Retired the shared-pool multi-chain format: withheld all 37 servable multi-chain rows (33 extreme: h2,h3,h4,h5 hard + x1-x5,xe1-xe30 extreme minus xe19/xe21 already withheld; 33 extreme + 4 hard = 37). needs_review_reasons=format-retired, a new token. Format decision, not a content-quality finding -- withheld rather than deleted, reversible. Extreme drops from 46 servable (33 multi-chain + 13 tree) to 13. Before withholding: ran against all 37 parents plus the full easy/linear pool. Of 74 candidate sub-chains, only 2 (volcano from xe8, vandal from xe20) produced genuinely new non-duplicate content after dedup against the WHOLE corpus by source_word, not just own easyWords set -- that set only checks difficulty=easy rows by name, and several existing derived puzzles (sugar/dv001 among them) are difficulty=hard, so two of the scripts own 4 raw candidates (sugar, barbarian) were already-existing words it did not know to skip. Inserted dv176 (volcano, from xe8/xe8b) and dv177 (vandal, from xe20/xe20b), both pending/parity-servable, confirmed by read-back before this withhold. The other 72 sub-chains either already have a derived row (28, from 20 of the 37 parents), were dropped by quality filters (13: 9 repeated-hidden-form, 1 uniform-family-parent, 3 era-inversion), or produced a word duplicating existing corpus content (31). Net: retiring the format loses almost no unique content once duplication is accounted for. Rescheduled 4 future daily_schedule dates off multi-chain rows first (see reschedule entry, same timestamp batch): x2/x3/x5/x4 -> xc10/xc11/xcomp8/xd2, all extreme tree. Bundle resync to follow so survival (which reads getMultiChainPuzzlesHard/Extreme via the SYNC path, not the D1 gate) actually stops drawing these.

Pool contains exact duplicate chips (tempo), so two slots are filled by indistinguishable tiles. Flag only -- sometimes correct, as when a borrowing did not change the form.

removed revealed-form chip(s) from options: "tempus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Weak puzzle and a cleared flag. The chain is unique[Modern English] < unique[French] < unicus[Latin], so both hidden nodes carry the SAME form and the pool contains two identical chips (unique, unique) -- the player places the same word twice with nothing to distinguish the slots. It scored 0.833 on mean_path_adjacent_similarity, above the 0.80 review band, and WAS flagged at insert; runVerify's enrichment branch then overwrote needs_review with passedValidation ? 0 : 1 and erased the flag. That overwrite is being fixed separately. Schedule 2027-01-03 re-pointed to e21 (money, single prior slot 2026-06-22). Row produced by the supervised first production run 2026-08-26.

revealed (payoff) node had empty meaning; added: "single, sole, only one". not currently servable (needs_review already set), fixed for content quality

removed revealed-form chip(s) from options: "unicus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*bainą" x1, "*beruz" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

meaning was a tautological echo of form on both sub-chains ("lemon"/"lemon", "orange"/"orange"); replaced with real glosses

removed revealed-form chip(s) from options: "līmūn" x1, "nāraṅga" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*tredaną" x1, "*dauþaz" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "comes stabuli" x1, "*marhskalk" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "carō" x1, "*langatīnaz" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Moneta" x1, "sal" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "φιλοσοφία" x1, "λόγος" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "yfesdrype" x1, "godsibb" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "jabara" x1, "ṣifr" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Reworded hook/hook_sub/origin_story to present the hashish-eaters derivation as the traditional account rather than settled fact, per Wiktionary, which labels it a folk etymology and gives asasiyyin (people faithful to the foundation) as the primary etymology. Chain and pool unchanged -- Hashshashin/hashish are real attested forms. NO state change: row remains, and stays excluded from serving pending a decision on whether to retire it. The reworded copy is staged, not player-visible.

removed revealed-form chip(s) from options: "ghāla" x1, "ḥashīsh" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Both sub-chains (muscle, mussel) share the identical PIE root *mus via Latin musculus -- structurally violates the multi-chain formats two-unrelated-chains premise. Retired rather than given a substitute sub-chain; authoring and sourcing a genuinely unrelated replacement is new content work, deferred. needs_review remains 1 from the Part 1 flag ; legacy.

