The tree, stage D4

Begun at D4a-01, 2026-10-24 (UTC). The taxonomy is read off the adjudicated ledger: the current ledger (the re-check, for all 705 rooms), whose lines were adapted against the model frozen at D3 (46384686…). Every count here comes from a script named beside it.

The model the tree hangs from, and two held proposals

The conceptual root is the global model as frozen at D3 (46384686d6173fdeb4c749667478c1363e675bc6b52bb6f4fdc211d387e68525), and the model in out/model/ is that version. At D4a-01, on J's instruction after the D3c interview, two amendments were made under rule 11 (moves 198 and 199; history lines 173 to 180). The D4 close-check refuses any model move after the D3 freeze (stage_D4.py, vlib.model_moved with no allowance). I asked (QD4-001), and J chose option (b): the same leg reverted both (move 203; history lines 181 and 182), and the model hash is 46384686 again. Both are held proposals: never served to any room, not re-checked, and not part of the model this tree hangs from. They are stated in full in "The held proposals" below. The distances do not read element text or status (adaptation_ledger.shared_lines reads the ledger and its rulings only), so neither the amendments nor the revert changed any distance, key or member here.

The statistic and the frequencies, read before any cut

• Tree-likeness. The delta score on the combined level-3 matrix is 0.4227 (Monte Carlo SE 0.0020; 20,000 nondegenerate quartets, seed 0), over 705 rooms and 26,487 shared lines (_views/tree-likeness.json). The weights (4, 2, 1, 0.5) and grades (1, 0.5, 0, -1) are prior.md section 7's, confirmed. Larger deltas mean less additive distances. The dendrogram is drawn (_views/dendrogram.nwk), but it is a chain: at 12 clusters one group holds 663 rooms, and the others are single rooms or small groups peeling off it (_views/dendrogram.txt). The dendrogram is a view of the dissimilarities, not a chronology and not a validated hierarchy, and it shows no partition of the rooms. The NeighborNet is drawn outside the run, if asked.
• Key frequencies (_analysis/d4a01_key_freq.py). At level 3 there are 5,237 keys: 2,772 held by one room, 1,509 by 2 to 4, 720 by 5 to 19, 208 by 20 to 99, 28 by 100 or more. Every room holds a key, and no room returned none. Corrections: 17 keys in 18 rooms, none held by more than 2 rooms below the bare target. Elaborations: 545 keys in 497 rooms. Specifications: 4,931 keys in all 705 rooms.
• The strongest signal is nearly empty. Shared corrections, the stage's strongest form signal, reach 5 rooms only at the bare target (restrict on P05 in 7 rooms, restrict on R01 in 5). The D3 revisions absorbed the first-return corrections, so the current ledger's groups rest on elaborations and specifications.

The sensitivity check (J after the D3c interview)

The kind of add-condition, add-step and restrict lines is not relied on (the blind check's kappa on them was 0.08). The distance was computed once more with one kind for those three ops (_analysis/d4a01_kind_collapsed.py). The combined distance hardly moves: Pearson 0.9965 and Spearman 0.9967 over 248,160 room pairs, mean absolute difference 0.0024, delta 0.4223 against 0.4227. The average-linkage trees agree at 2 and 3 clusters (ARI 1.00), at 5 (0.886) and 8 (0.922), and less at 12 (0.779). The per-kind views do depend on the kind: for add-step and add-condition the specification lines outnumber the elaboration lines about 2 to 1 and 6 to 1. So no branch takes its members from a per-kind view for those ops; keys are read in the all-kinds matrices.

The rules of the cuts, written before the first cut

• The minimum group size is 5, the default. prior.md sets none. I keep it, since 720 level-3 keys are held by 5 to 19 rooms, and a higher floor would drop most change keys before their lines are read.
• What keys a branch. An element's change key (op add-step, add-condition, link, restrict or contradict), read in the all-kinds matrix, at the deepest ladder level where 5 or more rooms meet on it. Arrow keys cannot key a branch, since the check resolves a key's target against the elements; they are context. Instantiate and absent keys specify the model as it stands and change nothing, so they define no branch.
• When a key becomes a branch. Only when the members' lines, read whole through _analysis/d4_key_lines.py, state one shared change to the element: a step, a condition or a link that the statement does not hold, or holds only in part. Otherwise the group is set aside below, with its reason.
• Subforms. A level-3 child key held by 5 or more of a branch's members, whose lines state a narrower change than the parent's.
• Corrections that meet only at the dimension or the bare target are groupings, not branches: P05's shape read and R01's joining (below).
• Cross-links. A room in two branches that do not nest is declared in crosslinks.jsonl by script, with both keys as the reason.
• Order. By element, from the largest change-key group at level 4 downward. A leg ends at a safe stop after a logged move.
• Ground. The keys, unless a cut's tier_history.jsonl line declares the members' material. The branch's kind field is the majority kind of its members' lines on the key, reported and not relied on.
• Rule 7. I hold every room's canvas outcome and cannot unknow it. No cut reads the canvas, the verdicts or the projections, and each cut is snapped before anything is projected onto it.

The cuts of D4a-01, in the order cut

• F01 (move 200, merging; branch named; level fourth; parent none). Keys: add-step:K01:T4-001. 103 members. Kind: elaboration (specification 60, elaboration 104), not relied on. Ground: keys. the play administered: steps that test, police, end and convert the April Fools frame. K01 states that the host sub is keyed as play for April Fools and that a claim in it may be a joke. In these rooms the frame gains steps that the statement does not hold. Members test whether a claim is a joke by its date, source or link. They police the licence's limits and argue about when it ends. Moderators enforce a takeover's rules by real removals and switch its styling off. A joke can turn into a sincere request or a lasting change. The steps belong to the sub's own play beside the canvas, not to canvas work.
• F02 (move 201, splitting; subform declared; level third; parent F01). Keys: add-step:K01:T3-023. 33 members. Kind: elaboration (elaboration 23, specification 20), not relied on. Ground: keys. the takeover run as a temporary regime. Within F01, a takeover's play becomes a temporary regime: its rules are enforced by real removals, bans and bots, members perform and test them, its styling is made usable or switched off, and its end is set.
• F03 (move 202, splitting; subform declared; level third; parent F01). Keys: add-step:K01:T3-095. 18 members. Kind: specification (specification 12, elaboration 6), not relied on. Ground: keys. the joke tested: members check a claim's truth. Within F01, members test whether a claim is a joke before they act on it: by its date, its source, its plausibility or its marks, and they repair a literal reading.
• F04 (move 204, splitting; subform declared; level third; parent F01). Keys: add-step:K01:T3-032. 13 members. Kind: elaboration (elaboration 15), not relied on. Ground: keys. the licence policed: objections to the play, its limits and its end. Within F01, members object to the play and police the licence: they set limits on taste and repetition, demand labels, deletion, restoration or better jokes, and dispute when the licence ends.
• F05 (move 205, splitting; subform declared; level third; parent F01). Keys: add-step:K01:T3-097. 12 members. Kind: elaboration (elaboration 13), not relied on. Ground: keys. the styling ended: ordinary use restored. Within F01, members and moderators end the prank's styling and restore ordinary use: they switch styles off, ask whether the sub has switched back, and clear what the prank left.
• F06 (move 206, splitting; subform declared; level third; parent F01). Keys: add-step:K01:T3-100. 8 members. Kind: elaboration (elaboration 8), not relied on. Ground: keys. the joke kept: a request that its change last. Within F01, members ask that a joke's change outlast the day: a temporary feature, mode, flair, banner or board is to become permanent, or a joke activity real.

Members are listed in branches.jsonl and in each forms/<id>.md, with every member line's how sentence, the elements the members carry unchanged, one representative concept per member with its ladder, and the mention graph among members. The branches are presented in the order cut, never by size.

Cross-links (move 207, relating; crosslinks.jsonl, 18 lines): F02 and F03: dankmemes, ontario, terraria, warframe, zelda; F02 and F04: bindingofisaac, dankmemes, monstercat; F02 and F05: ontario, schoolidolfestival; F03 and F04: AMD, canucks, dankmemes, magictcg; F03 and F05: minecraft, ontario; F04 and F05: anime_irl, nbaspurs. F06 shares no room with another subform. Every subform nests in F01, so a room in F01 and one subform is not a cross-link.

