A third specimen of the shared-tail class, and it is outside the repair the row prescribes

dag/gunbc/recurring_failure_mode/merge_region_excludes_shared_tail.dag

@@ -50,6 +50,16 @@ data merge_region_excludes_shared_tail: RecurringFailureMode = RecurringFailureM
     "Worked repair: `gunbc.rung_drop`, whose 30 rows are 30 files under `dag/gunbc/rung_drop/` enumerated by `gunbc.rung_drop.roster`. THE SHARED TEXT IS NOT ZERO AND THE HONEST COUNT IS THE ENUMERATIONS: a new drop still touches the roster line, and a new PROSE drop touches three further lists that exist only to keep the prose arm closed. Each is a whole line with no shared suffix, so each resolves additively -- which is the claim, and it is weaker than `disjoint edits no longer conflict`. ",
 
     "What this does NOT establish is enforcement -- nothing refuses a second row added to one of those files, so the default is safe and the shape is not walled.)",
+
+    "**THIRD INDEPENDENT INSTANCE, AND IT BREAKS THE REPAIR THIS ROW PRESCRIBES** (gunbc colo census carrier, 2026-09-08, silent-heron-303 integrating a carrier branch four lanes are based on). The severed unit was NOT a data row: it was a FUNCTION, `density_reads_never_exceed_facilities`, that both sides had edited. The conflict boundary cut THROUGH the body, so the HEAD side ended mid-function and the other side's closing braces sat BELOW the `>>>>>>>` marker. Replacing the HEAD side with a complete, correctly-braced version left those orphan closers stapled after it. Markers all gone, git satisfied, file structurally broken -- and PUSHED, then caught and repaired one commit later. ",
+
+    "THE REPAIR NAMED ABOVE -- make the appended unit ONE LINE so the shared suffix carries nothing -- DOES NOT REACH THIS. A roster row can be collapsed to one line; a function body cannot, and neither can a type with arms. So the one-line remedy covers the append case and leaves the EDIT case exactly as exposed, which is worth stating because this row's recognition rule asks what suffix appended units share and this specimen had no append at all. The generator is the same and the population is wider than the repair. ",
+
+    "**THE DETECTOR, AND WHY THE OBVIOUS FORM OF IT IS USELESS: A WHOLE-FILE BRACE COUNT CANCELS.** Two errors of opposite sign sum to zero, so a structurally broken file reads as balanced -- the same shape as every absence-that-reads-as-clean class in this ledger. What discriminates is a RUNNING count read per line, reporting the first line where the depth goes NEGATIVE, with string literals stripped first because citation prose contains braces. AND IT MUST BE RUN AGAINST THE PRE-MERGE VERSION OF THE SAME FILE AS A CONTROL: the finder measured 0 on the carrier's version and -2 on their resolution, which is what made the defect attributable to the resolution rather than to pre-existing prose. Without the control a non-zero number names no culprit. ",
+
+    "AND THE HARM IS WIDER THAN THE TOIL THIS ROW RECORDS, without contradicting it. The refusal is still loud at the module index, so silent wrongness is still correctly ruled out -- but the broken file here was a WITNESS module in a session with no .dag parser to hand, so nothing between the resolution and the push could observe it, and the natural local checks a lane runs are all downstream of a parse that never happened. Loud at the authority and silent at every point a lane actually looks is what let it reach a pushed head. ",
+
+    "**AND THE DETECTOR THAT SUBSUMES THE COUNTER ALREADY EXISTS, WHICH RELOCATES THE GAP FROM DETECTION TO REACHABILITY** (keen-newt-324, same night, checking fifteen files clean under the counter and then naming its ceiling). Compiling the closure catches an orphan-brace resolution BY CONSTRUCTION -- it is a parse refusal naming file and line, where the balance counter reports a number that needs a control to interpret at all. So the counter is not the wall; it is the pre-push check that needs no compiler. WHAT THE COMPILE DOES NOT REACH IS A TEST MODULE, BECAUSE NOTHING IMPORTS A TEST: a probe that imports one declaration from each module under test verifies every one of them and never opens the witness file, which is precisely the module the orphan braces landed in. The gap that let a broken file reach a pushed head is therefore not a missing analysis but an UNREACHABLE POPULATION -- the same shape as the coverage module that could not compile because nothing imported it, one ledger row over. The remedy is to name the test file as its OWN `--entry` rather than relying on a probe that imports what it tests. ",
   ],
 
   evidence: [],