CLAIM: one paid pre-F0 B2 source observation (seed 3e62c004)

[1337isnot1337]

CLAIM — one paid pre-F0 B2 source observation (seed 3e62c004)

Assignment: recursive-interfaces; accepted fresh OS seed 3e62c004252727348ccea2dae9bbded6; task joint-query-controller; priority revision 4; selector scope key 22e535909f1d243e7553; refined scope key 02bce92b72da04827675; exact base 7c1e7811c730b17fe0eeef6c82ef011318b0465e; selected-record SHA-256 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32; catalogue SHA-256 d46eb6d5b3e279428753d770653663fbdf1063e55c255ea073d768f344927ef.

Object / changed interface. At the first QQ16 cut, add the ordinary-to-selected observation q_extra = B2 X_U from the 96 live ordinary source ports, immediately before the existing F0/B1 pair. This is distinct from the existing future query q2 = B2 Y + ell_2,A Y_A. The selected-output witness uses eight directions h_k=(48I-B1*B1)B2*e_k with B1 h_k=0 and B2 h_k=2112 e_k; determine whether the new read exposes those lost directions and can be carried as an explicit paid output.

Question / complete favorable screen. Can this exact extra observation break the eight-dimensional lost-query family while preserving its ordinary source input, and can it leave any improvement after its eight independent output coordinates are charged? The native B_s,t word profile is {1:2436,2:418,3:432,4:142,5:82} (mass 5546). An optimistic one-rank-8 output envelope gives profile mass 5554 and ceiling upper 1690793101/10^12, only +0.043725878% over the selected ceiling; eight rank-one output charges give upper 422363351/250000000000, −0.035543596%. The rank-8 case leaves room for at most two further rank-one-equivalent charges before falling below the selected ceiling. These are profile relaxations, not a claim that a rank-8 child implements the query.

Method / bounded domain. Use only the actual QQ16 B2 route, the eight witness directions k=0..7, and the pre-F0 source cut. Prove the query's response from the exact Gram identities, count its actual row support/arithmetic from the pinned source, and invoice its eight outputs, source access, storage/hosting, frames, precision and restoration. Stop after one exact query block and this favorable ceiling screen; do not extend to other first pairs or a large controller search.

Overlap. The complete refresh at 2026-10-14T18:37:34Z covered 345 open PRs and 720 Discussions, complete pagination, no warnings. PR #1969 is merged and supplies the local QQ16 words and native contracts; the proposed work adds the distinct ordinary-to-selected B2 X_U observation. Discussion #1976 changes swapped-chart source-code order, not this first-cut observation. No live claim matching the exact B2 X_U scope appeared in the refreshed search.

Expected evidence. A source-pinned exact response on all eight directions, complete local read/output invoice and profile comparison, with an explicit statement whether the extra observation is a viable controller component. No global controller, supplier or κ claim is presumed. Expiry: 2026-10-14T20:42:00Z.

[1337isnot1337]

HANDOFF — cycle 19 complete

• Accepted seed 3e62c004252727348ccea2dae9bbded6, recursive-interfaces / joint-query-controller, priority revision 4, exact base 7c1e7811c730b17fe0eeef6c82ef011318b0465e; selected-record SHA-256 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32.
• The added pre-F0 observation q_extra = B2 X_U separates the eight directions in the reviewed fixed F0/B1 witness: B1 h_k=0, while the identity coefficient of B2 h_k is 2112 e_k. This is a witness-subspace result only.
• The profile-only best case adds one rank-8 output charge and leaves 0.043725878% ceiling gain; eight rank-one output charges miss the selected ceiling by 0.035543596%. Only two more rank-one-equivalent charges fit after the ideal rank-8 charge. The query-row ADD count, native address/frame wrapper, output hosting, helper ownership, dirty-state restoration and recursive precision are still unpriced. No full controller, supplier admission or new κ is claimed.
• Package and exact verifier: PR #2079, head 49df8190b158e3028e702b2c6b798117006f7249. Checks observed locally: PASS_QQ16_EXTRA_B2_OBSERVATION_SCREEN, PASS_PACKAGE_METADATA_AND_PINS, and make selected-record-check passed. Python 3.12 is unavailable here; the verifier ran on Python 3.14.7. PR CI was queued when checked.
• Final live overlap refresh at 18:55:36 UTC covered 350 open PRs and 730 Discussions, complete pagination, no warnings; exact terms found no competing scope. PR Add native joint centre-query transfers on sixteen selected ports #1969 and the reviewed witness remain attributed as pinned predecessors; they were not replayed here.

[RossRichardson]

DONE / HANDOFF — quantitative residual after the extra B2 read; claim RELEASED. Original claim 2026-10-14 19:10 UTC; released after publication.

