    """An item may name a family idea (`idea_id`) in place of a snippet
-    (milestone 463 step 5): it is judged against the idea's reference in the
-    shape's language, or its first when there is none in that language — an
-    idea is shared across languages. An explicit snippet_id wins."""
+    """An item may name a family idea (`idea_id`) in place of a snippet: it
+    is judged against the idea's reference in the shape's language, or its
+    first when there is none in that language — an idea is shared across
+    languages. An explicit snippet_id wins."""
     from scribe.services import family_adoption
 
     out = []
@@ -684,7 +684,7 @@ async def _resolve_ideas(user_id: int, items: list[dict]) -> list[dict]:
 
 
 async def _sync_family(user_id: int, project_id: int, snippet_ids: set[int]) -> list[dict]:
-    """The adoption ledger follows the shapes (milestone 463 step 5): after a
+    """The adoption ledger follows the shapes: after a
     judgment commits, the canon ideas whose references it touched — as the
     new target or the one it replaced — are answered from the shapes.
 
@@ -703,7 +703,7 @@ async def _sync_family(user_id: int, project_id: int, snippet_ids: set[int]) ->
 
 
 def rule_matches(row: CodeShape, *, path: str, pattern: str, kind: str) -> bool:
-    """Does a ledger row fall under a rule-form classification (#2868)?
+    """Does a ledger row fall under a rule-form classification?
     ``path`` is a file or a directory (everything beneath it), ``pattern``
     a shell glob on the symbol (``""`` = every symbol), ``kind`` narrows to
     sym/css. Pure, so the sweep's reach can be tested without a database."""
@@ -734,7 +734,7 @@ async def classify_shapes_where(
     reason_code: str | None = None,
     uses: list[int] | None = None,
 ) -> dict:
-    """The sweep form of classify_shapes (#2868): one judgment applied to
+    """The sweep form of classify_shapes: one judgment applied to
     every live row under ``path`` whose symbol matches ``pattern`` (and
     ``kind``). By default only unjudged rows are touched — `unclassified`
     and the sync's mechanical `scoped` stamp — a sweep must never silently
@@ -816,9 +816,9 @@ async def list_project_shapes(
     to rows the proposer has spoken about: "any", "canon" (an instance-of-#N
     suggestion), "derive" (a repeats-with-no-canon group), or one basis
     name (symbol/reference/text/signature/semantic). ``flag`` narrows to
-    the readout's asks (#2793): "divergence" (new where a canon dominates,
+    the readout's asks: "divergence" (new where a canon dominates,
     `diverges_from` names it), "recheck" (a judged shape whose body moved),
-    or "unused-css" (milestone 302: a css rule no template names).
+    or "unused-css" (a css rule no template names).
     """
     from sqlalchemy import func, or_
 
@@ -854,14 +854,14 @@ async def list_project_shapes(
     elif flag == "recheck":
         conds.append(CodeShape.recheck_at.isnot(None))
     elif flag == "unused-css":
-        # The consumer map's negative space (milestone 302): a live css rule
+        # The consumer map's negative space: a live css rule
         # no file's markup names. A candidate for deletion, surfaced — never
         # deleted — because the map reads templates only (a class built at
         # runtime, or used from a script, is invisible to it).
         conds.append(CodeShape.kind == "css")
         conds.append(~CodeShape.id.in_(select(CodeShapeConsumer.shape_id)))
     if uses:
-        # Consumers of a canon (#2870): rows with a uses edge to it, whatever
+        # Consumers of a canon: rows with a uses edge to it, whatever
         # shape they themselves are.
         conds.append(CodeShape.id.in_(
             select(CodeShapeUse.shape_id).where(CodeShapeUse.snippet_id == uses)
@@ -899,11 +899,11 @@ def _consumer_dict(row: CodeShape) -> dict:
 
 
 async def snippet_consumers(user_id: int, note_id: int) -> dict:
-    """The structured consumer map for one snippet (#2789): its `instances`
+    """The structured consumer map for one snippet: its `instances`
     (rows judged to conform) and `variants` (named departures, each carrying
     its why). Rows are filtered to projects the CALLER can read — a shared
     snippet must not become a side channel into someone else's project
-    layout. Empty lists mean "attach nothing" (#2483)."""
+    layout. Empty lists mean "attach nothing"."""
     from scribe.services import access
 
     async with async_session() as session:
@@ -916,7 +916,7 @@ async def snippet_consumers(user_id: int, note_id: int) -> dict:
                 )
             )
         ).scalars().all()
-        # The consumption edges (#2870): rows that USE this canon, whatever
+        # The consumption edges: rows that USE this canon, whatever
         # shape they are themselves — the call-site map.
         using = (
             await session.execute(
@@ -951,7 +951,7 @@ async def snippet_consumers(user_id: int, note_id: int) -> dict:
     return out
 
 
-# --- write-path stamping (#2791): hook evidence lands as rows ----------------
+# --- write-path stamping: hook evidence lands as rows -----------------------
 #
 # The write-path hook (plugin/hooks/scribe_prior_art.sh) fires on every
 # Write/Edit and already carries the two halves of a consumer-map row: the
@@ -959,7 +959,7 @@ async def snippet_consumers(user_id: int, note_id: int) -> dict:
 # it could not say on its own is WHICH canon the session is instantiating.
 # The pull stream answers that: a snippet the session opened in full
 # (get_snippet) and is now writing code that references or resembles is being
-# reused — and a reused canon's call site is an `instance` (note 2786).
+# reused — and a reused canon's call site is an `instance`.
 #
 # The rule, deliberately two-sided so it cannot fire on noise:
 #   PULLED   — a PULLED usage event by this user inside PULL_WINDOW. Offered-
@@ -968,9 +968,9 @@ async def snippet_consumers(user_id: int, note_id: int) -> dict:
 #              semantic arm scored it above the write-path threshold for this
 #              very payload. Either is evidence; the pull alone is not.
 # Both hold → every shape the hook named at that path, of the snippet's kind,
-# carries N as a PROPOSAL (milestone 439). Until then it became instance-of-N
-# with classified_by="hook" — a verdict nobody read, and the rows written that
-# way are what `stamps_to_review` still lists.
+# carries N as a PROPOSAL. Older rows stamped instance-of-N with
+# classified_by="hook" — a verdict nobody read — are what `stamps_to_review`
+# lists.
 #
 # A proposal is EVIDENCE, not judgment: it lands only on rows nobody has
 # judged, never touches a status, and is what the end-of-turn question shows
@@ -980,7 +980,7 @@ async def snippet_consumers(user_id: int, note_id: int) -> dict:
 # precision comes from the in-play test above, not from this window.
 PULL_WINDOW = timedelta(hours=6)
 