removed revealed-form chip(s) from options: "*mūs" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "kamára" x1, "pānis" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "ṣuffa" x1, "al-qubba" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "ʿarrafa" x1, "al-Khwārazmī" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "kômos" x1, "trágos" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*marhaz" x1, "turba" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*knabô" x1, "villa" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "shāh" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*péḷu-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "strephein" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "dulband" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "signum" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "porcus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

added Latin transliteration alongside untransliterated Greek in the kybernetes node meaning, per corpus sweep finding

removed revealed-form chip(s) from options: "κυβερνάω" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Genua" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Átlas" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Bethlehem" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Lakōnía" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "couvre-feu" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "ṣifr" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "húsbóndi" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "berserkr" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "زهر" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "pūpillus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Deleted fabricated second root *gwerh2- (id root-1), wired into the gravis confluence via children[] and a placeable chip. *gwerh2- and the revealed *gwrh2- are the same PIE root in two ablaut grades (full vs zero); the -e- is an ablaut vowel, not a different etymon. Confluence node keeps parentId=root-0; nodes 12->11; maxFanout 4 so extreme threshold still met. Chip removed.

removed revealed-form chip(s) from options: "gravis" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Two corrections in one write. (1) Deleted fabricated second root *dreu- (id root-1), wired into the *trewwaz confluence via children[] and a placeable chip: *dreu- and the revealed *deru- are the same PIE root in two ablaut grades (zero vs e-grade), not two roots meeting. (2) Dropped the traust/trust branch: Wiktionary derives trust from Old Norse traust, a Norse loan into English, not a native Old English descent alongside treowe/treowth/treow -- the branch asserted a line the tree could not support. source_word re-keyed 'trust'->'true' since trust is no longer in the tree. Combined effect nodes 11->8, one node below the nine forecast when the branch drop was approved on its own; maxFanout 3 so meetsExtremeThreshold still passes. Confluence node keeps parentId=root-0. Three chips removed.

removed revealed-form chip(s) from options: "*trewwaz" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "theos" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*déḱm̥" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*h₃dent-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "sarx" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*kʷetwóres" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "harena" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*h₃rey-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "vacca" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Sibling stacking across a language boundary: the tree wires tribunal as a CHILD of the Modern English word tribune. English tribunal is borrowed from Latin tribunal (the raised platform), and English tribune from Latin tribunus -- both Latin words derive in parallel from tribus, so the two English words are cousins, not parent and child. The structural sweep also flagged tribunus/tribuere/tributum under tribus in the same row; those three ARE genuine -tas/-ere/-tum derivations from tribus and are not defects. Only the tribunal<tribune edge is wrong. Row originates in the 2026-05-21T23:34:40.244Z seed insert (generated=0).

removed revealed-form chip(s) from options: "tribus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Deleted the council node. The tree hung council directly off PIE *kelh1- alongside the calendar family; Wiktionary derives council from conseil < consilium < consulo < PIE *selh1- (take, grab), with no connection to calare or kalendae. No legitimate calendar->council link exists.

removed revealed-form chip(s) from options: "kalendae" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "village [OF]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "villa" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*sal-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

Layer 1 fix: guest-family chain — inserted missing Middle English (gest) intermediate between guest and gestr

Layer 1 fix: inserted missing hostilis and Middle French hostile intermediates between hostis and hostile

removed revealed-form chip(s) from options: "*ghos-ti-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "Slověninŭ" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "franchise [OF]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "Frank" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "īre" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "τηλεσκόπος" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "*orbh-" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "via" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "tragōs" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "sincerus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "candēre" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "sal" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "clocc" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "qanû" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "musculus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "discus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "mus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "lapis" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "luna" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "gala" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "ripa" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "cappa" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "magazine [EME]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "magasin" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "sucre candi" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "coton" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "kahve" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "admiral [MidE]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "amiral" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "assassin [MFr]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "assassino" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "algorismus" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed fabricated undocumented tag chip(s) "paradis [MidE]", "paradis [OFr]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed revealed-form chip(s) from options: "paradeisos" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "zucchero" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "dæġes ēage" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "hlǣfdīġe" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

revealed (payoff) node had empty meaning; added: "loaf-ward — the keeper of the bread; head of a household". Old English hlaf (loaf) + weard (guardian/ward)

removed revealed-form chip(s) from options: "hlāford" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "godsibb" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed revealed-form chip(s) from options: "vindauga" x1 -- identical to an already-revealed node's own form, a chip the player can never legitimately place (already shown on the board). corpus-debt Part 2, 2026-09-01.