Groups read and set aside

• Two level-3 groups under F01. add-step on K01 at T3-001 (staging April Fools fake news, 37 rooms) and at T3-048 (reacting to official April Fools, 14 rooms). Their lines state several different steps (testing, an apology, a fake turned into serious talk, financial warnings, troubleshooting), so no narrower shared change holds them. Their rooms stay in F01.
• P05's shape read and R01's joining (cut rule R4). The re-check's corrections meet only at the bare target (restrict on P05 in 7 rooms and add-step on P05 in 2; restrict on R01 in 5). There the aggregation does the work, so they are groupings, not branches. A key holds the op, the target and the node, not the condition, so the keys cannot carry what separates the shape read from the trunk. D3's branch stays a held branch, read from the rooms' material (F-024), not a key branch.
• Arrow keys. The largest change keys are restrict lines on arrows: A04 at T3-010 (87 rooms), A11 at T3-004 (77), A05 at T3-010 (59), A13 at T3-010 (53). A branch cannot be keyed on an arrow (cut rule R1). They are context: most were ruled specifications that the arrows' bounds already state.
• Co-occurring keys. Among the change keys held by 5 rooms or more, the keys their rooms share beyond chance are mostly on the same third-order node (one theme lifting to several elements), so they are not independent definitions (_analysis/d4a01_candidates.txt, section B).

Where the pass stopped

J stops the run after this leg (QD4-001). The pass cut one element, K01, the largest element change key at level 4. These element change keys at level 4, held by 20 rooms or more, were not read for a cut: restrict:P05:T4-006 59; add-step:P14:T4-009 58; link:P03:T4-014 54; add-step:P08:T4-016 47; add-step:P13:T4-028 43; add-condition:P05:T4-013 43; add-step:P04:T4-008 42; add-step:P02:T4-006 41; add-step:P05:T4-013 40; add-step:P23:T4-010 39; add-step:P20:T4-002 36; add-step:P11:T4-026 36; add-step:P12:T4-027 33; restrict:K01:T4-001 32; add-condition:K01:T4-001 31; add-condition:P16:T4-017 30; add-condition:P13:T4-028 30; add-condition:P08:T4-016 29; add-condition:P14:T4-009 28; restrict:P14:T4-009 27; restrict:P13:T4-028 27; restrict:P05:T4-016 27; restrict:P05:T4-013 27; add-condition:P16:T4-016 27; add-condition:P23:T4-010 26; restrict:P23:T4-010 24; restrict:P19:T4-016 23; link:P05:T4-026 23; add-step:P04:T4-006 23; add-condition:P02:T4-006 22; add-step:P01:T4-005 20. Each would be read as F01 was, by cut rules R1 to R3. No comparison (section 5) and no projections (section 6) were built, since they come after the final cut, and this pass is not the final one.

The held proposals (QD4-001, option b)

Both were made at D4a-01 and reverted in the same leg. Neither was served to any room, and neither is re-checked. The model this tree hangs from holds P05 and A11 as frozen at D3.

• P05, move 198 (contrasting), history lines 173 to 176, reverted by line 181 (move 203). The statement would gain: "This holds where the keepers can learn of a planned change only by being told, and no informed member holds it." The status would be bounded. The bound would read: "It does not hold where the keepers read a planned change from its shape and stop repairing it with no telling: completeanarchy, polska, satania, halo, bangtan, thesilphroad (R-completeanarchy-c187508d7ae3-022, R-polska-dadda7358f9a-049, R-satania-84df196116f3-022, R-halo-b988854ba168-036, R-bangtan-7efe23da0d0c-070, R-thesilphroad-a3f86618e239-018), a candidate branch, 'the shape read' (F-024). It does not hold where informed members hold the change against keepers who do not know: armok, maryland, oneshot (R-armok-8852b2ab85f8-022, R-maryland-d1b6e2d1d5f3-017, R-oneshot-e840e127e9b2-013), a limit below the trigger, at three nodes. Amended at D4a-01, after the D3 freeze (46384686...); not re-checked: no room was served these words." Rooms: completeanarchy, polska, satania, halo, bangtan, thesilphroad, armok, maryland, oneshot. Lines: R-completeanarchy-c187508d7ae3-022, R-polska-dadda7358f9a-049, R-satania-84df196116f3-022, R-halo-b988854ba168-036, R-bangtan-7efe23da0d0c-070, R-thesilphroad-a3f86618e239-018, R-armok-8852b2ab85f8-022, R-maryland-d1b6e2d1d5f3-017, R-oneshot-e840e127e9b2-013.
• A11, move 199 (dropping working hypotheses), history lines 177 to 180, reverted by line 182 (move 203). The arrow from P02 to P04 would be withdrawn (status withdrawn, in_model false). Its reason, as written: "Rule 16 and rule 14: preserve what remains supported, and replace or withdraw what does not. No defensible formulation of A11 as an arrow survives the return. (1) The order between rooms never beat its null (111 of 185 rooms, 0.600; null p95 0.670; base_rate_beaten false), and C2 withdrew it, keeping a transition inside threads. (2) Inside threads, the return's census (D3a-38, re-run as _analysis/d4a01_a11_classes.json) shows the transition in 111 rooms and template first in 98 of the 228 rooms whose lines show both ends; across threads 19 more. The reverse order is nearly as common, so the arrow carries no direction. (3) The failure clause is present and stands alone in 117 rooms: talk only in 31 (berkeley, bih, bitcoin, edboxes, eve, FinalFantasy, guatemala, habs, HomestarRunner, indieheads, ireland, latvia, leagueoflegends, mexico, morrowind, msu, parrots, placeDE, placesnek, polska, portugal, quebec, samuraijack, slovenia, spain, suomi, titanfall, torontoraptors, unitedkingdom, wine, zootopia) and layout only in 86. (4) The mechanism's word 'settled' is contradicted by 261 accepted restrict lines whose 'how' says a template does not settle the design for good (rival templates, polls, later redesigns, placers who deviate). A reworded arrow (a layout answers design talk in its thread) would be a new claim, never served to any room, with its reverse nearly as common: a co-occurrence, not an arrow. What remains supported, that design talk and layouts meet in threads, is held by P02's and P04's statements and ranges ('a design argued over days and redone'; 'an official template the room asks everyone to follow against other versions'), which stand. The arrow's record keeps its evidence and reasoning." Rooms (the talk-only rooms of the failure clause): berkeley, bih, bitcoin, edboxes, eve, FinalFantasy, guatemala, habs, HomestarRunner, indieheads, ireland, latvia, leagueoflegends, mexico, morrowind, msu, parrots, placeDE, placesnek, polska, portugal, quebec, samuraijack, slovenia, spain, suomi, titanfall, torontoraptors, unitedkingdom, wine, zootopia. Lines (the first-return lines of the census classes talk only, layout only and template first, 216): 2007scape-012, 4chan-011, AMD-012, AmericanFlaginPlace-009, AvoArmy-011, AyyMD-041, CKCrusaders-041, CSUFoCo-015, DestinyTheGame-014, DevonUK-010, FinalFantasy-014, GuardiansOfDatBoi-009, HomestarRunner-015, MonkeyIsland-011, NewMexico-015, ScottPilgrim-015, Temple-041, TsundereSharks-048, adventuretime-041, aggies-050, albania-046, alberta-016, anaheimducks-015, animalcrossing-053, anime-012, argentina-051, armenia-030, bardmains-055, berkeley-038, berserk-012, bih-036, bisexual-012, bitcoin-012, blackmirror-046, borderlands-020, bostonbruins-052, brasil-011, canada-012, catalonia-009, catholicism-011, cfb-048, chibears-010, chicubs-010, cirkeltrek-013, clashroyale-045, completeanarchy-011, criticalrole-043, crusaderkings-039, cubers-014, cuboulder-008, czech-014, dankmemes-045, darksouls-042, detroitlions-049, edboxes-046, edmontonoilers-041, empiredidnothingwrong-013, england-045, ethereum-056, europe-050, eve-049, fivenightsatfreddys-012, flashtv-055, floridagators-035, flyers-035, formula1-048, forsen-010, ftlgame-044, fxKorea-010, gameattack-015, gamindustri-053, gatech-041, ghibli-010, globaloffensive-012, greenbaypackers-011, guatemala-010, habs-011, heyyea-010, hongkong-047, hsv-012, hyperlightdrifter-046, indieheads-056, indonesia-035, ireland-045, italy-035, katawashoujo-041, lakers-009, latvia-029, leagueoflegends-042, lifeisstrange-013, linux-015, littlewitchacademia-012, loliconsunite-010, lost-044, lovelive-041, lsu-049, luxembourg-012, madokamagica-041, magictcg-010, mechanicalkeyboards-042, metalgearsolid-013, mexico-041, miamidolphins-041, minecraft-017, mit-041, mizzou-018, mls-053, monalisaclan-013, morrowind-011, msu-011, muse-048, mylittlepony-011, nVade-013, naruto-010, nbaspurs-056, necrodancer-011, neopets-047, newzealand-046, nightinthewoods-012, nin-012, ockytop-042, opticgaming-012, overwatch-017, parahumans-013, parrots-013, pathofexile-012, pcmasterrace-013, pebble-007, penguins-055, philippines-052, pitt-041, placeDE-041, placecanada-013, placesnek-031, planetside-056, polska-054, portal-012, portugal-082, predators-010, purdue-051, puzzleanddragons-013, quebec-013, radiohead-010, rainbowgrid-006, rangers-045, re_zero-042, redpandas-018, rickandmorty-042, rit-011, rocketbeans-008, rocketleague-011, romania-015, roosterteeth-013, rpi-049, russia-052, rust-010, rutgers-012, samuraijack-047, scp-014, seahawks-014, serbia-051, singapore-016, slovenia-041, smite-010, southpark-012, spacexmasterrace-052, spain-035, ss13-016, ssbpm-012, stardewvalley-058, starryknights-041, stevenuniverse-012, straya-047, suomi-041, survivor-035, sweden-045, tampabaylightning-045, terraria-048, teslamotors-012, tf2-041, thebluecorner-016, theexpanse-012, theitalyplace-054, thepointing-043, thesimpsons-011, tinytardis-009, titanfall-013, torontoraptors-016, totallynotrobots-013, transformers-046, trees-013, turkey-055, twentyonepilots-012, ubc-046, ucf-050, uiuc-011, ultrahardcore-005, undertale-011, unitedkingdom-035, unturned-013, uofm-008, uofmn-035, uruguay-012, utahjazz-009, utaustin-013, uva-042, uwaterloo-011, vegan-043, virginiatech-043, weezer-053, wine-011, witcher-017, worldoftanks-015, xenoblade_chronicles-012, zelda-061, zootopia-012.