-# ── WHAT KIND OF THING a shape is, past `css | sym` (#4204) ──────────────
+# ── WHAT KIND OF THING a shape is, past `css | sym` ──────────────────────
 #
 # THE PROBLEM THIS EXISTS FOR. `kind` has exactly two values, so a frozen
 # dataclass, a module constant, a sync predicate, a class and an async service
@@ -1052,7 +1052,7 @@ def shape_form(signature: str, kind: str = "sym") -> str:
     if kind == "css":
         return "css"
     if kind == "file":
-        # A whole file (milestone 439): its own family, comparable only with
+        # A whole file: its own family, comparable only with
         # other files — a component is never told to build from a helper.
         return "file"
     sig = (signature or "").strip()
@@ -1101,10 +1101,10 @@ def forms_conflict(a: str, b: str) -> bool:
     speaks for every symbol in it; that needs positive agreement before
     asserting anything, which is `forms_agree`.
 
-    Demanding agreement everywhere was the first version of this and it was
-    wrong: it silenced the by-name path whenever a shape's definition was not
-    in the payload — an Edit rather than a Write — turning strong evidence
-    into no evidence for a reason that has nothing to do with the code.
+    Demanding agreement everywhere would silence the by-name path whenever a
+    shape's definition is not in the payload — an Edit rather than a Write —
+    turning strong evidence into no evidence for a reason that has nothing to
+    do with the code.
     """
     return bool(a) and bool(b) and a != b
 
@@ -1112,9 +1112,8 @@ def forms_conflict(a: str, b: str) -> bool:
 def shape_family(form: str) -> str:
     """The coarse category a form belongs to, or "" when the form is unknown.
 
-    WHY THIS IS SEPARATE FROM `shape_form`, and it is the correction to the
-    first version of this change. The two checks in this module ask opposite
-    questions, so they cannot share a burden:
+    WHY THIS IS SEPARATE FROM `shape_form`. The two checks in this module ask
+    opposite questions, so they cannot share a burden:
 
       STAMPING asserts "this IS that canon". Agreement in form is evidence
       FOR the claim, so `forms_agree` / `forms_conflict` — the precise level,
@@ -1122,24 +1121,21 @@ def shape_family(form: str) -> str:
 
       DIVERGENCE asserts "this is NOT the canon that dominates here — did you
       mean to?" A form MISMATCH is the premise of that prompt, not an
-      objection to it. Gating it on `forms_agree` inverted the check: it went
-      silent on exactly the mismatches it exists to catch, and the acceptance
-      case of #2793 — a hand-rolled sync `confirmDanger` in a directory where
-      an async confirm helper is canon — stopped being flagged.
+      objection to it. Gating it on `forms_agree` would invert the check: it
+      would go silent on exactly the mismatches it exists to catch, such as a
+      hand-rolled sync `confirmDanger` in a directory where an async confirm
+      helper is canon.
 
     So divergence gates at FAMILY level and only on contradiction. A sync
     function beside an async one is still a fair question. A frozen dataclass
-    told to build from an async service function is not a question at all,
-    and that is the #4204 prompt this removes.
+    told to build from an async service function is not a question at all.
 
-    This does NOT remove every false prompt #4204 recorded. Of the five, it
-    silences `Point` (a type, against a callable canon); `_p`, `get_point`,
-    `is_registered` and `sources_expected_to_emit` are callables like the
-    canon and still ask. At the signature level they are indistinguishable
-    from the #2793 case above, so nothing readable here can separate them —
-    only widening `kind` past `css | sym` (#4204 option 2) or comparing
-    meaning rather than form can. Stated here so the next reader does not
-    assume the gap is an oversight.
+    This does NOT remove every false prompt. A type against a callable canon
+    is silenced; an unrelated callable beside a callable canon still asks,
+    because at the signature level it is indistinguishable from the
+    `confirmDanger` case above. Only widening `kind` past `css | sym` or
+    comparing meaning rather than form can separate them. The gap is known,
+    not an oversight.
     """
     if form in ("fn", "async-fn"):
         return "callable"
@@ -1192,31 +1188,27 @@ _UNATTENDED_BY = ("hook",)
 def canon_coherence(members: Iterable) -> dict:
     """Does a canon's membership still agree on what the canon IS?
 
-    The verdict is NOT "do all the rows share a form". That was the first
-    version, and on this surface's first live day both canons it reported
-    were sound (#4220) while the one genuinely polluted canon had already
-    been cleaned by hand. Two things were wrong with it, and the fix for the
-    first nearly broke the second:
+    The verdict is NOT "do all the rows share a form". A form-agreement
+    check reports sound canons and misses polluted ones, for two reasons, and
+    the fix for the first can break the second:
 
     1. FAMILY, NOT FORM. A sync loop-starter and the async tick it schedules
-       are one shape written two ways; `shape_family` already collapses `fn`
-       and `async-fn` into `callable` for exactly this reason on the
-       divergence side, and this side never asked. Snippet #2849 — four
-       starters, three ticks, every row judged by an audit — scored 4/7 and
-       was called incoherent when nothing about it is.
+       are one shape written two ways; `shape_family` collapses `fn` and
+       `async-fn` into `callable` for exactly this reason on the divergence
+       side. A canon of four starters and three ticks is coherent, and a
+       form count would score it 4/7.
 
-    2. A METHOD OF A MEMBER IS NOT A STRANGER. Snippet #2844 is the
-       SQLAlchemy model convention and its own text covers the class AND the
-       `to_dict` the class must carry. Its 37 classes and 25 serialisers
-       scored 37/62 = 0.597, missing the bar by three thousandths for
-       containing exactly what it says it contains.
+    2. A METHOD OF A MEMBER IS NOT A STRANGER. A model convention can cover
+       the class AND the `to_dict` the class must carry. Thirty-seven classes
+       and twenty-five serialisers score 37/62 on form, under the bar, for
+       containing exactly what the canon says it contains.
 