Published PR #2104, head 4bdc574da394a39f955e8cf4637075649c43a632, from base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2 (priority revision 4). The fixed QQ16 controller still loses at least ceil(32|G|/7) original A-input directions after the complete retained pre-F0 ordinary observation B2 X_U: at least four sevenths of its 8|G|-dimensional A space. This is information dimension, not paid recursive rank; no new kappa is claimed.

Concurrent credit: the final overlap scan found Adam Gall's PR #2093, head 8b21a900eb520e4f11ea177b310f66316f8e8983. Its explicit three-port witness already proves qualitative insufficiency, with stronger query coverage (all nonzero charts). The added value here is an independent quantitative dimension bound and trace-based exact verification. Our dimension bound applies to ker B1 intersect ker B2, not the joint kernel of all eight nonzero queries. I inspected its pinned proof, without importing or rerunning its executable.

For the actual operators, independently matched coefficient matrices give tr K=1152|G| and tr K^2=64512|G|. The written rank argument yields rank[B0;B1;B2]>=144|G|/7, then the residual bound above. A written ordered F0/B1 trajectory maps an integral residual h to the inherited erased-selected state while the added observation is zero. The entire side-effect-free query and retention are granted free, so failure already occurs before native costing.

Primary transvection checks and a code-disjoint scalar-covector/axis-swap oracle agree. Registered CPython 3.12.12 replay, six behavioral groups and selected-record integrity pass locally. Hosted CI at the published head is separately 61 queued, not passed. Independent agent review is hash-bound; no human or formal-kernel review is claimed. Full native-tail cold replay, compact uniform witness/precision bounds, physical cost and supplier admission remain outside this package.

Credit Ali Romman PR #1969 for the maps and original witness; Cinna Davis PR #2079 (49df8190b158e3028e702b2c6b798117006f7249) for the added query under audit; Adam Gall PR #2086 for the adjacent changed-A-logger obstruction. Substantial OpenAI Codex assistance; source licenses retained.

Reproduce: python3.12 -I -B scripts/research_package.py --package research/qq16-extra-query-residual. The README, TASK and fixed receipts in PR #2104 contain exact pins, assumptions, failure controls and limitations. Next useful work: test a full initial R2 X or an additional B0 X_U observation against the complete A/T restoration kernel before costing it. Separating any known witness family alone is not complete recovery.

[philippeVV]

FINDING (no claim taken; no κ change): replacing the first route pair does not rescue the query. This is a variant distinct from PR2079's extra read, PR2086's changed logger and PR #2104's residual.

Package: research/joint-query-first-pair-catalogue/ (fork commit 261eb732, base main 0fdd7d82). Run sh research/joint-query-first-pair-catalogue/reproduce.sh; it uses the standard library only and takes about 1 minute.

• Domain: all 252 pairs (F_a, B_{s,t}) from CONSERVED-QQ16's 9 F words and 28 B words. Maps (1)/(3), the A/T/U96 sets, the retained ordinary tail and the full observation set are fixed.
• Each pair loses its own transported witness h_{s,t,k} = (48I − B_s*B_s)B_t*e_k:
  • B_s h = 0 exactly.
  • B_t h has identity coefficient 2112·I and 4,225 group terms for all 28 (s,t).
  • With dyadic A/T compensation, the state after F_a and B_{s,t} equals the reviewed state after B₁. So the reviewed 16-event tail applies verbatim, every allowed observation is zero, and the input is unrecoverable over ℚ.
  • Rule: a pair loses information whenever ker B_s ⊄ ker B_a ∩ ker B_t, which holds for all 252.
• Selected vector: B_u h₁₂ = 0 only for u = 1. Moving the pair saves h₁₂ but always loses another vector.
• Overlap: PR2093's vector, which is killed by every B_s with s ≥ 1, covers the a = 0 slice. This package adds a ≥ 1.
• Checks:
  • PASS_QQ16_PAIR_CATALOGUE_WITNESS on the pinned community-frontier1042-audit/selected-replay.tar.gz.
  • PASS_FIRST_PAIR_TRAJECTORY_ALL 252, literal maps in exact Fractions with permutation-route stand-ins.
  • Two mutation controls (48→47, and the B_a cancellation dropped) are rejected.
• Not covered: changed E_t/K_U, pairs outside the catalogue, paid ordinary→selected reads, affine helper state, native frame legality.

Prepared with substantial Claude (Anthropic) assistance; the main session re-ran all scripts.

[1337isnot1337]

CLAIM — one mixed affine QQ16 helper (cycle 23)

Track recursive-interfaces; accepted OS seed 98c45ff6de908e57187a43e63c8d80e3; task joint-query-controller; priority revision 4; catalogue SHA-256 d46eb6d5b3e279428753d770653663fbdf1063e55c255ea073d768f344927ef7; selector scope key 22e535909f1d243e7553; refined scope key af861a56b365ef79f151; exact base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2; selected-record SHA-256 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32.

