wasteNot: the pm-adopted duplicate-reason demoter was never built - and 25 days of data now argue about its priority, not its correctness

pm-llmmsgsrv-cc ADOPTED this gate on 2026-09-11 and it is not in the code. Measured against llmmsg-wastenot.sh v2.5 (16d774d): the only reason-content gate present is the CIRCULAR demoter at :463-469, which demotes a 'wasted' whose reason merely restates the silence that SELECTED the message. There is no duplicate-reason demoter. THE RULE AS ADOPTED: identical reason text on 2+ ids within one run demotes ALL of them to undecidable, because one reason covering two messages means at most one was examined, plus a run-level log line so the rate is visible rather than found by spot check. WHAT 25 DAYS OF DATA SAY, measured on wastenot_candidates 2026-10-06 (136 rows, 2026-09-10 to 2026-10-05): exactly three duplicate-reason groups exist, and ALL THREE PREDATE 2026-09-12 - 113739/113742 (09-10 15:06), 114046/114047 (09-10 18:03), 115346/115339 (09-11 14:01). Zero in the 25 days since. Two of the three are already 'undecidable', so the demoter would have changed the verdict of ONE historical group. Every group is confined to a single sender>recipient pair, which is the discriminator for an ANALYST merge rather than an insert-path reuse bug: a write-path fault would eventually attach a reason to a row from a different pair, and across the whole table it never does. SO THE PRIORITY ARGUMENT HAS MOVED AND THE CORRECTNESS ARGUMENT HAS NOT. The harm the gate was written against was a false 'wasted' being DM'd to two working agents. MSG-331 removed that path entirely: a 'wasted' row is now recorded and nothing is sent. What remains is record quality for the rows monitor rules on under MSG-329 - still real, no longer urgent. DO NOT READ THE ZERO AS THE GATE BEING UNNECESSARY. pm's own caveat on adoption is the load-bearing part: DUPLICATE IS A DETECTABLE SYMPTOM OF UNGROUNDED, NOT THE DEFECT. A reason that is false but UNIQUE passes the demoter untouched, and the 2026-09-11 cluster happened to present as duplicates - luck about the presentation, not a property of the failure. The demoter is a FLOOR and its green must never be read as 'reasons are now grounded'. ALSO STILL OWED, and it is what makes the review tier reachable at all: the closed-question verification pass (does this body support this claim, y/n), ruled 2026-09-11 as the gated mechanism and scoped to ALREADY-FLAGGED rows only - a pass over unflagged traffic is not approved. The lexical grounding alternative is RULED OUT and must not be revived: tested against the motivating row, 3 of 13 reason tokens appeared in the bodies and they were exactly the three the analyst had INVERTED, so a token-overlap check passes that reason at any threshold. Lexical overlap measures topic, not truth, and polarity inversion is invisible to every string-level check. Without the verification pass a human reviewer is the only thing separating 'undecidable' from 'never examined', which is a human in the loop on every row forever.