     THE TRAP, and it is why this is not simply a family histogram: grouping
-    by family and keeping a majority test makes the check BLIND. Before #2844
-    was cleaned it held 37 classes and 56 callables; as families that is
-    56/93 = 0.602, a clean pass, and the 31 rows that had no business being
-    there — Vue functions, route handlers, a dozen tests — would never have
-    been reported. A looser bar in the same shape is not a fix.
+    by family and keeping a majority test makes the check BLIND. The same
+    model canon polluted with 31 unrelated callables — Vue functions, route
+    handlers, tests — holds 37 classes and 56 callables; as families that is
+    56/93, a clean pass, and the 31 rows that have no business being there
+    are never reported. A looser bar in the same shape is not a fix.
 
     So the verdict is inverted. Instead of asking whether most rows agree, it
     asks how many rows the canon CANNOT ACCOUNT FOR:
@@ -1228,18 +1220,18 @@ def canon_coherence(members: Iterable) -> dict:
         of the readable rows make the canon incoherent.
 
     The majority vote abstains those methods, so a class's own serialisers
-    cannot outvote the classes and turn the members into the strangers. On
-    the real ledger the three populations separate completely: clean #2844
-    has 0 strangers in 62, #2849 has 0 in 7, and polluted #2844 had 31 in 93.
+    cannot outvote the classes and turn the members into the strangers. The
+    three cases above separate completely: the clean model canon has 0
+    strangers in 62, the loop canon 0 in 7, and the polluted one 31 in 93.
 
     Scoped to the REVIEW surface on purpose. `canon_form` still answers at
     the precise form level for stamping and divergence, where a sync helper
     beside an async canon is a fair question; nothing here changes what the
     ledger writes.
 
-    Returns a dict, deliberately. Widening a tuple return is the #4204 break
-    exactly — a 4-tuple grew a fifth field and one consumer went on unpacking
-    four, and Python said nothing until that line ran.
+    Returns a dict, deliberately. Widening a tuple return breaks every
+    consumer that still unpacks the old width, and Python says nothing until
+    that line runs.
     """
     forms: dict[str, int] = {}
     families: dict[str, int] = {}
@@ -1383,13 +1375,11 @@ async def recent_pulls(user_id: int, *, window: timedelta = PULL_WINDOW) -> dict
 
 # THE HOOK'S EVIDENCE, WRITTEN ONCE AND READ BACK BY ONE PARSER.
 #
-# The stamp's score used to exist only inside a prose sentence, which meant
-# nothing could ask "how strong was the evidence for this row?" without
-# re-deriving it. That is how 32 rows on one project and 334 on another sat
-# unexamined: the number was printed, never stored, never queried. Format and
+# The stamp's score is stored in a fixed format so something can ask "how
+# strong was the evidence for this row?" without re-deriving it. A score that
+# is only printed is never queried, and weak rows sit unexamined. Format and
 # parser live beside each other so a change to one that breaks the other is a
-# visible edit rather than a silent drift (rule 33's reasoning, one module
-# down).
+# visible edit rather than a silent drift.
 _REFERENCE_REASON = "hook: pulled #{sid}; payload references `{symbol}`"
 _RESEMBLE_REASON = "hook: pulled #{sid}; payload resembles it ({score:.2f})"
 _RESEMBLE_READ = re.compile(
@@ -1435,12 +1425,11 @@ async def suggest_write_path_instances(
 ) -> list[dict]:
     """Offer hook evidence as a PROPOSAL on the shapes being written.
 
-    This used to land the evidence as `instance` rows, classified_by=hook —
-    a permanent classification written with nobody reading it. That is what
-    poisoned two ledgers (132 and 488 rows below today's floor, #4608), and
-    what milestone 439 retires: the agent that wrote the code says what it is,
-    at the end of the turn, and this evidence is what it is shown when it
-    does. So the row keeps its status; the candidate canon goes into the
+    Evidence never lands as an `instance` row: that would be a permanent
+    classification written with nobody reading it. The agent that wrote the
+    code says what it is, at the end of the turn, and this evidence is what
+    it is shown when it does. So the row keeps its status; the candidate
+    canon goes into the
     proposal fields (`proposed_snippet_id`, `proposal_basis` "reference" for a
     by-name hit or "semantic" for a resemblance, `proposal_score`), where the
     end-of-turn question reads it as "looks like #N". A judged row is left
@@ -1477,7 +1466,7 @@ async def suggest_write_path_instances(
 
     # Which pulled canons are in play for this payload, by kind, ranked.
     #
-    # THE FORM OF THE CANON TRAVELS WITH IT (#4204), because what gets written
+    # THE FORM OF THE CANON TRAVELS WITH IT, because what gets written
     # here is an ASSERTION — "this shape IS that canon" — and it is permanent
     # until something re-judges it. The evidence below is payload-level: a
     # resemblance score is computed against the whole file, so without a
@@ -1499,12 +1488,10 @@ async def suggest_write_path_instances(
         elif resembles.get(sid, 0.0) >= _RESEMBLE_MIN:
             rank, why = 1, _RESEMBLE_REASON.format(sid=sid, score=resembles[sid])
         else:
-            # A SCORE BELOW THE FLOOR IS NOT WEAK EVIDENCE, IT IS NONE. This
-            # branch used to be `elif sid in resembles`, which took any score
-            # at all — 0.69 counted exactly as much as 0.95, and the number was
-            # printed into the reason line while never being tested against
-            # anything. A row written on that basis is indistinguishable
-            # afterwards from one written on real evidence.
+            # A SCORE BELOW THE FLOOR IS NOT WEAK EVIDENCE, IT IS NONE.
+            # Taking any score at all (`elif sid in resembles`) would count
+            # 0.69 exactly as much as 0.95, and a row written on that basis is
+            # indistinguishable afterwards from one written on real evidence.
             continue
         in_play.setdefault(kind, []).append(
             (rank, pulled_at, sid, why, shape_form(fields.get("signature") or "", kind))
@@ -1531,10 +1518,10 @@ async def suggest_write_path_instances(
             if not bucket:
                 continue
             row = by_key.get((name, kind))
-            # PER SHAPE, NOT PER FILE (#4204). The best candidate whose FORM
+            # PER SHAPE, NOT PER FILE. The best candidate whose FORM
             # matches this shape's — not simply the best candidate. A class and
-            # a function in one file can no longer be handed the same canon
-            # because the file as a whole resembled it.
+            # a function in one file are never handed the same canon because
+            # the file as a whole resembled it.
             #
             # The signature comes from the payload first: a shape being written
             # right now may have no ledger row yet, and then the code is the
@@ -1566,7 +1553,7 @@ async def suggest_write_path_instances(
                 by_key[(name, kind)] = row
             elif not (row.status in _MECHANICAL_TODO or row.classified_by == "hook"):
                 continue  # a judgment — or the canon itself — stands
-            # A proposal, never a status: the writer decides (milestone 439).
+            # A proposal, never a status: the writer decides.
             row.proposed_snippet_id = sid
             row.proposal_basis = "reference" if _rank >= 2 else "semantic"
             row.proposal_score = resembles.get(sid) if _rank < 2 else 1.0
@@ -1575,15 +1562,14 @@ async def suggest_write_path_instances(
                 "path": path, "symbol": name, "kind": kind,
                 "snippet_id": sid, "reason": why,
             })
-            # Every pulled canon the payload NAMES is a uses edge (#2870) — the
+            # Every pulled canon the payload NAMES is a uses edge — the
             # call-site fact, independent of which one the row is judged to be.
             await record_uses(
                 session, row,
-                # Indexed, not destructured: this list widened from 4 to 5 when
-                # the canon's form joined it (#4204) and a positional unpack
-                # here went on reading four. The unit lane never touches
-                # `record_uses`, so it stayed green and the integration lane
-                # was the only thing that said so.
+                # Indexed, not destructured: the bucket tuples carry the
+                # canon's form as a fifth field, and a positional unpack breaks
+                # silently whenever they widen. The unit lane never reaches
+                # `record_uses`; only the integration lane would catch it.
                 [t[2] for t in bucket if t[0] == 2],
                 basis="hook", evidence="write path: pulled the snippet, payload names its symbol",
             )
@@ -1592,7 +1578,7 @@ async def suggest_write_path_instances(
     return suggested
 