Grounds, and what would have changed the cuts

• Grounds. Every cut is grounded on keys alone ("ground": "keys" in each tier_history line). No member's material was read for a cut. The lines' how sentences were read only to test whether a key's rooms share one change (cut rule R2).
• Was any cut attractive because of what its members made? No. I hold every room's canvas outcome and cannot unknow it, but no cut read the canvas, the verdicts or the projections. F01 and its subforms are about the sub's own April Fools play, which the lines say is not canvas work. The first cut was chosen by the size of the element's change-key group at level 4.
• What would have made me choose differently. Shared corrections at the third or fourth order (none reach 5 rooms); a kind I could rely on for add-step, add-condition and restrict lines (J: not reliable, kappa 0.08), which would have let the elaboration view lead; or change keys that co-occur across different nodes in 5 rooms or more, which would have given multi-key branch definitions.
• Local validation. Not established. The labelers mark all 705 rooms unresolved, and prior.md section 3's measure (each second-order theme with a coordinating turn reached by an element or a validated adaptation, or listed unresolved) was not computed (J after the D3c interview).

The accounting (stage D4 section 4), against n_analysed

counts.py gives n_analysed = 705 (census 705, dropped 0). In a branch: 103 (F01 and its subforms). The unadapted group: 0 (no room returned none). Adapted but in no branch: 602. Unreturned at D2: 0. The four counts sum to 705, the analysed set. _views/distance.tsv holds 705 rooms.

Adapted but in no branch, each with why. Every one of these rooms holds accepted keys, so none is unadapted. None holds a key cut at D4a-01, because the pass reached one element, K01. The number after each room is how many level-3 change keys on elements it holds that 5 rooms or more share: the keys a later pass would have read. A room marked 0 holds no such key: its change keys are its own, or on arrows.