removed orphaned disconnected node n5, corrected n1 parentId to n2, trimmed redundant papir chip

removed fabricated undocumented tag chip(s) "papyrus [L]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

because the ETYMOLOGY IS WRONG, not because of any similarity score. The tree hangs saboteur (Modern English) as a child of sabotage (Modern English). Wiktionary gives saboteur as {{bor|en|fr|saboteur}} -- borrowed from French, not derived from the English noun. In French, saboteur and sabotage are parallel derivations of saboter, so the two English words are siblings borrowed separately, not parent and child. Same sibling-stacking shape as cursor/cursus in xb5 and recipere/accipere in xcomp2. NOTE the 1.00 adjacent-similarity pair that drew attention to this row (sabotage ModE <- sabotage French) is NOT the defect -- that is a legitimate unchanged borrowing; the metric pointed at the row and the fault was a different edge. Not scheduled on any date, so no daily_schedule slot is affected.

removed fabricated undocumented tag chip(s) "sabotage [Fr]", "sabot [Fr]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

removed fabricated undocumented tag chip(s) "alcohol [ML]" -- each duplicates the spelling of an already-correctly-needed bare chip, corresponds to no real chain node, and has no impostor_notes explanation (unlike the deliberate documented decoys found elsewhere in the same sweep). Found in corpus-wide tagged-chip recurrence check, 2026-09-01.

options had three near-identical hasard chips: hasard [MidE], hasard [OFr], and bare hasard. Chain data shows the Middle English and Old French nodes literally share the same form (hasard, no real spelling distinction) -- the bracket tags on two of the three were fabricated disambiguation with nothing behind them. Only one hidden slot needs hasard (the OFr node is already revealed). Removed both tagged duplicates, kept the one bare chip that matches the data. Found during live playtest 2026-09-01.

fabricated confluence, clearest case in the cohort: node confluence1 (virtus, Latin) claimed as descending from three parents including arete (Greek). Virtus and arete are conceptually parallel (both mean excellence) but there is no attested linguistic descent from Greek into this Latin word -- conceptual similarity presented as etymological ancestry. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

fabricated confluence: node confluence1 (paradeisos, Greek) claimed as descending from both pairi-daeza (Old Iranian/Avestan) and pardes (Hebrew/Aramaic) at once. Standard etymology treats these as separate loans from a common ultimate source, not a convergence through Greek. Duplicates xc4 (also paradise) which tells the real three-branch story correctly. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

mislinearization: node confluence1 (message [OFr]) has mittere and missus (Latin) as co-parents, but missus is literally the past-participle form of mittere -- same lemma, not two separate ancestors. Should be a linear mittere -> missus -> message chain, not a two-parent confluence. Milder than the other six structural findings in this cohort but the same underlying pattern. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

fabricated confluence: node confluence1 (animus, Latin -- breath/spirit, real root PIE *h2enh1-) claimed as also descending from *men- (the unrelated to-think family behind mental/mind). Cross-connects two different PIE root families. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

Retired, not repaired. The shared root is sound -- Wiktionary derives Greek tele (afar) from a locative of PIE *kwel- and explicitly says 'Compare telos (end)' -- but almost everything hung off it is wrong. (1) *tele- is not a PIE root; it is Greek tele, entered as a second revealed root. (2) telescope descends from tele as a SIBLING of telos, not through it (Wiktionary: telescopium < telescopos < tele + skopeo), yet the tree routes the telescope and telephone branches through the telos confluence node. (3) tele + phone has no Greek ancestor at all: telephone is from French telephone, affix tele- + -phone, a 19th-century coinage. Repair would mean deleting the root, revealing *kwel-, inserting a tele node and re-parenting two branches onto it -- authoring a new puzzle inside a broken row's shell rather than fixing this one.

fabricated confluence plus formula nodes: node conf (telos, Greek) claimed as child of two unrelated PIE roots (*kwel- and *tele-). Two branch nodes are compound formulas, not lexical forms -- tele + skopein and tele + phone -- with placeholder language values (Greek components, Coined) rather than real language names. The worst-structured row in the cohort. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

Layer 1 fix: relabeled ponton node from French to Middle French per Wiktionary

fabricated confluence: node conf (pons, Latin) given its real parent (*pent-, the standard root of pons/bridge) plus a fabricated second parent (*per-) with no attested connection. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