 
-# --- the mechanical proposer (#2792): the machine proposes, judgment classifies
+# --- the mechanical proposer: the machine proposes, judgment classifies
 #
 # Runs inside the coverage refresh — the one moment the shape BODIES exist
 # (the archive is in memory; the ledger stores fingerprints, never code).
@@ -1605,8 +1591,7 @@ async def suggest_write_path_instances(
 #   text       whitespace-insensitive containment either way between the
 #              body and the canon's recorded code (git-grep, in effect —
 #              a verbatim copy, or the canon's own call-site example);
-#   reference  the body calls/uses a canon's symbol — the call-site shape
-#              the P7 backfill classified by hand (#2790);
+#   reference  the body calls/uses a canon's symbol — the call-site shape;
 #   signature  a sym whose definition line, name blanked, resembles the
 #              canon's (the move_*/resequence family shape);
 #   semantic   the widest net: the body (concept-queried, as the write-path
@@ -1617,7 +1602,7 @@ async def suggest_write_path_instances(
 # A hit is a PROPOSAL on the row (proposed_snippet_id/basis/score), never a
 # classification: an agent confirms in batches (confirm_proposals) or judges
 # otherwise (classify_shapes clears it). Rows with no canon hit are grouped
-# by the derive-first rule (note 2786): the same body fingerprint in ≥2
+# by the derive-first rule: the same body fingerprint in ≥2
 # places, or the same name defined in ≥3 files, is "a repeating shape with
 # no canon — derive one first", carried as proposal_basis="derive" + a group
 # key so sessions see the consolidation candidates as one thing.
@@ -1627,7 +1612,7 @@ async def suggest_write_path_instances(
 # recur by convention, not by duplication).
 _DERIVE_MIN_DUP = 2
 _DERIVE_MIN_NAME = 3
-# CSS is never grouped by body (note 2917): classes for different purposes
+# CSS is never grouped by body: classes for different purposes
 # share declarations because the style system makes them alike — `.text-muted`
 # and `.pin-badge-auto` carrying the same `color: var(--fs-text-tertiary)` are
 # two meanings, not two copies. A CSS family is a NAME defined in more than
@@ -1639,40 +1624,36 @@ _DERIVE_MIN_NAME_CSS = 2
 _SEMANTIC_CAP = 150
 # The semantic arm has NO floor of its own: it uses the write-path surface's,
 # because it asks the write-path hint's question — "which canon does this body
-# resemble?" — of the same snippet documents (#4208).
+# resemble?" — of the same snippet documents.
 #
-# It had one, 0.8, set after the first live run paired alembic
-# upgrade()/downgrade() with unrelated canons at 0.68–0.75. That floor was
-# sized for a proposal nobody reads, and nobody-reads is not how proposals are
-# consumed: a judge reviews each semantic one (confirm_proposals refuses an
-# unnamed batch, and the tool says "semantic deserves a look"). Measured on
-# 2026-09-22 it proposed 0 of 150, while judged instances of the service-unit
-# canon score 0.68–0.71 against it. A wrong proposal costs the judge a look; a
-# floor that yields nothing costs the signal. Noise in this band is expected —
-# the score rides on the proposal so the judge can weigh it.
+# A higher floor (0.8) would be sized for a proposal nobody reads, and that is
+# not how proposals are consumed: a judge reviews each semantic one
+# (confirm_proposals refuses an unnamed batch, and the tool says "semantic
+# deserves a look"). Judged instances of a canon typically score 0.68–0.71
+# against it, so 0.8 proposes nothing. A wrong proposal costs the judge a
+# look; a floor that yields nothing costs the signal. Noise in this band is
+# expected — the score rides on the proposal so the judge can weigh it.
 #
 # How many hits the arm scans for an ALLOWED canon. Wider than the hint's
 # budget because disallowed snippets (other language families) share the
-# ranking: one measured true instance sat 4th, behind three it could not use.
+# ranking: a true instance can sit 4th, behind three it could not use.
 _SEMANTIC_LIMIT = 8
-# A MISS FROM THIS ARM IS NOT EVIDENCE, and nothing may read it as one (#4208).
-# #4208 briefly stored "compared, nothing cleared the floor" and let it
-# withdraw divergence prompts. It silenced real divergences exactly as it
-# silenced helpers: most judged instances of a canon score below any floor
-# that keeps noise out (0.66 and under, in the measurement above), and helpers
-# score 0.66–0.75 against snippets they have nothing to do with. A code body
-# against a prose-forward snippet document measures the wrong field (#2518);
-# no floor separates the two bands. The arm PROPOSES on a hit; its silence
-# says nothing.
+# A MISS FROM THIS ARM IS NOT EVIDENCE, and nothing may read it as one.
+# Letting "compared, nothing cleared the floor" withdraw a divergence prompt
+# silences real divergences exactly as it silences helpers: many judged
+# instances of a canon score below any floor that keeps noise out (0.66 and
+# under), and helpers score 0.66–0.75 against snippets they have nothing to do
+# with. A code body against a prose-forward snippet document measures the
+# wrong field; no floor separates the two bands. The arm PROPOSES on a hit;
+# its silence says nothing.
 #
-# THE FAMILY ARM (milestone 463 step 5). The arm above is held to the shape's
-# own project and language family, because across projects it was pure noise
-# (#2871). A family idea is the exception the hold was waiting for: an idea
-# shared across projects ON PURPOSE, whose reference may be in another
-# language (a Python throttle is an instance of a Go one). So the arm also
-# proposes the reference of a canon idea that reaches the shape's project
-# through a shared platform — any language, any project — under a stricter
-# rule of its own, measured 2026-10-06 (#4991) on cross-project pairs:
+# THE FAMILY ARM. The arm above is held to the shape's own project and
+# language family, because across projects it is noise. A family idea is the
+# exception: an idea shared across projects ON PURPOSE, whose reference may be
+# in another language (a Python throttle is an instance of a Go one). So the
+# arm also proposes the reference of a canon idea that reaches the shape's
+# project through a shared platform — any language, any project — under a
+# stricter rule of its own, calibrated on cross-project pairs:
 #
 #   true pairs   A py→go 0.728 (next 0.703)   B py→go 0.715 (next 0.675)
 #                D kt→kt 0.816 (next 0.734)   E go→go 0.610 (next 0.604)
@@ -1690,16 +1671,14 @@ _FAMILY_FLOOR = 0.70
 