2007scape (2); 49ers (6); aceattorney (0); AceOfAngels8 (0); AcePlace (5); achievement_hunter (0); adventuretime (3); aeon (2); aeskwadraat (2); afip (3); aggies (4); ainbowroad (4); albania (0); alberta (0); AmericanFlaginPlace (7); anaheimducks (6); anarchism (1); anarcho_capitalism (0); anarchocapitalism (1); androidgaming (3); animalcrossing (3); anime (4); aoe2 (1); apocalypserising (0); AprilKnights (2); araragi (1); archlinux (5); argentina (1); armenia (2); armok (0); aromantic (1); asexual (1); asexuality (3); asktransgender (2); auburn (1); australia (0); austria (0); austriahungary (0); AvoArmy (2); AyyMD (3); azerbaijan (2); azores (0); babymetal (8); bakchodi (2); bangladesh (2); bangtan (1); bannedfromanime_irl (0); bardmains (5); batman (1); baylor (0); bdsm (0); bees (0); belgium (2); berkeley (1); berserk (6); bigbang (0); bih (2); bisexual (8); bitcoin (2); blackmirror (2); bobross (0); borderlands (3); borussiadortmund (8); bostonbruins (3); brasil (5); Bravenewbies (1); Braveryjerk (0); bretagne (1); btc (1); buffalo (5); buffalobills (4); buffalowizards (1); bulgaria (2); calgaryflames (1); calvinandhobbes (1); cardinals (0); carletonu (0); catalonia (3); catalunya (3); catholicism (4); cavestory (4); CelticFc (1); cgpgrey (2); chess (1); chibears (2); chicago (2); chicubs (2); chile (4); ChronoTrigger (4); churchofhelix (0); cirkeltrek (0); CKCrusaders (2); Clannad (0); clashroyale (1); clg (3); cloud9 (3); clubpenguin (7); cmu (0); coldplay (6); colombia (1); colorado (2); columbia (1); completeanarchy (2); compsoc (1); Cornwall (1); corridor (0); COSI (0); costa_rica (1); counting (1); cowchop (4); croatia (0); crusaderkings (5); CryptoFR (2); crystalclear (2); CSUFoCo (1); cubers (5); cuboulder (0); czech (6); daftpunk (3); darksouls (2); DatBoi (1); davidbowie (2); day9 (1); DCFU (1); de (7); denmark (1); destiny (1); detroitlions (1); detroitredwings (2); DevonUK (2); diablo (2); DiamondClub (0); dmt (2); dofus (0); Dominican (1); donaldglover (1); Dorset (2); dragonquest (3); drawhelix (4); drugs (2); duckgame (2); ducks (0); Dustforce (2); dwarffortress (2); eagles (0); earthbound (5); edboxes (1); eder (1); edmontonoilers (2); eesti (1); egypt (1); ElSalvador (0); empiredidnothingwrong (2); england (0); enterthegungeon (2); epublica (3); EraseThePlace (3); esperanto (5); europe (3); eve (4); evedreddit (0); evex (4); factorio (4); faeria (9); farmcarrots (1); FaroeIslands (1); FederationPlace (4); Fez (3); FinalFantasy (2); finland (1); fireemblem (2); FlaggyFlag (3); FlipnoteStudio3D (1); floridagators (0); FloydVsVoid (11); flyers (4); forsen (7); foxes (4); foxstevenson (0); france (0); frankocean (7); frc (5); freedomplanet (4); ftlgame (3); fullcommunism (1); funhaus (5); furry (4); futanari (1); futurama (2); fxKorea (4); GabrielDropout (2); gameattack (2); gameboy (1); gamindustri (5); gatech (7); genderfluid (1); germany (1); ghibli (2); giantbomb (4); giivasunner (6); glasgow (1); globaloffensive (3); Goodboye (2); gorillaz (1); GranColombia (5); gratefuldead (0); gravityfalls (3); greece (2); greenbaypackers (1); GreenCorner (2); GreenEmpire (0); greenland (0); greenlattice (5); greyblob (4); GrowTheTree (1); GuardiansOfDatBoi (1); guatemala (0); h3h3productions (2); habs (2); halo (1); haskell (0); hatsune (3); hawks (6); hearthstone (3); hellointernet (4); hentai (3); heyyea (2); homelab (1); HomestarRunner (2); homestuck (5); Homura (0); Honduras (1); hongkong (3); hoodr (0); hsv (3); httyd (2); hungary (3); hyperlightdrifter (6); iastate (3); ibo (1); ice_poseidon (0); iceland (1); ik_ihe (2); india (5); indonesia (5); ireland (1); israel (2); italy (3); Jamaica (2); japan (4); jerma985 (3); k_on (2); kamenrider (4); katawashoujo (1); kcroyals (5); kilian (2); kindafunny (2); kingdomhearts (1); konosuba (0); korea (7); KState (5); kurdistan (1); kuwait (1); lakers (2); landry (2); latin (2); latvia (0); leafs (3); leagueoflegends (0); lebanon (3); lego (4); lexington (0); lifeisstrange (1); linux (5); liquicity (1); lithuania (2); littlebusters (0); littlewitchacademia (6); livecounting (3); liverpoolfc (2); lofihiphop (3); loliconsunite (3); loltyler1 (4); lost (2); lovelive (3); lsu (5); luciomains (5); Lugarix (0); luxembourg (2); Madeon (0); madokamagica (6); malaysia (6); malta (1); mamamoo (1); ManningPlace (1); Mario (1); marvel (1); maryland (3); masseffect (3); math (1); mcgill (1); mcgillrobotics (0); mechanicalkeyboards (4); megaman (5); metalcore (3); metroid (3); mexico (2); miamidolphins (2); MiamiHurricanes (0); mindcrack (6); minnesotavikings (1); mit (3); mizzou (3); monalisaclan (7); monero (6); MonkeyIsland (3); monsterhunter (1); montenegro (2); Moondye7 (4); mordhau (1); morrowind (6); msu (5); muse (2); mylittlepony (5); mysterydungeon (5); n64 (0); nascar (4); natsume (1); necrodancer (5); neopets (3); nerf (4); neu (2); neutralmilkhotel (2); NewMexico (4); newzealand (2); nin (3); nlsscirclejerk (3); nonbinary (0); norge (1); Northwestern (0); norway (3); notredame (5); NovaScotia (0); npr (3); nuclearthrone (2); nVade (2); nyu (1); ockytop (3); ohio (3); Okami (3); OKState (3); oltp (0); oman (0); OnePiece (4); onepunchman (2); oneshot (4); OneTrueEmilia (5); opticgaming (4); orderofthebluecorner (1); OrderOfTheKoi (2); osu (4); osugame (4); overwatch (4); pakistan (4); palestine (1); pansexual (0); parabola (2); parahumans (4); paranatural (2); parrots (3); partyparrot (1); pathofexile (4); patriots (2); paydaytheheist (1); pebble (0); penguins (5); pepe (2); Peru (2); peterdraws (1); pgang (0); philippines (2); physical_removal (1); picturegame (1); Pigifs (0); PinkUnion (2); PinkVomitMonster (1); pitt (4); placecanada (3); placeDE (3); placehearts (1); placehfarts (0); PlaceNL (5); placeportalpattern (5); placesnek (3); placestart (3); placetardis (5); pokemon (1); PokemonSage (0); poland (5); polska (5); polyamory (2); portal (4); porterrobinson (5); portugal (4); PORTUGALCARALHO (0); predators (7); prequelmemes (10); Produce101 (0); ProfessorLayton (3); purdue (5); purplecorner (3); PurplePixels (2); pusheen (4); puzzleanddragons (3); python (4); qotsa (0); quebec (6); quq (0); radiohead (3); rainbowgrid (1); rangers (4); raocow (2); ravens (3); re_zero (4); ReBoot (0); redhotchilipeppers (3); Redox (0); redpandas (4); redvsblue (1); RelientK (2); rhythmheaven (1); rickandmorty (4); ripcity (6); riskofrain (2); rit (4); rocketbeans (3); rocketleague (0); rojava (2); roosterteeth (3); rotmg (5); rpi (5); russia (2); Russiaclan (2); rutgers (5); sabres (3); sadboys (5); sakartvelo (0); sakuragakuin (1); satania (4); sbu (1); scotland (0); ScottPilgrim (0); scp (3); seolhyun (3); serbia (1); seventeen (0); sheffield (2); shield (3); shishamo (1); shitredditsays (2); shrooms (0); signal (0); simonfraser (2); singapore (5); sips (6); sixers (3); skyrim (3); Slovakia (0); slovenia (2); smite (1); snsd (1); sonicthehedgehog (4); sooners (4); southafrica (5); southpark (2); soylent (4); spacexmasterrace (4); spain (2); splatoon (2); spongebob (4); SquareSpiral (4); ss13 (2); ssbpm (2); starbound (2); starcitizen (3); stardewvalley (6); starlightstage (8); starryknights (5); startrek (0); starvstheforcesofevil (3); stlouis (3); stlouisblues (0); strangerthings (1); straya (1); Summit1G (3); suomi (4); superman (1); sweden (2); SwissNeutralityZone (4); switzerland (4); SydneyFC (0); tagpro (5); taiwan (2); tampabaylightning (1); teamcube (2); TeamDignitas (1); TeamYellowPlace (2); telegram (3); Temple (3); tenma (1); teslamotors (2); tetris (1); texas (3); texastech (1); tf2 (1); thailand (4); the_pack (1); THE_R (1); TheAdventuresofTintin (0); TheB1G (0); theblackvoid (1); thebluecorner (3); theexpanse (6); TheFarLeftSide (0); Thegreatblackhole (1); TheGreatGreenBox (1); TheHulk (5); TheID (0); theitalyplace (1); thepointing (2); thesilphroad (1); thetreesnetwork (0); timbers (3); tinytardis (2); tokyo (1); toolband (3); torontoraptors (4); touhou (3); traaaaaaannnnnnnnnns (4); TransFlagPlace (1); transformers (3); transgender (1); TreePlace (1); trees (2); truestl (2); TsundereSharks (1); turkey (4); twentyonepilots (2); twice (5); twitch (3); ubc (2); ubreddit (3); ucf (1); udub (2); ufl (1); uiowa (1); uiuc (2); ukplace (2); ukraine (3); UMiami (0); undertale (4); unitedkingdom (1); UNLincoln (0); unturned (5); uofm (3); uofmn (3); uoft (2); UPenn (2); uruguay (0); utahjazz (3); utaustin (3); uva (2); uvic (3); vegan (2); VelvetUnderground (2); vermont (5); vietnam (1); Villanova (1); vim (3); vinesauce (4); virginiatech (4); vocaloid (3); wales (1); wallstreetbets (2); wario (0); warriors (4); warthunder (2); watchmen (0); weezer (3); wine (3); winnipegjets (3); WisconsinBadgers (2); witcher (3); worldpowers (3); WPI (2); wsu (0); xcom (3); xenoblade_chronicles (6); yerbamate (0); yumenikki (0); yuyuyu (4); zootopia (7).

The cuts of D4a-02 (the second pass), in the order cut

The pass read the 31 element change keys at level 4 that D4a-01 left, each held by 20 rooms or more, with the dated amendment R1a to R1c (the level, keys read together, restated statements set aside). Each key's lines were read whole in _analysis/d4a02_keys/ (d4_key_lines_bynode.py). The cuts were written by _analysis/d4_cut_batch.py from the specs _analysis/d4a02_cut_F07.json to F35.json, each logged as a move and snapped when it was written.