Retired, not repaired. Three independent defects. (1) Fabricated second root *ghew- (to pour) wired into the *dumbaz confluence via children[] and placeable as a chip: Wiktionary derives dumb via *dumbaz from *dhewbh-, and *ghew- appears nowhere in the line. (2) *daubaz (deaf) is a SIBLING of *dumbaz -- both descend from *dhewbh- directly -- but the tree hangs deaf beneath dumb. (3) The source_word's own branch is unattested from the tree's own root: Wiktionary gives dump as ME dumpen, 'probably from Old Norse dumpa, of uncertain origin, possibly imitative', with no descent from *dumbaz. Deleting the fabricated root would leave two wrong edges and a source_word not reachable from its root, so there is no spine to repair. -- it was never re-verified.

fabricated confluence: node conf (*dumbaz, Proto-Germanic) claimed as child by two unrelated PIE roots at once (*dhewbh- and *ghew-). A real ancestor lineage is singular; this presents unresolved scholarly alternatives as simultaneous descent. Found in seed-cohort audit 2026-09-01, replacing bare manual token with this note.

revealed (payoff) node had empty meaning; added: "a crowd, the common people". not currently servable, fixed for content quality

revealed (payoff) node had empty meaning; added: "a young squire or junior vassal not yet holding his own fief — origin disputed". not currently servable, fixed for content quality; etymology genuinely contested, gloss hedged accordingly

revealed (payoff) node had empty meaning; added: "woman of Magdala — the given name of Mary Magdalene". not currently servable, fixed for content quality

revealed (payoff) node had empty meaning; added: "witness". not currently servable, fixed for content quality

revealed (payoff) node had empty meaning; added: "to give birth, to beget". not currently servable, fixed for content quality

revealed (payoff) node had empty meaning; added: "a stringing together of words; talk, discourse — from PIE *ser- "to join, link"". not currently servable (needs_review already set), fixed for content quality

revealed (payoff) node had empty meaning; added: stomachos gloss (throat, gullet - from stoma mouth + -achos)

revealed (payoff) node had empty meaning; added: "to give birth to a child". not currently servable (needs_review already set), fixed for content quality

revealed (payoff) node had empty meaning; added: "a buying back; a ransoming, redemption". not currently servable (needs_review already set), fixed for content quality

revealed (payoff) node had empty meaning; added: "star". PIE root of astrum/astron; both descendant nodes already glossed "star", the root itself was empty

meaning was a tautological echo of form ("lemon"/"lemon"); replaced with a real gloss

added Latin transliteration alongside untransliterated Greek in the kybernetes node meaning, per corpus sweep finding

Deleted. The elixir path hides elixir[Modern English] and elixir [ML][Medieval Latin] -- two distinct RAW strings that both render as elixir once the editorial tag is stripped, so the pool needs two identical chips and either satisfies either slot. The repeated-hidden-form rule initially compared raw forms and passed it; it now compares stripped forms, which is what the player sees. Editorial tags disambiguate nodes inside a parent tree and do not survive into a derived row, so they must not be trusted to distinguish forms. Row was inert and unscheduled.

UNSOLVABLE as enriched: the chain hides captain/capitain/capitaine/capitaneus/caput but the generated chip pool is captain, capitain, capitaine, caput, capitale, chef, capitan -- there is NO chip for capitaneus. isEasySolvable would discard it at serve time, so its scheduled date would have fallen through to the DAILY_PATTERN rotation and silently served a different puzzle. Verification passed (all 5 links attested); the defect is in the enrichment-generated options pool, which no gate checked before insert. Schedule 2026-12-14 re-pointed to e20 (tragedy, corrected, single prior slot 2026-06-18) so the date is not left resolving through the fallback. Row produced by the supervised first production run 2026-08-26.

Retired. The sincere chain presents Latin sincerus as descending from PIE *sem-, which the adversarial pass flagged and which does not hold: the derivation is genuinely disputed. The sine cera (without wax) story is folk etymology; the *sem-kero- compound from *sem- is one hypothesis among several (others connect it to *segh- or to Oscan-Umbrian cognates) and is not settled. Same class as e42 (sterling) and e40 (carnival): where the literature has no winner, no single-node edit makes the chain correct, so the row is withdrawn rather than repaired. SCHEDULE: this row held a FUTURE slot on 2026-09-17, which retiring it would have converted into a silent DAILY_PATTERN fallback -- the e41 failure mode. That date has been repointed to e4 (jungle) in the same write batch. ; it was never re-verified.