 
 def _semantic_priority(row) -> tuple:
-    """Order for the capped semantic pass: new shapes first (#4208, measured
-    on the first live refresh after the proposer version was bumped).
+    """Order for the capped semantic pass: new shapes first.
 
     The pass is capped (`_SEMANTIC_CAP`), and `flag_divergence` acts only on
     shapes NEW since the previous refresh — and skips one the proposer matched
     to the dominant canon. In plain row order the new shapes are the highest
     ids, the back of the queue, so whenever the backlog exceeds the cap (every
-    `_PROPOSER_VERSION` bump queues the whole todo at once) they were flagged
-    before the arm could propose for them: 1,533 rows queued, 150 checked, and
-    every new shape left unread.
+    `_PROPOSER_VERSION` bump queues the whole todo at once) they would be
+    flagged before the arm could propose for them.
 
     So: the human todo (`unclassified`) before `scoped`, newest first within
     each. A row with no creation time sorts last in its group.
@@ -1707,27 +1686,19 @@ def _semantic_priority(row) -> tuple:
     created = row.created_at.timestamp() if row.created_at is not None else float("-inf")
     return (row.status != "unclassified", -created)
 # Bump when a basis's rule changes: rows remember the (body, ruleset) they
-# were examined under, so a tightened rule re-examines everything once.
-# v3: language-family gate on the sym bases, reference stoplist, semantic
-# restricted to the shape's own project (#2871).
-# v4: the semantic arm recorded its misses too (#4208). Retired — a miss is
-# not evidence (see _SEMANTIC_LIMIT) — but the proposals themselves did not
-# change, so no re-examination is owed; a stale miss basis is inert.
-# v5: the semantic arm's floor dropped from 0.8 to the write-path floor (#4208),
-# and the signature floor from 0.8 to 0.75 (#4306), so rows examined and
-# found nothing for must be read once more.
-# v6: the family arm (milestone 463 step 5) — a canon idea's reference on a
-# shared platform, any language, as the top hit at _FAMILY_FLOOR.
+# were examined under, so a tightened rule re-examines everything once. A
+# change that does not alter what is proposed needs no bump. A stale "miss"
+# basis on an old row is inert: a miss is not evidence (see _SEMANTIC_LIMIT).
 _PROPOSER_VERSION = 6
 # Signature resemblance floor, name blanked (difflib ratio) — and a length
 # floor, because `def NAME():` resembles `def NAME(x):` at 0.95 while saying
 # nothing; a family shape has parameters to resemble.
 #
 # A proposal is read before it is confirmed, so this admits the band where
-# true instances live rather than the band where nothing is wrong (#4306).
-# Measured 2026-09-22 over 459 judged instance rows against every same-family
-# canon: true instances score 1.0 or ~0.795 (a model class whose mixin list
-# differs from the canon's), and 0.8 cut the whole second band. At 0.75 all
+# true instances live rather than the band where nothing is wrong. Over
+# judged instance rows against every same-family canon, true instances score
+# 1.0 or ~0.795 (a model class whose mixin list differs from the canon's), and
+# 0.8 cuts the whole second band. At 0.75 all
 # of them clear; the price is `async def NAME(user_id: int, …) -> dict`
 # look-alikes (35 wrong-canon pairings vs 14 at 0.8), which read as wrong at
 # a glance. Below 0.75 the next true row is at 0.6 behind 150+ wrong ones.
@@ -1749,12 +1720,11 @@ class Canon(NamedTuple):
     language: str = ""   # the snippet's recorded language; "" = unknown, no gate
 
 
-# Language families: the sym bases only propose within one. The 2026-08
-# audit (#2871) found every cross-language hit wrong — a Python tool-module
-# canon named `register` proposed for Vue `handleSubmit`s that call
-# `authStore.register()`, and a TS store's `register` matched it by symbol;
-# Minstrel/Forge TS canon proposed for Python bodies by resemblance. CSS is
-# its own kind and is not gated here.
+# Language families: the sym bases only propose within one. Cross-language
+# hits are wrong — a Python tool-module canon named `register` would be
+# proposed for Vue `handleSubmit`s that call `authStore.register()`, a TS
+# store's `register` would match it by symbol, and TS canon would be proposed
+# for Python bodies by resemblance. CSS is its own kind and is not gated here.
 _FAMILY_BY_LANG = {
     "python": "py", "py": "py",
     "typescript": "js", "ts": "js", "tsx": "js", "javascript": "js", "js": "js",
@@ -1796,8 +1766,7 @@ def same_family(path: str, canon_language: str) -> bool:
 