• F07 (move 208, merging; branch named; level fourth; parent none). Key: restrict:P05:T4-006. 59 members. Kind: specification (specification 56, elaboration 1, correction 2), not relied on. the telling clause not shown at the design: informed members reject the planned change. P05's added clause says that a planned change is repaired away until its keepers are told. In these rooms the lines limit the clause where the design is settled. About half say that members who know of the change reject it, so repair or recolouring enforces a competing design, and telling does not settle it. The others say the record does not show the untold keepers, the agreed change or the loss that the clause needs. The frozen clause holds here only in part.
• F08 (move 209, merging; branch named; level third; parent none). Key: add-step:P14:T3-028. 50 members. Kind: specification (specification 55, elaboration 1), not relied on. machine hands kept on one target: installing, teaching and updating every runner. P14 says the room adjusts its machine hands. In these rooms the adjusting is a step of its own: members install the script and teach it, and they keep every runner's target current, because a stale target undoes the room's own new design. A revision must reach the operator and the target file, not only the people.
• F09 (move 210, merging; branch named; level fourth; parent none). Key: link:P03:T4-014. 54 members. Kind: specification (specification 53, elaboration 2), not relied on. the move returns to building: relocation followed by renewed construction. The model runs from building to keeping and, after a loss, to the move (P10), but no arrow runs from the move back to building. In these rooms it does: after a relocation the room recruits, designs and builds again at the new spot, and members report the new work.
• F10 (move 211, merging; branch named; level third; parent none). Key: add-step:P08:T3-010. 38 members. Kind: specification (specification 35, elaboration 4), not relied on. a deal carried out as work: prepared in the room, then implemented and relayed. P08 says rooms propose, settle and contest deals and hold each other to them. In these rooms a deal also takes work around it: members prepare it inside the room (local consensus, a contact, checks on the counterpart's credibility), and they carry it out by moving placed work, changing templates and bot targets, and relaying the terms to placers who do not know them.
• F11 (move 212, merging; branch named; level fourth; parent none). Key: add-step:P13:T4-028. 43 members. Kind: specification (specification 31, elaboration 12), not relied on. carrying on as renewed work, begun before the close. P13 says that near the close and after it the room decides what of its effort to keep. In these rooms carrying on is work in its own right, and it starts while placement is live: on another canvas the room recruits, builds and defends again, members remake the work in other media, and the mark is adopted into the room before the close.
• F12 (move 213, splitting; subform declared; level third; parent F11). Key: add-step:P13:T3-054. 22 members. Kind: specification (specification 11, elaboration 11), not relied on. the spine restarted on another canvas. Within F11, the room carries its work to a successor canvas, and there the spine runs again: recruiting, choosing a site, building, defending and losing work.
• F13 (move 214, splitting; subform declared; level third; parent F11). Key: add-step:P13:T3-057. 9 members. Kind: specification (specification 8, elaboration 1), not relied on. the work remade in other media, often before the close. Within F11, members remake the canvas work elsewhere (fan art, games, physical copies, a quoted text), and this often begins while the original is still live.
• F14 (move 215, splitting; subform declared; level third; parent F11). Key: add-step:P13:T3-156. 7 members. Kind: specification (specification 7), not relied on. the mark adopted as the room's emblem while it is still placed, damage included. Within F11, the room takes the mark into its own sub (banner, flair, a room feature) during placement, and it may adopt the damaged form as fitting.
• F15 (move 216, merging; branch named; level fourth; parent none). Key: add-step:P04:T4-008. 42 members. Kind: specification (specification 28, elaboration 15), not relied on. the plan kept current: correcting, updating and re-circulating the template. P04 says members circulate a template or coordinates and ask strays to keep to them. In these rooms the template needs upkeep of its own: members correct the reference against the canvas, replace stale plans, make updates noticeable, route helpers to the current version and message friendly placers who still follow a superseded one.
• F16 (move 217, merging; branch named; level third; parent none). Key: add-step:P23:T3-014. 35 members. Kind: specification (specification 35), not relied on. the chat kept open: invitations renewed, entrants screened, outcomes relayed to the thread. P23 (a candidate) says members move planning into a chat and invite helpers. In these rooms the chat's access is upkeep: members replace expired invitations and broken links, screen who comes in, and relay decisions back to the thread for those outside.
• F17 (move 218, merging; branch named; level fourth; parent none). Key: add-step:P11:T4-026. 36 members. Kind: specification (specification 32, elaboration 5), not relied on. the marking provisional and contested: success judged and credit disputed while the work goes on. P11 says that when a work is up the room marks it: announces, celebrates and credits. In these rooms marking comes while the work is still live and is open to dispute: success is judged by what survives, completion is provisional, upkeep goes on, and credit is challenged and corrected.
• F18 (move 219, splitting; subform declared; level third; parent F17). Key: add-step:P11:T3-009. 15 members. Kind: specification (specification 12, elaboration 3), not relied on. success judged while upkeep continues. Within F17, the room judges success by recognizable survival rather than a perfect mark, treats completion as a milestone, and keeps the obligation to maintain.
• F19 (move 220, splitting; subform declared; level third; parent F17). Key: add-step:P11:T3-043. 14 members. Kind: specification (specification 13, elaboration 1), not relied on. credit contested during the work. Within F17, members dispute and correct who made the work while it is still being placed: accusations, denials, competing labour claims and corrected omissions.
• F20 (move 221, merging; branch named; level third; parent none). Key: add-step:P12:T3-117. 10 members. Kind: specification (elaboration 4, specification 7), not relied on. the close as news: late arrivals told that placement has ended. P12 says the room records what happened and marks the end. In these rooms the end itself must travel: members who did not know offer to rebuild or ask for help, and others tell them that placement has ended and correct the stale call.
• F21 (move 222, merging; branch named; level fourth; parent none). Key: restrict:K01:T4-001. 32 members. Kind: specification (specification 30, elaboration 2), not relied on. the canvas call not a joke: the April Fools frame kept apart from the canvas work. K01 says that a claim in the host sub, a call to the canvas included, may be a joke. In these rooms the play stays with the sub's own business: its fakes and takeovers do not make the canvas calls jokes, and the frame does not decide what the canvas work was.
• F22 (move 223, merging; branch named; level fourth; parent none). Key: add-condition:P16:T4-017. 30 members. Kind: specification (elaboration 3, specification 28), not relied on. help conditional on the helper's own work and spare hands. P16 says rooms help one another, trade help and turn requests away. In these rooms help waits on conditions the statement does not hold: the helper's own work must be safe or finished, the helper must have pixels and time to spare, and the terms (space, non-interference, recognizable help) must hold.
• F23 (move 224, merging; branch named; level fourth; parent none). Key: add-condition:P13:T4-028. 30 members. Kind: specification (specification 25, elaboration 6), not relied on. the close scoped to the original canvas: continuation renewed, uncertain and contested. P13 places the decision near the close and after it. In these rooms the stop at hour 73 binds only the original canvas: on successor canvases placement goes on, and continuation is often one member's attempt, contested or unfinished, and conditioned on tolerable upkeep and access.
• F24 (move 225, merging; branch named; level fourth; parent none). Key: restrict:P14:T4-009. 27 members. Kind: specification (specification 27), not relied on. machine hands debated, not shown placing. P14 says members add placements with scripts, bots and spare accounts, and argue over them. In these rooms only the arguing part shows: scripts are offered, debated, refused or attributed to outsiders, and the room's own machine placement is not shown.
• F25 (move 226, merging; branch named; level fourth; parent none). Key: restrict:P13:T4-028. 27 members. Kind: specification (specification 29, elaboration 1), not relied on. carrying on in talk: proposals and single attempts, not shown taken up. P13 says the room decides what to keep. In these rooms carrying on shows as invitations, proposals and one member's attempt; collective uptake and durability are not shown, and some reuse begins during live placement.
• F26 (move 227, merging; branch named; level third; parent none). Key: restrict:P05:T3-010. 23 members. Kind: specification (specification 22, correction 1), not relied on. the telling clause not shown at the border: informed members contest the terms. As in F07, P05's added clause holds only in part, here where the planned change is a border or compromise agreed with a neighbour: members who know the terms contest them or refuse the concession, or the record does not show untold keepers or the loss.
• F27 (move 228, merging; branch named; level fourth; parent none). Key: restrict:P05:T4-013. 27 members. Kind: specification (specification 27, correction 1), not relied on. the telling clause not shown in keeping: interference seen, its cause and loss not. As in F07, P05's added clause holds only in part, here in watching, repairing and defending: interference with a planned change is seen, but who was told, whether keepers knew, and whether the change was lost for want of telling are not, and informed disagreement also occurs.
• F28 (move 229, merging; branch named; level third; parent none). Key: add-condition:P16:T3-010. 22 members. Kind: specification (specification 20, elaboration 3), not relied on. help at the border conditional: on the home work, the terms and harm to a third room. As in F22, help waits on conditions the statement does not hold, here where neighbours settle borders: the helper's home work comes first, help depends on stopping attacks or on sparing a work, it can entail destroying a third room's mark, and offers often stay offers.
• F29 (move 230, merging; branch named; level third; parent none). Key: add-condition:P23:T3-014. 25 members. Kind: specification (specification 25), not relied on. the chat's reach conditional: access, and the return of decisions to the thread. P23 (a candidate) says members move planning into a chat and bring its decisions back. In these rooms the chat's reach depends on working access, and those outside it learn only what returns to the thread; its decisions do not bind those who cannot see them.
• F30 (move 231, merging; branch named; level third; parent none). Key: restrict:P23:T3-014. 24 members. Kind: specification (specification 24), not relied on. the chat seen only through its invitations and returned reports, not its choices. P23 (a candidate) says the room's design, borders and diplomacy can be settled in the chat. In these rooms the record shows only the outside of the chat: invitations, venue proposals and reports that come back. Its deliberation, membership and authority are not shown.
• F31 (move 232, merging; branch named; level third; parent none). Key: restrict:P19:T3-010. 22 members. Kind: specification (specification 22), not relied on. hosting as one-off permission: requests and leave, no standing host rule. P19 says one room's work lives on another's mark under the host's rule, with hosts ruling on requests. In these rooms hosting shows only as a guest's request and a one-off permission, often provisional or contested; a standing host rule, an approval procedure or durable guest status is not shown.
• F32 (move 233, merging; branch named; level third; parent none). Key: link:P05:T3-009. 21 members. Kind: elaboration (elaboration 18, specification 3), not relied on. the marking returns to keeping: celebration followed by renewed defence. The model runs from building to marking (A07) and from building to keeping (A03), but not from marking back to keeping. In these rooms it does: a celebration or completion post draws renewed defence calls, repair and night cover in its own thread.
• F33 (move 234, merging; branch named; level fourth; parent none). Key: add-step:P04:T4-006. 23 members. Kind: specification (specification 19, elaboration 4), not relied on. the plan kept current while the design settles. As in F15, the template needs upkeep of its own, here while the design is still being settled: members check details against one another's instructions, correct and replace the reference, retire outdated posts and message overwriters, so that one updated plan replaces competing ones.
• F34 (move 235, merging; branch named; level fourth; parent none). Key: add-condition:P02:T4-006. 22 members. Kind: specification (specification 16, elaboration 6), not relied on. the design sized to the hands: scale and detail conditioned on crew, upkeep, space and colours. P02 says the room settles what its mark will show. In these rooms the design is conditioned by what the room can build and keep: its scale and detail depend on the crew it can assemble, the upkeep it can sustain, the space it can protect and the colours available.
• F35 (move 236, merging; branch named; level third; parent none). Key: add-step:P01:T3-002. 18 members. Kind: elaboration (elaboration 17, specification 1), not relied on. the call made workable: answers ask for a task, and callers supply it. P01 says the room calls members and kindred subs to place. In these rooms answering a call is an exchange: a willing member asks what to do, where to start or for backup, and the caller supplies instructions, location cues and visibility (a sticky, other venues), reaching people beyond the room.