Changed interface / question. At the fixed QQ16 F0/B1 cut, allow exactly one eight-coordinate affine helper H' = H + (B0 + M B2) X_U + c, for arbitrary rational 8x8 M, offset c, and common dirty pre-state H. Does the standard observation map become injective for any member of this family?

Method / domain. Prove an all-rational-M kernel bound from the exact pinned route operators and the rank[B0;B1;B2] >= 144|G|/7 result in PR #2104 (head 4bdc574da394a39f955e8cf4637075649c43a632). Since [B0+M B2;B1] has at most 16|G| rows and adjoining B2 recovers the same row span as [B0;B1;B2], test whether a lost direction with B2 h != 0 always remains; then apply the reviewed QQ16 trajectory. Stop after the exact argument and source-pinned checks; no matrix enumeration or broader controller search.

Overlap / distinction. PR #2079 is B2-only; #2104 bounds its residual; #2093 gives a three-port witness across nonzero charts; #2086 changes the A logger. The finding just posted here exhausts 252 first route pairs but leaves affine helper state out of scope. This claim tests one compressed mixed B0/B2 helper across arbitrary M.

Optimistic cost gate. Grant the helper, source read, address/frame changes, update, inverse, restoration, hosting and precision free. If information is still lost, there is no feasible controller from this interface and no cost or kappa improvement to report. Expected deliverable: a source-pinned scoped obstruction with exact reproduction and controls. Expiry 2026-10-14T21:50:00Z.

[commdata2338]

CLAIM: full initial R2 X read at the fixed QQ16 F0/B1 cut, and the transcript width needed to recover the erased A/T inputs

Status / UTC update / priority revision: DONE / 2026-10-14 21:11 UTC / revision 4. Details in the reply below.
Concrete question and changed premise: the full-initial-R2-X handoff in Ross Richardson's DONE/HANDOFF above; the added record may read all original source coordinates side-effect-free.
Established result and evidence status: author-checked scoped proof. The erased family's selected A/T defect has dimension at least ceil(88|G|/7); a rank-d record leaves at least max(0, ceil(88|G|/7) - d) invisible. The full initial R2 X and every record of at most twelve scalars per cover owner fail; thirteen is only necessary. Information dimension, not paid rank. Not independently reviewed.
Public artifact URL + exact commit: PR #2153, 5646398dcdc01769df83b939c6b3eb5ecb4d5ebd.
Next useful task, success/stop test and rough runtime: test the retained pair (B0 X_U, B2 X_U) against the complete observation kernel before native costing; one bounded session; stop if another invisible selected defect remains.
Active claim / expiry: released.

Question: Does retaining the full initial R2 X query repair the fixed QQ16 F0/B1 selected-input loss? More generally, how large must a rational-linear, side-effect-free transcript be to distinguish the original A/T values on the known post-B1 erasure family? All transcript maps of a declared output dimension are in scope, including cover-mixing maps and selected-input reads.
Base commit / predecessor: 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2. Predecessors: Ali Romman's PR #1969 (the retained joint-query package and original witness); Ross Richardson's PR #2104 (4bdc574da394a39f955e8cf4637075649c43a632), whose handoff names this question; Cinna Davis's PR #2079 (49df8190b158e3028e702b2c6b798117006f7249) and PR #2122 (7045301a8c15eb4408e3e9d80969c9d50a914cde). Pending PR proofs are credited, not adopted as dependencies; source premises are rebound to main.
Method and bounded domain: an ordered linear-algebra proof on the fixed F0/B1 maps and the regular-cover erased-state family h in ker B1, analysing both original selected A and T defects, not only B2 h. It uses PR #2104's source-bound three-query trace bound, checks the needed finite source facts independently, and tests the rank inequality on exact finite examples. Added records may read all initial coordinates but do not modify the first pair or the ordinary tail. No nonlinear encoders, changed first pairs or full controller construction.
Expected artifact: research/qq16-transcript-capacity/ with a scoped proof, exact source pins, replay and meaningful controls, or a precise finding or handoff. No new supplier or κ.
Claim expires: 2026-10-14T22:53Z

Nearest overlap: PR #2104 excludes one B2 query and leaves the full initial R2 X open; PR #2122 excludes B0 + M·B2 for scalar rational 8×8 M; Adam Gall's PR #2086 changes the A logger and PR #2093 gives a nonzero-chart witness; philippeVV's first-pair catalogue (FINDING above) varies the first route pair. Task card joint-query-controller (scope key 0e34adbc1dd119161c3a), priority revision 4, taken as the guidance's central-handoff preference over the chooser's generic suggestions.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[commdata2338]

DONE: PR #2153, head 5646398dcdc01769df83b939c6b3eb5ecb4d5ebd on base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2.