 # Reference basis: generic verbs name too many unrelated things to count a
 # bare mention as a call site of THIS canon (`register`, `load`, `save` …).
-# The symbol basis still catches a second definition of such a name; the
-# call-site relation for these becomes a `uses` edge once #2870 lands.
+# The symbol basis still catches a second definition of such a name.
 _REFERENCE_STOPLIST = frozenset({
     "get", "set", "put", "post", "load", "save", "run", "main", "init", "setup",
     "register", "restore", "reset", "toggle", "close", "open", "submit", "handler",
@@ -1836,7 +1805,7 @@ def text_contains(body: str, code: str) -> bool:
 
 
 def reference_canons(kind: str, path: str, symbol: str, body: str, canons: Iterable[Canon]) -> list[int]:
-    """Every canon this body NAMES (#2870) — the uses edges the proposer can
+    """Every canon this body NAMES — the uses edges the proposer can
     write mechanically: same kind, same language family, symbol not in the
     generic-verb stoplist, and not the shape's own name."""
     norm_sym = _norm_symbol(symbol)
@@ -1903,7 +1872,7 @@ def match_canon(
 
 async def canon_catalog(user_id: int) -> list[Canon]:
     """Every snippet this user can browse, as matchable canon — own projects
-    and shared ones alike, because family canon counts (note 2786)."""
+    and shared ones alike, because family canon counts."""
     from scribe.models.note import Note
     from scribe.services.access import browsable_notes_clause
     from scribe.services.coverage import extract_definitions
@@ -2022,10 +1991,9 @@ async def propose_for_repo(
     if canons is None:
         canons = await canon_catalog(user_id)
     by_key = {(d[0], d[1], d[2]): d for d in definitions}
-    # The semantic arm is the widest net and, across projects, was pure noise
-    # in the 2026-08 audit (#2871): it is held to the shape's own project and
-    # language family. The precise bases (symbol/text) still reach family
-    # canon in other projects (note 2786).
+    # The semantic arm is the widest net and, across projects, is noise: it
+    # is held to the shape's own project and language family. The precise
+    # bases (symbol/text) still reach family canon in other projects.
     sym_canons = [c for c in canons if c.kind == "sym" and c.project_id == project_id]
 
     def semantic_allowed(path: str) -> set[int]:
@@ -2062,7 +2030,7 @@ async def propose_for_repo(
             if row.proposed_at is not None and row.proposed_sha == examined_as:
                 continue
             if row.status not in _MECHANICAL_TODO:
-                # A judged row gets no proposal — but its uses edges (#2870)
+                # A judged row gets no proposal — but its uses edges
                 # are a fact about the body, judged or not: the consumer map
                 # of a canon must include the call sites someone already
                 # classified. Mark it examined so the scan runs once per body.
@@ -2083,7 +2051,7 @@ async def propose_for_repo(
             row.proposal_group = group
             row.proposed_at = now
             row.proposed_sha = examined_as
-            # Consumption is recorded for every canon the body names (#2870),
+            # Consumption is recorded for every canon the body names,
             # whatever the row is then judged to be.
             used = reference_canons(row.kind, row.path, row.symbol, body, canons)
             if used:
@@ -2126,7 +2094,7 @@ def derive_groups(
     (kind `sym`) identical bodies in ≥2 places group as `dup:<sha>` and the
     same name defined in ≥3 files groups as `name:sym:<symbol>`, the copy
     before the name. CSS groups by name only — the same class defined in
-    ≥2 files is `name:css:<symbol>`; its body never groups it (note 2917)."""
+    ≥2 files is `name:css:<symbol>`; its body never groups it."""
     by_sha: dict[str, list[tuple[str, str, str]]] = {}
     by_name: dict[tuple[str, str], list[tuple[str, str, str]]] = {}
     for path, kind, symbol, sha in rows:
@@ -2233,7 +2201,7 @@ def proposal_summary(
         g["files"] = len(files[key])
         if key in consumers:
             g["consumers"] = consumer_summary(consumers[key])
-    # Body-identical groups first (#2872): the things an audit actually
+    # Body-identical groups first: the things an audit actually
     # consolidated were identical bodies under different names/files; a
     # name repeated across modules is usually convention. Within a tier,
     # the group spread over more files is the bigger copy.
@@ -2251,7 +2219,7 @@ def proposal_summary(
 def derive_new_summary(
     rows: Iterable[CodeShape], *, since: datetime | None, top: int = 3
 ) -> dict:
-    """The arrival-moment drift signal (#2899): derive-grouped rows FIRST
+    """The arrival-moment drift signal: derive-grouped rows FIRST
     SEEN after ``since`` — the previous refresh's stamp, the same one
     flag_divergence uses. "Since the last refresh, N more copies joined a
     duplicate family" is the sentence that makes the derive queue a thing
@@ -2341,7 +2309,7 @@ async def confirm_proposals(
     return out
 
 
-# --- the divergence readout (#2793): button B where button A is canon -------
+# --- the divergence readout: button B where button A is canon --------------
 #
 # Three answers the ledger can now give mechanically:
 #   DIVERGENCE  a shape NEW since the previous refresh, in a directory+kind
@@ -2355,27 +2323,21 @@ async def confirm_proposals(
 #   RECHECK     an instance/variant whose body moved since it was judged is
 #               flagged recheck_at (sync) — the judgment stands, re-confirm it.
 
-# A canon dominates a directory+kind when at least this many siblings are
-# judged (canonical/instance) and this share of them answer to one snippet.
 # How much a payload must resemble a canon before the hook SUGGESTS it as the
-# shape's canon — and, for the rows stamped before milestone 439, the bar below
-# which a stored stamp is weak (`is_weak_stamp`). Until milestone 439 this was
-# the bar for the hook to ASSERT, unattended, that a shape IS an instance; the
-# hook no longer asserts anything, and the history below is why.
+# shape's canon — and the bar below which a stored hook stamp (an `instance`
+# row written unattended, classified_by="hook") is weak (`is_weak_stamp`).
+# The hook itself never asserts a classification.
 #
-# WHY A FLOOR AT ALL. There was none: any score the semantic arm produced
-# counted, so 0.69 asserted as confidently as 0.95, and the score went into
-# the reason line without ever being compared to anything. Those rows are
-# permanent — they feed `dominant_canon`, which then tells the next writer in
-# that directory what to build from — so a thin stamp does not stay thin, it
-# compounds.
+# WHY A FLOOR AT ALL. Without one, any score the semantic arm produced would
+# count, 0.69 as confidently as 0.95. A stamp feeds `dominant_canon`, which
+# tells the next writer in that directory what to build from, so a thin stamp
+# does not stay thin, it compounds.
 #
 # 0.80 rather than the retrieval floors near 0.70, deliberately. Those bars
 # decide whether to SHOW someone a record, where being wrong costs a glance.
-# This one decides whether to RECORD a claim about the codebase unattended,
-# where being wrong costs a wrong instruction to everyone who writes in that
-# directory afterwards. An unattended write should need more evidence than a
-# suggestion, not the same.
+# This one weighs a claim about the codebase, where being wrong costs a wrong
+# instruction to everyone who writes in that directory afterwards. A claim
+# should need more evidence than a suggestion to read, not the same.
 _RESEMBLE_MIN = 0.80
 
 
@@ -2384,12 +2346,11 @@ def is_weak_stamp(row) -> bool:
     the floor it would have to clear today?
 