Groups read at D4a-02 and set aside

Eight keys did not pass cut rule R2. Their lines state several different changes, or restate what the element's statement holds (R1c).

• add-condition on P05 at T4-013 (43 rooms): members' time and sleep, a stale display, repair that keeps an alteration, upkeep beside growth, upkeep of others' work, and the evidence for the planned-change clause.
• add-step on P02 at T4-006 (41 rooms): polls, provisional fixes, feasibility tests and adopted alterations, most held by 'the room argues its design'.
• add-step on P05 at T4-013 (40 rooms): checking whether a pixel was intended, shifts and handoffs, ranking repair against extension, defence during building; partly held by 'puts upkeep before growth'.
• add-step on P20 at T4-002 (36 rooms): finer forms of the megathread, the removals by rule and the venue that the statement holds.
• add-step on P12 at T4-027 (33 rooms): several steps of recording; its child T3-117 states one and is cut on its own (F20).
• add-condition on K01 at T4-001 (31 rooms): offence and hurt, who can see the joke, and the frame kept apart from the canvas; 15 of its rooms are in F01.
• add-condition on P08 at T4-016 (29 rooms): several conditions of deals; the commonest, that a deal binds only placers who know and accept it, is held by P21 and A19.
• add-condition on P14 at T4-009 (28 rooms): mostly F08's change (targets kept in step) and the rest of several kinds.

Families: one change at several nodes

A branch is keyed at one node, so a change the rooms make in several settings appears as several branches. The ladder joins them only at the dimension or the bare target, where a group is a grouping, not a branch (cut rule R4). The families are named here for D5:

• P05's added clause holds only in part: F07 (where the design is settled), F26 (at the border), F27 (in keeping). In F07 about half the lines say that informed members reject the planned change, and the rest that the record does not show the untold keepers; F26's and F27's lines say one or the other. Together they hold 108 rooms. This is the trunk's limit beside F-024's branch, 'the shape read', and the held proposal of move 198.
• Keeping the shared instruction current: F15 and F33 (the template), with F08 (the machine targets). A stale plan or target undoes the room's own new work, as P05's known-form repair does with people.
• Help conditional: F22 and F28.
• Carrying on: F11 with its subforms F12 to F14, F23 and F25.
• The chat: F16, F29 and F30.
• New links the model lacks: F09 (the move back to building) and F32 (the marking back to keeping).

The overlap

476 rooms are in at least one branch. By the number of branches they are in (subforms counted): 1: 200, 2: 127, 3: 92, 4: 36, 5: 11, 6 or more: 10. crosslinks.jsonl holds 759 lines over 348 pairs of branches that do not nest. The branches cut across one another rather than partition the rooms: a room makes several element-level changes at once. They are facets of the rooms' adaptations, not exclusive forms, and D5 reads them so.

Where the cuts stop

The cuts stop with F35. Every element change key at level 4 held by 20 rooms or more has been read: D4a-01 read K01's, and D4a-02 the other 31. Keys held by 5 to 19 rooms at level 4 were not read for a cut, since J's credit limit (QD4-001) ends the run soon. The rooms they hold, and in no branch, are named below with their counts. This is a stopping rule by key size, not a claim that the smaller keys hold no form.

The comparison (QD4-002)

The comparison was declared in notes-D4.md before its script ran. It did not run: the engine stops on the 335 second-order themes placed on the two residue third-order themes (T3-017 and T3-133), which no fourth-order theme gathers. The question is with J (QD4-002). The projections follow the comparison and are not built either.

The accounting at D4a-02 (stage D4 section 4), against n_analysed

n_analysed = 705. In a branch: 476. The unadapted group: 0 (no room returned none). Adapted but in no branch: 229. Unreturned at D2: 0. The four counts sum to 705, the analysed set.

Adapted but in no branch, each with why. Each holds accepted keys, so none is unadapted, and none holds a key that was cut. The number after each room is how many element change keys at level 4, held by 5 to 19 rooms, it holds: the keys below the pass's size rule. A room marked 0 holds none: its change keys are its own or on arrows, or its lines are specifications only.