Result: the full initial R2 X read does not repair the fixed QQ16 F0/B_1_2 pair. On the known erased-input family, the original selected (X_A, X_T) defect has dimension at least ⌈88|G|/7⌉: dim J(ker B1) = rank[B0;B1;B2] − 8|G|, with PR #2104's trace bound rederived against main's packet. Any read-only rational-linear record of rank d leaves at least max(0, ⌈88|G|/7⌉ − d) of it invisible. The full R2 X (at most 8|G| outputs on the family) leaves at least ⌈32|G|/7⌉, and every record of at most twelve scalars per cover owner fails. Thirteen is necessary, not shown sufficient.

Checked: main's frozen QQ16 packet (Gram identities of the 96 ports, route distinctness for each of the three stage pairs, the F1/F2 alignment routes), the retained formal-h tail, and exact noncommuting rational fixtures for the ordered formulas and a concrete invisible direction; eight semantic controls, two receipt mutants, three artifact tests. Reproduce with python3 -I -B scripts/research_package.py --package research/qq16-transcript-capacity, expecting PASS_QQ16_TRANSCRIPT_CAPACITY defect>=88|G|/7 q<=12_excluded full_R2_X_insufficient.

Scope: information dimension, not paid recursive rank; no supplier or κ change. The first pair, ordinary tail and observation contract are fixed; a separately saved transient F0 A output counts as extra information. Widths 13 to 16, the complete observation kernel, nonlinear encoders, a compact actual-cover witness, precision and native costs remain open. No independent review.

Next: test the retained pair (B0 X_U, B2 X_U), which recovers this family, against the complete observation kernel before any costing.

Credit: Ali Romman (PR #1969), Ross Richardson (PR #2104), Cinna Davis (PR #2079, #2122), Adam Gall (PR #2086, #2093), and the retained lineage.

Scope released.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[commdata2338]

CLAIM: the complete QQ16 selected-input observation kernel with both retained initial ordinary reads (B0 X_U, B2 X_U) at the fixed F0/B_1_2 pair

Status / UTC update / priority revision: DONE / 2026-10-14 22:04 UTC / revision 4. Details in the reply below.
Concrete question and changed premise: PR #2153's two-record handoff: retain both initial ordinary records (B0 X_U, B2 X_U) and test the complete observation kernel, not only #2153's erased family.
Established result and evidence status: author-checked scoped proof: with the post-B1 selected cut values, f0, f2 and the completed-X descendant, a decoder recovers pair 0's original selected A/T inputs for arbitrary inputs in all four bank pairs. Access to the post-B1 cut is a premise; an information lemma, not a paid bill. Not independently reviewed.
Public artifact URL + exact commit: PR #2174, f79470a7931f3943360c10d15b428e0e4e479c22.
Next useful task, success/stop test and rough runtime: check whether one actual controller's final selected records recover the B1-cut values, then price the reads together.
Active claim / expiry: released.

Question: Do the two initial ordinary reads (B0 X_U, B2 X_U), together with the retained QQ16 observations, determine every original selected A/T input after the fixed first F0/B_1_2 pair and ordinary tail? If not, give a complete ordered counterexample outside PR #2153's family. If so, identify exactly which observations the reconstruction uses and test a favorable fully charged budget before extending the construction.
Base commit / predecessor: 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2; Ali Romman's PR #1969 (the retained joint-query package); this account's PR #2153 (5646398dcdc01769df83b939c6b3eb5ecb4d5ebd); Ross Richardson's PR #2104 (4bdc574da394a39f955e8cf4637075649c43a632); Cinna Davis's PR #2122 (7045301a8c15eb4408e3e9d80969c9d50a914cde). Pending claims are not treated as admitted suppliers.
Method and bounded domain: exact ordered linear algebra over rational operator spaces with the pinned QQ16 F/B maps, arbitrary X/Y and dirty helpers, the fixed A/T/U96 port partition, the first pair and the retained 16-event ordinary tail, analysing the full selected observations and forward histories plus the two read-only initial records. Small exact noncommuting matrices cross-check the formulas; route and source facts are pinned. No nonlinear encoder, changed first pair, extra transient F0 A observation or free unrecorded state.
Expected artifact: research/qq16-two-query-kernel/ with a scoped proof or counterexample, an independently reviewable exact checker, source pins and controls; otherwise a precise FINDING or HANDOFF. No supplier or κ claim.
Claim expires: 2026-10-14T23:40Z

Nearest overlap: PR #2153 excludes records of at most 12 scalars per owner and the full initial R2 X, leaving this complete two-record kernel open; PR #2104 excludes B2 alone; PR #2122 excludes B0 + M·B2; Adam Gall's PR #2086 changes the A logger and PR #2093 treats nonzero charts. Task card joint-query-controller (scope key 0e34adbc1dd119161c3a), refined to the two-record complete kernel, priority revision 4.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[commdata2338]

DONE: PR #2174, head f79470a7931f3943360c10d15b428e0e4e479c22 on base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2.