Deleted the formula node 'niht + mare' (Old English compound) and relabelled its child from nihtmara to *nihtmare, promoting it to revealed root. The row carried TWO consecutive Old English nodes and neither matched the source: Wiktionary gives nightmare < ME nyghtmare < OE *nihtmare, asterisked, plus a compound gloss of night + mare. nihtmara appears nowhere in that line, so it was a fabricated intermediate rather than a spelling variant, and the fix is a deletion plus a relabel rather than a straight replacement. The asterisk is retained deliberately: *nihtmare is an attested reconstruction stated via {{inh}}, handled the same way proto forms are elsewhere in the corpus. Chain 4->3; nihtmara removed from the chip pool. NOTE for follow-up: the surviving Middle English node reads nihtmare, which now differs from the root only by the asterisk and which Wiktionary spells nyghtmare -- flagged, not changed, as it was outside the approved edit.

Corrected the Middle English node from nihtmare to nyghtmare, closing a loose end flagged when this row was corrected on 2026-08-24. Wiktionary: nightmare < ME nyghtmare < OE *nihtmare. The previous spelling made the ME node differ from the Old English root only by the asterisk, which is also the entire explanation for this row scoring 0.94 on mean-path adjacent similarity -- audited for sibling-stacking and there is none; the two nodes are a genuine parent-child pair, one of them simply mis-spelled. Chip pool entry updated nihtmare -> nyghtmare. Only scheduled date is 2026-07-16, in the past. corrected.

Deleted the formula node 'dæg + ege' (Old English compound) and relabelled its child from dægesege to the attested dæġes ēage, promoting it to revealed root. A straight relabel of the formula node was not possible: dægesege IS dæġes ēage without diacritics, so the chain already carried the target form and the replacement would have duplicated it. Wiktionary: daisy < ME dayesye < OE dæġes ēage (day's eye), the flowers closing at night. Chain 4->3; dægesege removed from the chip pool now that the node is revealed.

Retired as the DUPLICATE of the daisy pair, in place of ge20260518-6 which was the instructed target. Two premises behind that instruction turned out not to hold. (1) The chains are no longer different: ge20260518-6 was generated with the correct dæġes ēage root on 2026-05-18, and today's correction to e41 (replacing the formula node dæg + ege) converged e41 onto exactly the same three nodes -- so ge20260518-6 was never serving an uncorrected version; it was right first. (2) ge20260518-6 holds a FUTURE daily_schedule slot (2026-12-12) whereas e41's only slot (2026-07-27) is past, so retiring ge20260518-6 would have silently converted a scheduled date into a DAILY_PATTERN fallback -- the documented failure mode. ge20260518-6 also carries the richer enrichment (origin_story 342 chars vs 210). Retiring e41 satisfies the same intent (one daisy puzzle, the correct chain) without breaking a schedule slot or discarding the better row. Trivially reversible if the other choice is preferred.

Deleted fabricated second root *kurso- (id root-1), wired into the cursus confluence via children[] and a placeable chip. *kurso- is not a PIE root; it is the Latin o-grade stem of curro, itself descended from the revealed root *kers-. A derived stem presented as a co-ordinate ancestor. Confluence node keeps parentId=root-0; nodes 9->8; maxFanout 3 so extreme threshold still met. Chip removed.

Sibling stacking: the tree wires cursor as a CHILD of cursus. Both are Latin derivations of currere -- cursor is the agent noun (one who runs), cursus the fourth-declension verbal noun (a running) -- so they are siblings formed in parallel from the same verb, and neither descends from the other. Same defect shape as the *daubaz/*dumbaz edge that retired xd1 and the recipere/accipere edge that withheld xcomp2. Note this row already received a content-fix earlier this week (fabricated second root *kurso- deleted); that correction was sound and is unaffected -- this is a separate, second defect in the same hand-seeded row, found by the structural sweep rather than the convergence sweep. Row originates in the 2026-05-18T00:54:54.245Z bulk seed insert (generated=0).