2007scape (3); aceattorney (1); AceOfAngels8 (0); achievement_hunter (2); aggies (2); ainbowroad (6); albania (0); alberta (0); anarchism (2); anarcho_capitalism (2); anarchocapitalism (0); androidgaming (1); apocalypserising (0); araragi (2); argentina (2); armenia (3); aromantic (2); asexual (1); asktransgender (1); auburn (2); austria (4); austriahungary (1); azores (0); bangladesh (2); bannedfromanime_irl (0); batman (1); bdsm (1); bees (0); bigbang (0); bih (1); bitcoin (2); blackmirror (2); bobross (0); brasil (3); Bravenewbies (1); Braveryjerk (0); bretagne (3); bulgaria (5); cardinals (1); carletonu (0); chicubs (3); churchofhelix (0); cirkeltrek (1); CKCrusaders (4); Clannad (1); cmu (1); completeanarchy (3); compsoc (1); Cornwall (1); COSI (0); costa_rica (3); counting (1); CryptoFR (2); cuboulder (1); daftpunk (3); DatBoi (1); detroitredwings (2); DevonUK (1); diablo (3); DiamondClub (1); dmt (2); dofus (0); Dominican (4); Dorset (1); drugs (1); duckgame (1); ducks (0); eagles (0); edmontonoilers (7); ElSalvador (2); england (1); europe (3); evedreddit (0); FinalFantasy (2); finland (1); fireemblem (2); FlipnoteStudio3D (1); floridagators (2); foxstevenson (2); france (1); futanari (1); GabrielDropout (2); gameboy (1); genderfluid (1); germany (3); giantbomb (4); glasgow (1); Goodboye (4); gratefuldead (2); greenbaypackers (2); GreenCorner (2); GreenEmpire (0); greenland (0); GrowTheTree (2); guatemala (1); halo (2); haskell (0); homelab (1); HomestarRunner (2); Homura (1); Honduras (2); hoodr (1); hsv (3); iastate (2); ibo (2); iceland (1); ireland (2); k_on (3); landry (2); latin (4); latvia (0); leafs (3); leagueoflegends (2); lexington (1); littlebusters (0); lost (2); Lugarix (0); Madeon (0); madokamagica (4); malta (3); math (1); mcgillrobotics (0); MiamiHurricanes (1); mizzou (1); monsterhunter (1); mordhau (2); mysterydungeon (2); n64 (0); natsume (2); neutralmilkhotel (2); nonbinary (0); norge (0); Northwestern (4); norway (4); NovaScotia (0); npr (3); nVade (5); nyu (3); oltp (1); oman (0); oneshot (4); orderofthebluecorner (1); overwatch (4); pansexual (3); parabola (1); paranatural (5); partyparrot (2); pathofexile (5); peterdraws (1); pgang (0); Pigifs (1); PinkUnion (5); PinkVomitMonster (1); placecanada (4); placehearts (1); placehfarts (0); placesnek (4); PokemonSage (0); Produce101 (0); ProfessorLayton (4); qotsa (1); quq (0); rainbowgrid (4); ReBoot (1); Redox (0); redvsblue (1); rit (5); rocketleague (3); rojava (3); roosterteeth (3); Russiaclan (4); sakartvelo (0); sakuragakuin (3); ScottPilgrim (1); seventeen (1); shrooms (0); signal (0); singapore (4); Slovakia (1); snsd (3); sonicthehedgehog (5); startrek (0); stlouisblues (2); straya (3); Summit1G (3); suomi (4); sweden (2); SydneyFC (0); teamcube (5); TeamDignitas (1); TeamYellowPlace (2); telegram (1); tetris (0); texastech (1); the_pack (2); TheAdventuresofTintin (1); TheB1G (1); Thegreatblackhole (3); TheGreatGreenBox (1); TheID (0); theitalyplace (1); thesilphroad (1); thetreesnetwork (1); traaaaaaannnnnnnnnns (7); transgender (1); TreePlace (2); trees (3); truestl (3); ufl (2); uiowa (2); UMiami (0); UNLincoln (0); uruguay (2); uva (5); vegan (2); VelvetUnderground (1); vermont (2); vietnam (1); Villanova (1); wales (3); wario (0); watchmen (0); winnipegjets (2); worldpowers (2); wsu (1); xenoblade_chronicles (6); yerbamate (0); yumenikki (1); yuyuyu (3).

The comparison (section 5), read at D4a-03

Run at D4a-03 on chunk-comparison-v1.1 (engine sha256 c8d2eb8c7c30...), a departure declared before the run (departures.jsonl line 10; QD4-002), with the stage file's command and defaults: a = 1.0, seeds [0, 1], 200 bootstrap draws and 100 permuted copies per seed. It reads the concept store, the placements, the tiers, the dropped cases and the prior. No verdict, canvas field or branch entered its feature space. Its features are 28 fourth-order themes over 705 rooms; empty compositions: 0. It is a non-adaptation baseline that shares the inductive coding structure, not theory-free evidence.

• All rooms (705 rooms): status supported-diagnostic-cut; k carried 4. At k 4: seed 0: silhouette 0.024, real stability mean 0.151 and median 0.154, null 0.099 and 0.082, margins +0.052 and +0.072, null 95th percentile 0.220; seed 1: silhouette 0.024, real stability mean 0.162 and median 0.160, null 0.108 and 0.099, margins +0.054 and +0.061, null 95th percentile 0.208. The k that clear under both seeds: 3, 4, 5, 6, 7, 8, 9, 10, 11, 12. The largest silhouette at any k under seed 0 is 0.051. The median own-data weight is 0.990. Against the carried Ward/CLR labels at k 4, the Jaccard presence and Hellinger share partitions give ARI -0.007 and -0.008, and a/10 and 10a give 0.145 and 0.164.
• Without the low-information rows (626 rooms): status supported-diagnostic-cut; k carried 2. At k 2: seed 0: silhouette 0.097, real stability mean 0.173 and median 0.146, null 0.074 and 0.017, margins +0.099 and +0.129, null 95th percentile 0.380; seed 1: silhouette 0.097, real stability mean 0.184 and median 0.144, null 0.060 and 0.014, margins +0.124 and +0.130, null 95th percentile 0.307. The k that clear under both seeds: 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12. The largest silhouette at any k under seed 0 is 0.097. The median own-data weight is 0.991. Against the carried Ward/CLR labels at k 2, the Jaccard presence and Hellinger share partitions give ARI 0.014 and -0.005, and a/10 and 10a give 0.143 and 0.556.
• The tier check (descriptive, not proof of form): at k 4, dimension ARI 0.110 and third-tier ARI 0.170. The third tier has 195 columns and 13.52 nonzero per room (6.9 per cent), so its profiles are sparse.
• Low-information rooms (fewer than five placed themes): 79, kept in the primary run and removed in the second.
• The clusters (two-letter codes; no mechanism labels; best-match Jaccard against the branches is descriptive, and ARI against the branches is undefined, since they overlap):
  • VA: 176 rooms, 0 low-information; largest profile shares T4-001 0.256, T4-013 0.112, T4-016 0.099; median concepts 182; mean residue share 0.019; best-match branch F01 (Jaccard 0.187).
  • IT: 292 rooms, 72 low-information; largest profile shares T4-001 0.241, T4-006 0.090, T4-016 0.077; median concepts 56; mean residue share 0.023; best-match branch F01 (Jaccard 0.135).
  • TX: 155 rooms, 5 low-information; largest profile shares T4-013 0.120, T4-006 0.112, T4-016 0.101; median concepts 91; mean residue share 0.010; best-match branch F08 (Jaccard 0.096).
  • CH: 82 rooms, 2 low-information; largest profile shares T4-006 0.127, T4-016 0.112, T4-013 0.087; median concepts 187; mean residue share 0.011; best-match branch F08 (Jaccard 0.100).

The residue share per room (QD4-002)

v1.1 counts a second-order theme placed only on the residue themes T3-017 (carrying on ordinary sub talk) and T3-133 (recovering a hacked sub) with the unplaced concepts, apart from the composition. 199 rooms hold residue concepts (335 themes, 1982 concepts). The residue share is the room's residue concepts over its placed and residue concepts: median 0.000, mean 0.018, maximum 0.503. 78 rooms are at 0.05 or more, 39 at 0.10 or more and 7 at 0.30 or more. Rooms at 0.20 or more: counting 0.50, Madeon 0.44, SydneyFC 0.42, aeskwadraat 0.39, Braveryjerk 0.33, wine 0.32, androidgaming 0.30, soylent 0.29, mls 0.23. For these rooms the composition leaves out a large part of what the coders coded, the room's own ordinary business.

The volume check (section 5), and the size of the rooms

The volume field is the comparison's own count of unique concepts per room (median 101; quartile cuts 50, 101, 209). Percentiles are against random groups of the same size from the 705 rooms (2,000 draws, seed 0).

• The volume verdict for the clusters. The cluster labels explain 0.255 of the variance in log concepts (eta squared), at the 100.0th percentile of the permutation null (null maximum 0.026). By concept quartile (smallest first): VA 3 34 61 78; IT 137 66 48 41; TX 31 64 35 25; CH 5 13 33 31. So the carried partition largely separates rooms by how much the coders coded.
• Branches and size. In a branch, by quartile: 63 122 149 142; in no branch: 113 55 28 33. 11 of 35 branches have a median room size at or above the 95th percentile, and 0 at or below the 5th. A larger room holds more lines and more keys, so it has more chances to share one; the branches lean to the larger rooms for that reason, at least in part. A correlation or this null alone cannot show that a branch's difference is independent of size.