     ONE PREDICATE FOR "WEAK", read by the review surface that lists these rows
-    and by every reader that would otherwise take them as evidence. The floor
-    guards new writes only; the rows stamped before it (#4204) are still
-    stored, and until an agent re-judges them they must not speak. They are
-    not withdrawn here — `stamps_to_review` still lists them for judgment —
-    they are simply not counted, which is the module's rule for evidence it
-    cannot trust: quieter, never more confident (#4608).
+    and by every reader that would otherwise take them as evidence. Rows
+    stamped below the floor stay stored, and until an agent re-judges them
+    they must not speak. They are not withdrawn here — `stamps_to_review`
+    lists them for judgment — they are simply not counted, which is the
+    module's rule for evidence it cannot trust: quieter, never more confident.
     """
     if getattr(row, "classified_by", None) not in _UNATTENDED_BY:
         return False
@@ -2397,6 +2358,8 @@ def is_weak_stamp(row) -> bool:
     return score is not None and score < _RESEMBLE_MIN
 
 
+# A canon dominates a directory+kind when at least this many siblings are
+# judged (canonical/instance) and this share of them answer to one snippet.
 _DENSITY_MIN_JUDGED = 3
 _DENSITY_SHARE = 0.6
 
@@ -2436,9 +2399,9 @@ def dominant_canon(rows: Iterable[CodeShape]) -> tuple[int, int, int] | None:
     counts: dict[int, int] = {}
     judged = 0
     for r in rows:
-        # A weak stamp is not a judgment: counted, it let a YAML CI snippet
-        # "dominate" a Svelte directory and every write there was told it
-        # diverged from it (#4608).
+        # A weak stamp is not a judgment: counted, it could let a YAML CI
+        # snippet "dominate" a Svelte directory, and every write there would
+        # be told it diverged from it.
         if is_weak_stamp(r):
             continue
         if r.status in ("canonical", "instance") and r.snippet_id is not None:
@@ -2466,12 +2429,11 @@ async def canon_density(
     `shape_form` returns it, NOT a family. `families_conflict` coarsens both
     sides itself, and handing it a family makes `shape_family("callable")`
     return "" so nothing is ever excluded: the narrowing silently becomes a
-    no-op that still reads as applied. It narrows the DENOMINATOR
-    (#4208). Without it the base rate was computed over every code symbol in
-    the directory as one bucket: "372 judged siblings" counted a dataclass, a
-    CSS-less constant and an async service unit as three comparable things,
-    and the share that came out of that was a statement about a population
-    nobody had asked a question about.
+    no-op that still reads as applied. It narrows the DENOMINATOR. Without
+    it the base rate is computed over every code symbol in the directory as
+    one bucket, counting a dataclass, a constant and an async service unit as
+    three comparable things, and the share that comes out is a statement about
+    a population nobody asked a question about.
 
     EXCLUDES ONLY A KNOWN CONFLICT, using `families_conflict` — the same
     predicate the divergence gate uses, so the two cannot drift apart. A
@@ -2482,10 +2444,10 @@ async def canon_density(
     this module goes the same way — quieter, never more confident.
 
     NOT narrowed to the candidate's exact form, for the reason `shape_family`
-    gives at length: `fn` beside `async-fn` is the acceptance case of
-    milestone #2793, not noise. Bucketing the denominator by form would take
-    the async canon out of a sync candidate's count and silence that flag —
-    the same inversion the first form gate made, one layer down.
+    gives at length: `fn` beside `async-fn` is a fair question, not noise.
+    Bucketing the denominator by form would take the async canon out of a
+    sync candidate's count and silence that flag — the inversion a form gate
+    makes, one layer down.
     """
     directory = _dir_of(path)
     async with async_session() as session:
@@ -2508,8 +2470,8 @@ async def canon_density(
     # The canon's form travels with the count, read from its OWN judged rows.
     # `dominant_canon` keeps its three-value shape: it answers "what dominates
     # here", which is still a true and separately useful question, and its
-    # test pins that arithmetic. What changed is that nobody acts on the count
-    # alone any more.
+    # test pins that arithmetic. Callers act on the count together with the
+    # canon's form, never on the count alone.
     return dom[0], dom[1], dom[2], canon_form(siblings, dom[0])
 
 
@@ -2526,13 +2488,13 @@ async def write_time_divergence(
     out: list[dict] = []
     if not shapes:
         return out
-    # DENSITY IS NOW PER CANDIDATE, not per kind (#4208): the denominator
+    # DENSITY IS PER CANDIDATE, not per kind: the denominator
     # excludes siblings whose family contradicts what is being written, so it
     # cannot be computed until the candidate's own form is known. Cached on
     # (kind, family) — a write names a handful of shapes and they collapse to
-    # one or two buckets, so this is the same one-or-two queries as before.
+    # one or two buckets, so this stays one or two queries.
     #
-    # The cost is that the row load below no longer sits behind an early exit
+    # The cost is that the row load below does not sit behind an early exit
     # on "nothing is dense here". That is one indexed lookup on
     # (project_id, path), and it has to happen first regardless: the
     # candidate's signature comes from its stored row when the payload does
@@ -2575,7 +2537,7 @@ async def write_time_divergence(
             row.status != "unclassified" or row.proposed_snippet_id == sid
         ):
             continue
-        # IS THIS EVEN THE SAME CATEGORY OF THING AS THE CANON? (#4204)
+        # IS THIS EVEN THE SAME CATEGORY OF THING AS THE CANON?
         # Without this the line was a pure base rate: "most things here are X,
         # so be X", with the candidate never examined at all. It told a frozen
         # dataclass to build from the `async_session` service canon.
@@ -2603,13 +2565,13 @@ _DERIVE_FILES_SHOWN = 4
 async def write_time_derive(
     project_id: int, path: str, shapes: list[tuple[str, str]]
 ) -> list[dict]:
-    """The in-band DERIVE check (#2900): for each (kind, name) the hook
+    """The in-band DERIVE check: for each (kind, name) the hook
     named at ``path``, what the ledger already knows about that name
     elsewhere in the project —
 