Layer 1 fix: guest-family chain — inserted missing Middle English (gest) and Old Norse (gestr) intermediates between *gastiz and guest

pending audit, not a decided defect. Flagged by the sibling-stacking detector: the chain wires hospes as a CHILD of hostis. Both descend from PIE *ghos-ti- (stranger, guest) as parallel Latin outcomes -- hospes via *hosti-potis (guest-master) -- so hospes is a sibling of hostis, not its descendant. Same shape as the xd1 edge. Row originates in the 2026-05-18T00:54:54.245Z bulk seed insert (generated=0). Withheld pending Part 3 audit;

pending audit, not a decided defect. Flagged by the sibling-stacking detector: the chain wires recipere as a CHILD of accipere. Both are Latin prefixed compounds of capere (re- + capere, ad- + capere) and are therefore siblings -- neither derives from the other -- which is the same shape as the *daubaz/*dumbaz edge that retired xd1. Row originates in the 2026-05-18T00:54:54.245Z bulk seed insert (generated=0), the source of every defect found this week. Withheld rather than corrected because the disposition is Part 3 audit output, not a spot fix.

Deleted the formula node 'pre + scribere' (Latin compound), relabelled its child from prescribere to praescribere and re-parented it directly onto the *skribh- root. A straight relabel was not possible: prescribere already sat in the chain as the formula node's child, so replacing the parent with praescribere would have created two nodes for one Latin verb. Wiktionary: prescribe borrowed from Latin praescribere, from prae- (before) + scribere (to write) -- the ae spelling is the lemma. Nodes 6->5; chip pool lost 'pre + scribere' and its prescribere entry was rewritten to praescribere.

Replaced the formula node 'manu + factura' (Latin) with the attested Latin manufactus. The originally proposed target, manufactura, was NOT used because it already sits in this chain as the Medieval Latin node one level up -- relabelling to it would have produced two nodes for one word. Wiktionary gives a real form one level deeper: manufacture < Middle French manufacture < Medieval Latin manufactura, from manufactus, a compound of manu (ablative of manus, hand) and factus (past participle of facio). manufactus is therefore both attested and distinct from its child. Revealed root, so no chip changed. Nodes unchanged at 4.

Three corrections in one write. (1) Relabelled the revealed root from *peh2- to *peh2k-: Wiktionary derives Latin pax from Proto-Italic *paks, from PIE *peh2k-s, root *peh2k- (to join, to attach). *peh2- (to protect) is a different root and gave pastor, not pax. (2) Deleted the second root *pag-, which was wired into the pax confluence via children[] and placeable as a chip. It was NOT the same root in older notation: *pag-/*peh2g- is the sibling variant giving pangere, pagan and compact, not pax. (3) Deleted the formula node 'ad + pax' and re-parented apaisier onto pais. No Latin or Vulgar Latin verb stands in this line: Wiktionary gives appease < ME apesen < Old French apeser, and the Old French entry apaisier is confixed a + pais + ier -- so apaisier descends from pais, which was already present in the tree. An invented *appacare was avoided in favour of the attested parent. Nodes 12->10, single root, hard tier. Two chips removed.

Deleted the formula node 'pes + grus' (labelled Latin) and re-parented its child directly onto the *ped- root. No Latin compound of pes and grus is attested -- they are two separate Latin words -- so there was no form to replace it with. The surviving child was checked rather than promoted as-is, per instruction: Wiktionary gives pedigree as 'From Anglo-Norman pe de grue, a variant of Old French pie de gru (foot of a crane), from Latin pes + grus'. The direct ancestor of pedigree is therefore the Anglo-Norman form, so the node was relabelled from 'pie de grue' (Old French) -- a spelling matching neither attested variant -- to 'pe de grue' with language Anglo-Norman. Nodes 7->6; chip pool lost 'pes + grus' and its 'pie de grue' entry was rewritten.

Replaced the formula node 'mort + gage' (Old French compound) with the attested Old French phrase 'mort gage'. Wiktionary: mortgage < ME morgage and Middle French mortgage, from Anglo-Norman morgage, from Old French mort gage (dead pledge) -- rendered there as the two-word phrase, which is what this node now carries. Chip pool entry updated. Node count unchanged at 10.

Replaced the formula node 'e + ducere' (Latin compound) with the attested Latin verb educere. Wiktionary derives educate from Latin educatus, perfect passive participle of educo (educare), 'further from an intensive/frequentative formed on e- + duco' -- so educere is the attested compound standing between *deuk- and educare, and is a distinct lemma from its child educare rather than a restatement of it. Chip pool entry updated from 'e + ducere' to educere. Node count unchanged at 7.

Replaced the formula node 'ambi + ire' (Latin compound) with the attested Latin verb ambire. Wiktionary: ambition < ME ambicioun < OF ambition < Latin ambitio, from ambio (to go around, solicit votes). ambire is the infinitive of that verb and is distinct from its child ambitio, so no duplicate is created. Infinitive form chosen to match house style elsewhere in the corpus (scribere, ducere, pacificare). Revealed root, so no chip changed. Node count unchanged at 4.