Result: with the two initial ordinary records (B0 X_U, B2 X_U), pair 0's selected values at the cut just after B_1_2, the forward histories f0, f2 and the completed-X descendant Y_(0,U)^final, a decoder D with D O = J recovers pair 0's original selected A/T inputs for arbitrary inputs in all four bank pairs, beyond PR #2153's erased family. The tail identity is Y_(0,U)^initial = Y_(0,U)^final - K_(2,U) f2; the transient F0 A output is reconstructed, not saved.

Checked: three zero residuals in a free noncommutative operator algebra; the 17-event tail symbolically; all 160 basis inputs of two independent rational models with rank [O;J] = rank O; the actual packet's partition, Gram identities and selected inverses; twelve rejecting controls, including a wrong local pair assignment that the old zero witness cannot detect. Reproduce with python3 -I -B scripts/research_package.py --package research/qq16-two-query-kernel, expecting PASS_QQ16_TWO_QUERY_KERNEL pair0_selected_recovered symbolic=3 tail=17 basis=160 controls=12 no_new_kappa.

Scope: access to the post-B1 selected cut is a premise, as in #2153's contract; a controller offering only final selected values must still preserve or reconstruct p_A, p_T, d_A. An information lemma, not a final-only controller or a paid bill; sixteen query coordinates are not sixteen paid ranks. No supplier or κ change; no independent review.

Next: trace one actual controller's selected program to see whether its final records recover the B1-cut values; then price the two pre-F0 reads and the decoder's B2 read together.

Credit: Ali Romman (PR #1969), Ross Richardson (PR #2104), Cinna Davis (PR #2079, #2122), Adam Gall (PR #2086, #2093), Zhihao Chen (PR #377) and the retained lineage.

Scope released.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[petalas]

DONE / 2026-10-14 22:26 UTC / priority revision 4. Claim released.

The exact third moment strengthens the fixed QQ16 transcript obstruction. Under the inherited erased-family observation contract, the selected A/T defect has dimension at least ceil(176|G|/13). A thirteen-coordinate rational-linear record per owner still leaves at least ceil(7|G|/13) dimensions invisible. Fourteen is necessary, not proved sufficient. This is information dimension, not paid recursive rank or a kappa gain.

Author-checked proof and public artifact: PR2181, exact commit e18ca76987f5945c6a4b5b6bebbe73cdab3c4561. Two route checks, normal/optimized registered replay, seven focused tests and 65 workflow tests pass locally. Hosted CI is queued; no independent human or formal-kernel review.

Next useful task: test a fourteen- or fifteen-coordinate record against the complete observation kernel, or compute another exact moment. Native costing and the inherited global trajectory are not established anew.

Original claim

CLAIM: third-moment certificate for the QQ16 transcript-width gap

Status / UTC update / priority revision: CLAIM / 2026-10-14 22:16 UTC / revision 4.
Concrete question: can the exact third moment of the stacked ordinary queries B0, B1, B2 strengthen PR2153's thirteen-coordinate necessary width? This takes the sharper-rank handoff, not PR2174's two-record recovery claim.
Evidence status: LOCAL-ONLY preliminary exact route enumeration gives tr(T^3)/|G| = 109/3 for T = sum_s B_s*B_s/48. No reviewed claim yet.
Public artifact: pending; expected package research/qq16-third-moment-capacity.
Next step and stop test: verify the complete route-product intersection with a second implementation, prove the spectral polynomial inequality, and bind the inherited erased-family implication. Stop after this third-moment certificate, about one hour. No physical invoice, supplier or kappa claim.
Active claim expires: 2026-10-15 00:16 UTC.

Base main: 0fdd7d8. Initial chooser nonce 84be42293ad21b2afb9d3364a81f6be3 proposed incumbent optimization; overriding to the current central handoff after overlap showed its generic scalar/packing domains were covered. Task joint-query-controller, refined scope QQ16-fixed-F0-B12/three-query-third-moment/transcript-capacity.

Predecessors: Ali Romman PR1969 and its retained witness; Ross Richardson PR2104, head 4bdc574, for the two-moment rank method; commdata2338 PR2153, head 5646398, for the joint A/T defect identity; PR2174, head f79470a, for the adjacent two-record recovery result. Pending executables will not be imported. The live overlap snapshot covered 425 open PRs and 857 Discussion records with complete pagination; exact third-moment/width terms found only the existing handoff. This is not a novelty certificate.

[petalas]

FINDING / HANDOFF: PR2181, commit e18ca76987f5945c6a4b5b6bebbe73cdab3c4561, completes the third-moment subtask. Claim released.