The projections (section 6), after every cut was snapped

python3 tools/projections.py (2,000 permutations, seed 0) wrote projections-D4.jsonl, one line per branch and per cluster. Its verdict for a room is my last ruling at D3, else the readers' agreed verdict, so the first return and the re-check are mixed. The canvas sentence it carries: "The atlas records what survived the last hour of a 97-hour event; it is not a measure of skill, and a large area is not evidence of good organizing (contract section 5)." Each line below gives the verdict counts (composes, with residue, does not compose, cannot tell), the percentile of the does-not-compose share, the median area in pixels with its percentile, and the group's median room size with its percentile.

• F01: 54, 49, 0, 0; does-not-compose percentile 43.5; median area 610.5 px, percentile 67.1; median size 201 concepts, percentile 100.0.
• F02: 18, 15, 0, 0; does-not-compose percentile 47.4; median area 789.0 px, percentile 89.0; median size 223 concepts, percentile 100.0.
• F03: 9, 9, 0, 0; does-not-compose percentile 48.8; median area 349.0 px, percentile 9.4; median size 158 concepts, percentile 93.1.
• F04: 4, 9, 0, 0; does-not-compose percentile 49.0; median area 957.5 px, percentile 88.8; median size 243 concepts, percentile 99.3.
• F05: 6, 6, 0, 0; does-not-compose percentile 49.0; median area 706.5 px, percentile 67.8; median size 183 concepts, percentile 94.7.
• F06: 5, 3, 0, 0; does-not-compose percentile 49.3; median area 1535.25 px, percentile 95.0; median size 308 concepts, percentile 99.5.
• F07: 21, 38, 0, 0; does-not-compose percentile 45.9; median area 770.0 px, percentile 95.3; median size 145 concepts, percentile 98.8.
• F08: 21, 29, 0, 0; does-not-compose percentile 46.6; median area 821.5 px, percentile 96.2; median size 123 concepts, percentile 87.8.
• F09: 22, 32, 0, 0; does-not-compose percentile 45.5; median area 464.25 px, percentile 16.2; median size 127 concepts, percentile 91.9.
• F10: 16, 22, 0, 0; does-not-compose percentile 47.4; median area 696.0 px, percentile 77.7; median size 93 concepts, percentile 27.3.
• F11: 18, 25, 0, 0; does-not-compose percentile 46.7; median area 985.5 px, percentile 99.2; median size 144 concepts, percentile 96.2.
• F12: 9, 13, 0, 0; does-not-compose percentile 48.6; median area 777.0 px, percentile 82.8; median size 114 concepts, percentile 68.6.
• F13: 2, 7, 0, 0; does-not-compose percentile 49.2; median area 3202.0 px, percentile 100.0; median size 239 concepts, percentile 98.2.
• F14: 4, 3, 0, 0; does-not-compose percentile 49.5; median area 588.0 px, percentile 50.7; median size 132 concepts, percentile 71.0.
• F15: 20, 22, 0, 0; does-not-compose percentile 46.8; median area 554.75 px, percentile 39.0; median size 86 concepts, percentile 13.8.
• F16: 13, 22, 0, 0; does-not-compose percentile 47.8; median area 1127.5 px, percentile 99.5; median size 129 concepts, percentile 85.2.
• F17: 12, 23, 0, 1; does-not-compose percentile 47.0; median area 601.25 px, percentile 51.9; median size 104 concepts, percentile 52.5.
• F18: 4, 11, 0, 0; does-not-compose percentile 48.9; median area 462.0 px, percentile 28.9; median size 123 concepts, percentile 71.1.
• F19: 6, 7, 0, 1; does-not-compose percentile 49.1; median area 563.0 px, percentile 45.4; median size 89 concepts, percentile 30.6.
• F20: 2, 7, 0, 1; does-not-compose percentile 49.5; median area 839.0 px, percentile 77.5; median size 90 concepts, percentile 34.3.
• F21: 21, 10, 1, 0; does-not-compose percentile 97.4; median area 406.0 px, percentile 10.8; median size 116 concepts, percentile 74.4.
• F22: 13, 17, 0, 0; does-not-compose percentile 47.6; median area 360.5 px, percentile 6.6; median size 109 concepts, percentile 62.4.
• F23: 8, 22, 0, 0; does-not-compose percentile 47.6; median area 2096.0 px, percentile 100.0; median size 147 concepts, percentile 94.5.
• F24: 12, 15, 0, 0; does-not-compose percentile 48.0; median area 655.0 px, percentile 69.7; median size 144 concepts, percentile 91.6.
• F25: 15, 11, 1, 0; does-not-compose percentile 98.0; median area 1237.5 px, percentile 99.1; median size 196 concepts, percentile 99.5.
• F26: 10, 13, 0, 0; does-not-compose percentile 48.4; median area 1008.0 px, percentile 95.9; median size 135 concepts, percentile 86.0.
• F27: 9, 18, 0, 0; does-not-compose percentile 48.0; median area 1389.5 px, percentile 99.4; median size 196 concepts, percentile 99.8.
• F28: 11, 11, 0, 0; does-not-compose percentile 48.6; median area 1005.5 px, percentile 96.2; median size 120 concepts, percentile 74.0.
• F29: 8, 17, 0, 0; does-not-compose percentile 48.2; median area 1800.0 px, percentile 100.0; median size 238 concepts, percentile 100.0.
• F30: 7, 17, 0, 0; does-not-compose percentile 48.2; median area 836.5 px, percentile 88.0; median size 139 concepts, percentile 88.4.
• F31: 12, 10, 0, 0; does-not-compose percentile 48.6; median area 663.25 px, percentile 67.0; median size 161 concepts, percentile 96.1.
• F32: 5, 16, 0, 0; does-not-compose percentile 48.5; median area 1158.0 px, percentile 98.0; median size 140 concepts, percentile 86.5.
• F33: 10, 13, 0, 0; does-not-compose percentile 48.4; median area 633.0 px, percentile 61.0; median size 86 concepts, percentile 22.6.
• F34: 9, 13, 0, 0; does-not-compose percentile 48.6; median area 442.0 px, percentile 22.7; median size 95 concepts, percentile 37.1.
• F35: 8, 10, 0, 0; does-not-compose percentile 48.8; median area 374.25 px, percentile 12.8; median size 98 concepts, percentile 43.9.
• VA: 64, 112, 0, 0; does-not-compose percentile 38.4; median area 1016.0 px, percentile 100.0; median size 182 concepts, percentile 100.0.
• IT: 165, 116, 1, 10; does-not-compose percentile 79.8; median area 312.0 px, percentile 0.0; median size 56 concepts, percentile 0.0.
• TX: 73, 82, 0, 0; does-not-compose percentile 39.3; median area 642.0 px, percentile 84.2; median size 91 concepts, percentile 5.4.
• CH: 28, 54, 0, 0; does-not-compose percentile 44.3; median area 914.25 px, percentile 99.8; median size 187 concepts, percentile 100.0.

Reading. The one room that does not compose, picturegame (out-of-scope), lies in F21 and F25, so their does-not-compose percentiles are high; every other group's is near the middle. The canvas percentiles follow room size: the groups of large rooms (VA, CH, F11, F23, F29) have large median areas, and IT, the cluster of small rooms, has the smallest. No cut rests on the canvas or the verdicts; each cut was snapped before these were computed (rule 7).

What was drawn, and what was not

• Drawn as views, from the carried distance: the average-linkage dendrogram (_views/dendrogram.nwk), a chain with no partition, and the leaf-to-root view (_views/adaptation-tree.html, .svg, .json), whose root is the frozen global model and whose JSON carries the branches and cross-links. Neither is read as a chronology, an ancestry or a validated hierarchy.
• Not drawn: the NeighborNet (outside the run, if asked), and any tree of forms. The branches overlap (759 cross-links), so they are not a tree; they are facets of the rooms' adaptations.

The final accounting (stage D4 section 4)

Against n_analysed = 705: 476 rooms in a branch, 0 in the unadapted group, 229 adapted but in no branch, 0 unreturned; the four sum to 705. The rooms in no branch are named, with their counts, in the accounting at D4a-02 above; no branch changed since.