Check a partial supersession's date and refuse the older note form, as the review of this change found 0269's note on 0118 went round the check

00-META/checks/records.py

@@ -269,6 +269,10 @@ PARTIAL_MARK = re.compile(
 )
 
 
+# The form these notes had before the check: it named the later record but nothing read it.
+NARROWED_MARK = re.compile(r"^> \*\*Narrowed, not replaced — [^\n]*\[ADR \d{4}\]\(", re.M)
+
+
 def check_partial_supersession(failures, records):
     """A record superseded in part says so under its heading, and its superseder names it.
 
@@ -276,6 +280,8 @@ def check_partial_supersession(failures, records):
     21 `supersedes-in-part:` links had no note on the record they change, so a reader who arrived
     at the older record by a citation read a rule that no longer held. Both directions are
     checked: the field on the newer record and the dated note on the older one, naming each other.
+    The note is dated by the newer record's `date:`, and the older form of the note, which named a
+    later record and was read by nothing, is refused so it cannot be a way round this check.
     """
     for number, record in records.items():
         claimed = record["front"].get("supersedes-in-part", [])
@@ -295,7 +301,22 @@ def check_partial_supersession(failures, records):
                     f"ADR {number} supersedes this in part; this record carries no "
                     f"'> **Superseded in part — YYYY-MM-DD, by [ADR {number}](…).**' note",
                 )
-        for _, by in PARTIAL_MARK.findall(record["text"]):
+        for line in NARROWED_MARK.findall(record["text"]):
+            failures.add(
+                "partial-supersession",
+                rel(record["path"]),
+                f"a note names a later record in the old form ({line.strip()[:60]}…); write it as "
+                f"'> **Superseded in part — YYYY-MM-DD, by [ADR NNNN](…).**' and declare it in that "
+                f"record's `supersedes-in-part:`",
+            )
+        for date, by in PARTIAL_MARK.findall(record["text"]):
+            if by in records and date != str(records[by]["front"].get("date", "")):
+                failures.add(
+                    "partial-supersession",
+                    rel(record["path"]),
+                    f"the note for ADR {by} is dated {date}; ADR {by} is dated "
+                    f"{records[by]['front'].get('date', 'nowhere')}",
+                )
             if by not in records:
                 failures.add("partial-supersession", rel(record["path"]), f"note names no such record: ADR {by}")
                 continue

02-DECISIONS/0009-modules-and-the-graph.md

@@ -43,7 +43,7 @@ hand.
 
 ### What a domain module turns out to be, and why it is not the one refused above
 
-> **Narrowed, not replaced — 2026-10-06, by [ADR 0226](0226-the-private-network-is-assigned-by-its-own-name-and-the-proxy-names-its-public-issuer.md).** The catalogue no longer
+> **Superseded in part — 2026-10-06, by [ADR 0226](0226-the-private-network-is-assigned-by-its-own-name-and-the-proxy-names-its-public-issuer.md).** The catalogue no longer
 > holds a domain module: `networking` was retired, and every machine is assigned the private network's
 > own module. A module with requirements and no files stays a thing a module may be; "why it is worth
 > having" below describes what this record decided for `networking`, not what runs.

02-DECISIONS/0118-undeclaring-gives-a-unit-back-the-state-it-was-found-in.md

@@ -80,7 +80,7 @@ records the state it first found the unit in, and undeclaring returns the unit t
   manager: running before the mesh arrived, running after it leaves.
 - **A unit the mesh started is stopped again**, and one it enabled is disabled again — the packet
   filter a converge loaded, which returning to adopted unloads.
-> **Narrowed, not replaced — 2026-10-09, by [ADR 0269](0269-what-the-meshs-own-declaration-stopped-is-started-again-when-it-goes.md).**
+> **Superseded in part — 2026-10-09, by [ADR 0269](0269-what-the-meshs-own-declaration-stopped-is-started-again-when-it-goes.md).**
 > "Never started on the way out" has one exception. The host now records what the mesh declared beside
 > what it found. A unit found running or enabled that the mesh's own declaration held stopped or disabled
 > is started or enabled again when that declaration goes, but only while it is still exactly as the mesh

02-DECISIONS/0269-what-the-meshs-own-declaration-stopped-is-started-again-when-it-goes.md

@@ -5,6 +5,8 @@ date: 2026-10-09
 deciders: jochen
 reconstructed: false
 extends: 02-DECISIONS/0118-undeclaring-gives-a-unit-back-the-state-it-was-found-in.md
+supersedes-in-part:
+  - 02-DECISIONS/0118-undeclaring-gives-a-unit-back-the-state-it-was-found-in.md
 ---
 
 # 269. What the mesh's own declaration stopped is started again when it goes

02-DECISIONS/0276-merges-are-assembled-in-a-rolling-window-and-each-batch-is-delivered-by-one-walk-with-one-commit-per-repository.md

@@ -18,7 +18,8 @@ extends: 02-DECISIONS/0249-a-delivery-groups-order-rules-have-a-precedence-and-a
 > Decision 4's "one walk at a time", and decision 1's window for every merge, no longer hold: a lone merge is
 > cut at once, and walks that share no repository, module or edge of the module graph run at once, one
 > first-node judgement per node at a time. Two walks of one repository never at once, no walk superseded once
-> it started, and the window for a merge that waits for its delivery group stand as decided here.
+> it started, and the window for a merge that waits for its delivery group or arrives while a batch assembles
+> stand as decided here.
 
 ## Context
 

02-DECISIONS/README.md

@@ -81,7 +81,8 @@ heading and newest first, a note in the form
 > as decided here.
 
 dated by the newer record's `date:`. `00-META/checks/records.py` (`partial-supersession`) fails when
-either half is missing. Saying "this narrows" or "this replaces" an older record in prose alone is not
+either half is missing, when the note's date is not the newer record's, and on a note in the older form
+*Narrowed, not replaced*. Where the note sits is the reviewer's to see in the diff. Saying "this narrows" or "this replaces" an older record in prose alone is not
 enough: on 2026-10-11, 6 records did only that, and 11 of 21 declared links had no note on the record
 they changed.