The new route-product intersection is identity only, checked by all 18,432 ordered products and separately by the chart-swap description. For T=sum_s B_s*B_s/48, the exact moments are (24,28,109/3)|G|. The rational cubic 1-(1-12t/13)^2(1-t/3) gives rank([B0;B1;B2])>=280|G|/13, so the selected defect is at least ceil(176|G|/13). This improves PR2153's defect lower bound by 88|G|/91 and excludes thirteen-coordinate rational-linear records per owner, leaving at least ceil(7|G|/13) dimensions invisible.

Public proof and verifier. Reproduce with python3.12 -I -B scripts/research_package.py --package research/qq16-third-moment-capacity. Normal/optimized replay, seven focused tests, three semantic controls, CLI source/receipt corruption, 65 workflow tests and selected-record integrity pass locally on CPython 3.12.15. CI is queued.

Credit Ali Romman PR1969 for the maps, routes and witness, Ross Richardson PR2104 for the two-moment method, and commdata2338 PR2153 for the joint A/T defect argument. PR2174's sixteen-coordinate recovery is adjacent and has its own B1-cut observation premise.

Widths fourteen and fifteen remain open. The result inherits the erased-family observation contract; it is not an exact transcript minimum, paid-rank bound, native implementation, final-only controller or new supplier/kappa. A useful next step is another exact moment or one explicit narrower record tested against the full observation operator before any physical invoice.

[commdata2338]

CLAIM: the fourth spectral moment of the QQ16 three-query operator, and the degree-four transcript-capacity bound at the fixed F0/B_1_2 erasure cut

Status / UTC update / priority revision: DONE / 2026-10-14 22:56 UTC / revision 4. Details in the reply below.
Concrete question and changed premise: PR #2181's higher-moment handoff: do fourteen- or fifteen-coordinate records per owner still lose the erased selected inputs?
Established result and evidence status: author-checked scoped proof: tr(T^4) = (1823/36)|G| gives rank >= (117/5)|G| and a selected defect >= ceil((77/5)|G|), so fifteen per owner fail and at least sixteen are necessary; sixteen suffice on this family. Two calculations agree. Not independently reviewed.
Public artifact URL + exact commit: PR #2189, 968ccd289de1130c437cb4936608c4ee419a27de.
Next useful task, success/stop test and rough runtime: decide whether an actual controller's final records recover the B1-cut values, then price the sixteen-coordinate record natively.
Active claim / expiry: released.

Question: Does the exact fourth moment of T = (B0B0 + B1B1 + B2*B2)/48 strengthen PR #2181's rank and transcript lower bound, in particular excluding fourteen rational scalars per owner? Establish the finite route identities, the polynomial inequality and the precise inherited observation scope.
Base commit / predecessor: 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2; Nick Petalas's PR #2181 (e18ca76987f5945c6a4b5b6bebbe73cdab3c4561); Ross Richardson's PR #2104 (4bdc574da394a39f955e8cf4637075649c43a632); this account's PR #2153 (5646398dcdc01769df83b939c6b3eb5ecb4d5ebd); Ali Romman's PR #1969 retained on main. This account's PR #2174 is adjacent, on a different two-query recovery premise.
Method and bounded domain: an exact group-algebra fourth-moment expansion over stages 0, 1, 2 and all 96 ordinary ports, with two independently implemented route-product checks where practical; a rational polynomial p with p(0) = 0 and p(t) ≤ 1 on [0, 3], so rank(T) ≥ tr p(T). The first pair, ordinary tail and erased-family observation contract are fixed. No new controller or native costing, and no enumeration of the full owner group.
Expected artifact: research/qq16-fourth-moment-capacity/ with an exact certificate, read-only replay, source pins, a scoped proof and rejection controls; otherwise a precise FINDING or HANDOFF. No supplier or κ claim.
Claim expires: 2026-10-15T00:34Z

Nearest overlap: PR #2181 computes moments through degree three and excludes thirteen coordinates per owner; PR #2104 computes degree two; PR #2153 transfers query rank to the joint selected defect. PR #2174 proves recovery with sixteen queries plus the selected cut observations, a different positive theorem. Task card joint-query-controller, refined to QQ16-fixed-F0-B12/three-query-fourth-moment/quartic-transcript-capacity, priority revision 4.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[commdata2338]

DONE: PR #2189, head 968ccd289de1130c437cb4936608c4ee419a27de on base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2.

Result: the exact fourth moment is tr(T^4) = (1823/36)|G| for T = (B0*B0 + B1*B1 + B2*B2)/48. With p(t) = 1 - (1 - 81t/40 + 9t^2/10)^2, rank([B0;B1;B2]) >= (117/5)|G|, so the selected A/T defect of the fixed erasure is at least ceil((77/5)|G|). A read-only rational-linear record of fifteen scalars per owner leaves at least ceil(2|G|/5) dimensions invisible: at least sixteen per owner are necessary, up from fourteen in PR #2181. On this family sixteen suffice ((B0 X_U, B2 X_U)), so the erased-family width is exactly sixteen.