Replaced the formula node 'tragos + oide' (Greek) with the attested Ancient Greek compound tragoidia. Wiktionary: tragedy < ME tragedie < OF tragedie < Latin tragoedia < Ancient Greek tragoidia, from tragos (male goat) + oide (song). The node is the revealed root, so no chip changed. Greek script matches the ten existing Ancient Greek forms already in the corpus. Node count unchanged at 4.

Deleted the formula node 'ban + fyr' (Middle English compound) and promoted banefire to revealed root. Same duplicate-in-disguise as e11: banefire IS the compound of ban + fyr, so the formula node restated its own child. Wiktionary lists bonefire, beane fyre and bone fyre among the obsolete alternative forms, so the surviving spelling is attested. Chain 4->3; banefire removed from the chip pool now that it is revealed.

Deleted the formula node 'tres + palus' (Latin) and promoted trepalium to revealed root. No replacement was possible because the formula IS its own child: trepalium is the attested Late Latin compound of tres + palus, so the node duplicated the one beneath it. Wiktionary carries travail only as far as Old French travail and marks the tripalium account as a possibility rather than a settled derivation, which the chain still reflects honestly at three nodes. Chain 4->3; trepalium removed from the chip pool now that it is revealed.

Retired, not repaired. The revealed root 'caro + vale' (flesh + farewell) is the folk etymology, not a derivation. Wiktionary: carnival < Middle French carnaval < Italian carnevale, 'possibly' from the Latin phrase carnem levamen (meat dismissal), with two further competing proposals named (carnualia; carrus navalis) and no winner picked. Same shape as e42/sterling: the answer is genuinely contested in the literature, so no single-node edit can make the chain correct. Withdrawn from service instead.

Retired, not fixed in place. The sterling chain asserts a single settled etymology that does not exist: the derivation of sterling is genuinely disputed -- the Easterling/Hanseatic-trader account, the small-star (steorra) account, and the starling-on-the-coin account are all live in the literature and Wiktionary does not pick a winner. A puzzle whose answer is contested cannot be made correct by editing one node, so there is nothing to repair; the chain is withdrawn from service instead. removes it from all four servable gates. -- it was never re-verified, and marking it 'corrected' would misstate what happened.

Layer 3 fix: added folk-etymology hedge to salary origin_story (Roman-soldiers-paid-in-salt story is disputed by modern historians), matching e1/x1/e87 convention

Deleted fabricated second root 'salarium (practice)' (id root-1), wired into the salarium confluence via children[] and a placeable chip. The parenthetical was an editorial gloss about the salt-ration story, not a headword or an etymon -- the tree had salarium converging with a node that was literally itself plus a comment. Confluence node keeps parentId=root-0 (*sal-); nodes 11->10; hard tier, no extreme threshold to meet. Chip removed.

Deleted fabricated second root *lewk- (id root-1), wired into the lux confluence via children[] and a placeable chip. *lewk- and the revealed *leuk- are the SAME PIE root -- PIE notation writes the glide as either u or w. This pair initially escaped the convergence sweep for exactly that reason and was only caught after folding w->u in the comparison. Confluence node keeps parentId=root-0; nodes 9->8; maxFanout 3 so extreme threshold still met. Chip removed.

Deleted fabricated second root *gneh3- (id root-1), which was wired into the natura confluence node via its children array and was a placeable chip. *gneh3- (to know) and the revealed root *genh1- (to beget) have the same consonant frame and differ only by ablaut grade -- one root written twice, not two roots meeting. natura descends from *genh1-. Confluence node keeps parentId=root-0 (valid); tree stays single-rooted. Node count 8->7, maxFanout 3 so meetsExtremeThreshold still passes independently of the removed confluence. Chip removed from options.

source_word was "doctor [mode]" -- stray editorial bracket annotation. Set to "doctor". Metadata only, no player-visible change.

source_word was the literal string "hb15" -- a data-entry bug, not a headword. Set to "regal", the leaf of the first branch of this *reg- family, matching the sibling convention (hb7=law, hb11=mead, hb13=bane, hb14=audio). Noted separately: this tree has no Modern English node at all, its leaves are Old French regal/direct, which is unusual for the hb series but out of scope here.