       family  the name sits in a derive-first group (code: identical body
               in N files or the same name in ≥3; CSS: the same class in
-              ≥2 files, note 2917): "this is a known family with no canon
+              ≥2 files): "this is a known family with no canon
               — derive it now, don't add a copy";
       canon   a `canonical` row of that name at another path: "this is
               canon #N at <path> — reuse, don't redefine".
@@ -2660,7 +2622,7 @@ async def write_time_derive(
             "size": len(members) + (1 if here is not None else 0),
         }
         if kind == "css":
-            # What renders the family (milestone 302): the members' consumer
+            # What renders the family: the members' consumer
             # files, the row at `path` included when it already exists.
             ids = [r.id for r in members] + ([here.id] if here is not None else [])
             edges = await consumers_of(ids)
@@ -2702,10 +2664,10 @@ async def flag_divergence(project_id: int, *, since: datetime | None) -> int:
                 if r.diverges_from is not None:
                     # A flag is a mechanical prompt resting on one premise —
                     # "this canon dominates here". When the premise no longer
-                    # holds (the rows that elected it were weak stamps, #4608,
-                    # or were re-judged away) the prompt is false, and a false
-                    # prompt on every write is how the flag stopped being
-                    # read. Withdrawing it judges nothing: the row stays
+                    # holds (the rows that elected it were weak stamps, or
+                    # were re-judged away) the prompt is false, and a false
+                    # prompt on every write is a flag nobody reads.
+                    # Withdrawing it judges nothing: the row stays
                     # unclassified, and a judgment is never touched here.
                     if dom is None or dom[0] != r.diverges_from:
                         r.diverges_from = None
@@ -2716,16 +2678,16 @@ async def flag_divergence(project_id: int, *, since: datetime | None) -> int:
                     continue
                 if r.proposed_snippet_id == dom[0]:
                     continue  # the proposer already says "instance of the canon"
-                # NOT the converse (#4208). A semantic MISS does not mean "not
-                # the canon": measured, true instances of a canon rarely clear
-                # the arm's floor either (see the note above _SEMANTIC_LIMIT).
-                # A helper beside a canon and a genuine divergence — #2793's
-                # sync `confirmDanger` beside an async confirm canon — are the
-                # same case to every signal this ledger holds, so the prompt
-                # asks, and the judge answers it (`exempt` with a reason
-                # retires a false one for good).
-                # The same structural test the write-time check applies
-                # (#4204). The sweep and the hook must agree about what counts
+                # NOT the converse. A semantic MISS does not mean "not the
+                # canon": true instances of a canon often miss the arm's floor
+                # too (see the note above _SEMANTIC_LIMIT). A helper beside a
+                # canon and a genuine divergence — a sync `confirmDanger`
+                # beside an async confirm canon — are the same case to every
+                # signal this ledger holds, so the prompt asks, and the judge
+                # answers it (`exempt` with a reason retires a false one for
+                # good).
+                # The same structural test the write-time check applies. The
+                # sweep and the hook must agree about what counts
                 # as divergence, or an audit contradicts the line the writer
                 # was shown at the keyboard.
                 if families_conflict(shape_form(r.signature or "", r.kind), cform):
@@ -2736,7 +2698,7 @@ async def flag_divergence(project_id: int, *, since: datetime | None) -> int:
     return flagged
 
 
-# --- the end-of-turn question (milestone 439): what did this turn write that
+# --- the end-of-turn question: what did this turn write that
 # nobody has judged? ---------------------------------------------------------
 #
 # The write hooks keep a session ledger of the definitions each turn wrote; at
@@ -2822,21 +2784,21 @@ def stamps_to_review(rows: Iterable[CodeShape], *, top: int = 10) -> dict:
     """Judged rows whose EVIDENCE no longer meets the bar the ledger now
     holds, and canons whose own rows no longer agree what they are.
 
-    THIS CHANGES NOTHING, AND THAT IS THE POINT. Every row here was written
-    unattended by the write-path hook on a similarity score, and the damage
-    that caused was not the score being wrong once — it was a machine
-    asserting a permanent classification with nobody reading it. A sweep that
-    silently un-asserted them would be the identical mistake with a larger
-    blast radius and a cleaner conscience. So this reports; an agent reads the
+    THIS CHANGES NOTHING, AND THAT IS THE POINT. Every `weak` row was written
+    unattended by the write-path hook on a similarity score, and the harm in
+    that is not a score being wrong once — it is a machine asserting a
+    permanent classification with nobody reading it. A sweep that silently
+    un-asserted them would be the identical mistake with a larger blast
+    radius and a cleaner conscience. So this reports; an agent reads the
     evidence, decides, and records the decision through `classify_shapes`
     under its own name. The ledger should be able to say "these look wrong"
     without being able to act on it alone.
 
     Two lists, because they are different questions:
 
-    `weak` — rows stamped on a resemblance below `_RESEMBLE_MIN`. The floor
-    arrived after they did (#4204), and a guard at the point of classification
-    does not undo the classifications already stored (lesson #4202). Rows a
+    `weak` — rows stamped on a resemblance below `_RESEMBLE_MIN`. A guard at
+    the point of classification does not undo classifications already stored,
+    so rows written before the floor need reading. Rows a
     person or an audit judged are never listed, however old: a human judgment
     is not weak evidence, it is a different kind of evidence.
 
@@ -2845,14 +2807,13 @@ def stamps_to_review(rows: Iterable[CodeShape], *, top: int = 10) -> dict:
     thing; when its own members are a class, three getters and a dozen tests,
     that claim has stopped being true, and every base-rate reading built on
     it is reading noise. `canon_form` already makes such a canon fall silent
-    — this is what makes it VISIBLE, which is the half that was missing.
+    — this is what makes it VISIBLE.
 
     What counts as disagreement is `canon_coherence`, and it is deliberately
     looser than "one form": it compares FAMILIES, and it does not hold a
-    method against the class it is defined beside. Both corrections came from
-    this surface's first live day, when the only two canons it reported were
-    both sound (#4220). A review surface whose output is noise is one that
-    stops being read, which costs more than the check was ever worth.
+    method against the class it is defined beside. A review surface whose
+    output is noise stops being read, which costs more than the check is
+    worth.
 
     `strangers` names the rows that do not fit, not the first dozen members:
     the reader's question is which ones are wrong. `unattended` counts rows