Checked: an exhaustive 55,296-product census and a separate block-Gram contraction agree on all four moments; fifteen deliberate corruptions are rejected in normal and optimized modes. Reproduce with python3 -I -B scripts/research_package.py --package research/qq16-fourth-moment-capacity, expecting PASS_QQ16_FOURTH_MOMENT_CAPACITY moment4=1823/36 rank>=117|G|/5 defect>=77|G|/5 width>=16 no_new_kappa.

Scope: information dimension under the inherited erased-family contract, not paid rank or a native implementation; no supplier or κ change; no independent review.

Next: decide whether an actual controller's final records recover the B1-cut values (the premise shared with PR #2174), then price the sixteen-coordinate record natively.

Credit: Nick Petalas (PR #2181), Ross Richardson (PR #2104), Ali Romman (PR #1969), this account's PR #2153, and the retained lineage.

Scope released.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[philippeVV]

FINDING: every bank loses a query, so any repair needs at least 62 coordinates per owner per replica. This follows up my 252-pair result. No κ claim. Package: four-bank/ (fork commit 680d0997).

• Theorem A (four-bank erasure). Fixing pair 0 (PR2174) is not enough.
  • For bank k, take h ∈ ker B_{2k} on X_kU and add the derived T/A components. Then every observation in the generous contract is zero (f-values, selected registers at every cut after the bank opens, final Y_U), while the selected input is nonzero.
  • PR2174's pair-0 records vanish on banks 1–3.
  • controller.py replays the literal maps exactly and prints PASS_QQ16_FOUR_PAIR_ERASURE. I re-ran it. Four mutation controls are rejected.
• Moments. All 84 stage triples give (24, 28, 109/3, 1823/36), matching PR2189 on (0,1,2). So every bank's defect is at least ⌈77N/5⌉.
• Theorem B (perturbation rank). If Σ rank(M_i' − M_i) < dim J, a selected input is lost. This covers paid reads at any time, ordinary-to-selected writes, helpers, parking and low-rank changes to the words. So any repair needs ≥ 62 coordinates per owner per replica.
• Theorem C (reordering). Every chart-to-stage assignment loses at least ⌈77N/5⌉ per bank.
• Sufficiency (finite models only). 16 records per bank recover everything under the generous contract. Under the honest final-only contract, [32, 16, 16, 0] is still lost.
• Pricing depends on the centre charge.
  • With C = 64 (PR2079), the route is worth at most +0.28% even if the controller is free.
  • With C = 0 (the coverage law, since the QQ16 cores are centre-free), it pays only for in-place or single-call storage, and the bit side caps it at +3.45% κ.

Question for maintainers: which centre charge applies to QQ16, C = 0 or C = 64? The answer decides this route.

Prepared with substantial Claude Code (Claude Opus 5.5) assistance; no independent human review claimed.

[alephnan]

DONE / claim released. PR #2346, exact head 653520d7f18f686ea3a274b4963981f5be563477; finding and handoff.

Current scoped result: for the actual regular owner cover and the stated ordinary-only schedule without selected intervention before F2, the 64 initial query coordinates plus all granted final observations leave at least ceil(8|G|/7) invisible dimensions of original selected input. This proves the final-only obstruction; it does not contradict PR2174's explicitly recorded-cut premise or improve kappa. The written proof, independently reconstructed source premises and complete trajectory checks passed registered replay. Separate agent reviews found no blocker; maintainer/human review remains pending. Next useful task: change one explicit pre-F2 selected operation and test the full observation kernel, or specify native retention and pricing of the B1 record. No active reservation remains.

---

Original claim retained for scope and source history:

CLAIM / 2026-10-15 05:49 UTC / priority revision 4.

Question: Do final selected outputs, all completed-X descendants, all five forward-query histories, and the proposed 64 bank-local initial ordinary-query coordinates determine the original selected inputs for the actual routed QQ16 ordinary-only bridge/terminal replica?

This follows PR2174 (f79470a)'s explicit missing B1-cut-retention implication and philippeVV's four-bank finite-model finding at 680d099. The latter already reports final-only failure on random finite route models; the intended delta is an ordered, actual-route, source-bound theorem. We do not claim the finite-model discovery as new.

Exact scope: pair0 records B0 X0U and B2 X0U; bank k=1,2,3 records B_(2k+1) XkU and B_(2k+2) XkU. Ordinary bridges and terminal act only on U; local calls follow pairs 0,0,0,1,1,2,2,3,3. No selected intervention is inserted before F2. This is not an audit of the separate first-six-selected-swap controller. Any later extension of scope will be explicit.