Dropped the auctor/auctour/author branch. The tree claimed PIE *au- as the common root of both audio and author; Wiktionary derives auctor from augeo (to increase) < PIE *h2ewg-, explicitly not from a root meaning to hear. Surface phonetic coincidence only.

Dropped the homicidium/homicide branch. The tree claimed PIE *gwhen- as the common root of both bane and homicide; Wiktionary derives homicidium from caedo < PIE *keh2id- (cut, hew), a different root from *gwhen- (strike, slay).

Dropped the mad branch and replaced the malformed root. The root node was the literal string "skr. मद् (mad-)" with language Sanskrit -- an editorial citation, not a headword -- and it was claimed as ancestor of both mead and mad. Wiktionary derives mad from OE gemaeded < PGmc *maidijana < PIE *mey- (change), unrelated to *medhu- (honey/mead). Root replaced with a properly formed PIE *medhu.

Dropped the lectura/lecture branch. The tree claimed PIE *legh- as the common root of both law and lecture; Wiktionary derives lecture from lectura < lego < PIE *leg- (gather, read), a different root from *legh- (lie down), which gives law/lay/lair. A false-cognate pairing.

Deleted the node "exploser" (claimed Old French) from the explode chain. Wiktionary: "First recorded around 1538, from the Latin verb explodere" -- no Old French intermediary, and exploser is not attested as Old French. Chain is now explode <- explodere <- plaudere.

Deleted fabricated PIE root *yewg- from the jungle chain; jangala (Sanskrit) promoted to revealed root and removed from the chip pool. Wiktionary gives jangala a substrate (Dravidian or Munda) origin, explicitly not Indo-European, so no PIE root exists to claim. Linear twin of the x4 defect corrected on 2026-08-24.

Removed vandalisme (1794) as a false intermediate in the vandal branch -- Wiktionary dates English vandal (destroyer sense) to the 1660s, predating and not derived from the French coinage. Corrected vandal era to 1660s-present. Retained vandalisme in the pool as a crossover impostor with a new note. Also removed this rows own pre-revealed roots (barbaros, *Wandalaz) from options -- dead chips that trimPool already stripped at serve time., Lanes now 4/3.

Dropped fabricated PIE root *yewg- from the jungle branch; jangala (Sanskrit) promoted to revealed root and removed from the chip pool. Wiktionary gives jangala a substrate (Dravidian or Munda) origin, not Indo-European. Same defect class as e4., Lanes now 4/3.

pool column was missing macron-accented chip forms for two unrevealed nodes -- had liber (bare) and deliberare (bare, no macrons) but the chain needs exact-match chips liberare and deliberare with macrons for nodes b3-x/b3-0. isTreeSolvable treats macron/non-macron as different strings, so the puzzle was unsolvable. A prior bundle-only patch apparently fixed this once but was lost when the bundle got resynced from D1 -- fixing D1 directly this time so it cannot recur. Added the two correctly-accented forms to pool (kept the 3 existing impostors library/libra/libation). Scheduled for 2026-10-02.

corrected cappella language+era Late Latin(200-700) -> Medieval Latin(600-1500) (Wiktionary: Early Medieval Latin), updated matching enrichment text

Layer 3 fix: added folk-etymology hedge to salary origin_story (Roman-soldiers-paid-in-salt story is disputed by modern historians), matching e1/x1/e87 convention

Layer 3 fix: added folk-etymology hedge to salary origin_story (Roman-soldiers-paid-in-salt story is disputed by modern historians), matching e1/x1/e87 convention

Layer 3 fix: added folk-etymology hedge to salary origin_story (Roman-soldiers-paid-in-salt story is disputed by modern historians), matching e1/x1/e87 convention

Layer 3 fix: added folk-etymology hedge to salary origin_story (Roman-soldiers-paid-in-salt story is disputed by modern historians), matching e1/x1/e87 convention

Layer 1 fix: inserted missing French gynécologie intermediate between gynē and gynecology

Layer 3 fix: corrected origin_story claim that gynecology was coined directly from Greek -- actual chain is via French gynécologie per Layer 1 fix

Layer 1 fix: guest-family chain — inserted missing Old Norse (gestr) intermediate between *gastiz and gest; relabeled existing gest node from Old Norse to Middle English

Layer 3 fix: rewrote hook/origin_story/meaning_drift to match actual chain content (removed references to host/hostile which are not part of this puzzle)