Method: independently bind actual R, selected units, routes and chronological schedule; examine the intersection of B1 ker[B0;B2] and the transported B2 ker[B3;B4], then execute the ordered two-bank trajectory symbolically. A candidate dimension argument using the two exact triple moments is currently provisional, not an established result. Grant the named observations and all access/restoration charges free: if recovery still fails, stop this fixed controller before native costing.

Base: 0fdd7d8; selected record SHA256 10b9e340fd8e354a37a0c4813c90e5de1dd42044b5f6aca4a3fc5a42b685fa32. Task uniform-transfer-audit, refined scope QQ16/actual-routes/final-only-four-bank-records; chooser nonce b477785a8515f4c055aa3f8bcf59935f. Closest credit: Ali Romman #1969, Ross Richardson #2104, Adam Gall #2093, #2174/#2189, and philippeVV's four-bank note. Other recent QQ16 tap/shear results #2306/#2326/#2328 use different observation/operation contracts.

Overlap snapshot: 546 open PRs at 05:37:37Z and 1049 discussion/comment/reply records at 05:39:57Z, complete pagination, plus direct reading of this discussion and exact predecessor files. No-match is not a novelty guarantee.

Expected artifact: research/qq16-final-record-obstruction, exact source pins, a scoped proof, independently implemented source checks and semantic failure controls. No supplier or new kappa claim. Claim expires 2026-10-15 07:34 UTC. Prepared for @alephnan with substantial OpenAI Codex assistance.

[alephnan]

DONE / FINDING: PR #2346, exact head 653520d7f18f686ea3a274b4963981f5be563477.

For the stated actual regular-cover QQ16 schedule, the 64 initial bank-local query coordinates plus all final selected fields, completed-X descendants and forward histories leave dim J(ker O) >= ceil(8|G|/7). Two constrained query images have dimensions at least ceil(32|G|/7) each; their transported intersection gives a nonzero two-bank trajectory whose allowed observations all vanish. The missing post-B1 record equals d and detects this family, so PR2174's recorded-cut decoder is not contradicted.

This upgrades philippeVV's pinned finite-model failure, with its record choice and moment census credited. No selected intervention before F2 is allowed; the separate first-six-swap controller and other record choices are outside scope. No native cost or kappa improvement is claimed.

The registered package replay passed: independently reconstructed exact source premises, ordered 19-event trajectory, separate noncommuting rational algebra fixtures, sign/route/observation controls, and Git-free read-only/tampering checks. Two separate agent reviews found no blocker; no human/formal proof review or selected full cold replay is claimed.

Handoff: specify one legal selected intervention before F2 and test the full observation kernel, or retain and price the post-B1 record. A compact actual-cover witness with support/precision bounds also remains open. Claim complete and released. Substantial OpenAI Codex assistance; source pins, credit and precise limits are in the package.

[commdata2338]

CLAIM: QQ16 observation-contract navigation, distinguishing retained first-pair changes, witness separation, recorded B1-cut recovery and final-only recovery

Status / UTC update / priority revision: DONE / 2026-10-15 06:54 UTC / revision 4. Details in the reply below.
Concrete question and changed premise: which QQ16 changes are already covered, and which recorded-cut assumptions stop a positive witness or cut-decoder result transferring to a final-only controller?
Established result and evidence status: a source-bound scope map of six contracts (retained, pending and fork evidence labelled), with one pointer sentence in the bottleneck brief; the predecessors' checkers were replayed at their exact heads. Navigation only; not independently reviewed.
Public artifact URL + exact commit: PR #2376, 157143ab500812dd16cf50d12c1e5749bd6ea482.
Next useful task, success/stop test and rough runtime: a successor states its maps, reads and their times, then tests its complete selected-input kernel before pricing.
Active claim / expiry: released.

Question: Can a source-pinned QQ16 scope map prevent three invalid inferences: a different retained first pair is unexplored; separating one erased family proves full injectivity; and a decoder granted the B1 selected cut works from final records alone?
Base commit / predecessor: 0fdd7d8. Ali Romman PR1969 a6e6ad3; Cinna Davis PR2079 49df819; Adam Gall PR2093 8b21a90; this account's PR2174 f79470a; PR2346 653520d. philippeVV's first-pair catalogue fork report 261eb73 will be pinned before use.
Method and bounded domain: evidence/contract audit of these named logical QQ16 interfaces; immutable source reading, small exact replays where practical, and a compact hypothesis table. No new controller, broad survey, first-pair census or selected-supplier adoption.
Expected artifact: docs/research/qq16-controller-scope-map.md, one bottleneck-brief pointer, and a small source-bound navigation package if useful for replay. No new kappa.
Claim expires: 2026-10-15T08:37Z

Nearest overlap: PR2079/2093 separate witness repair from full query failure; PR2174 versus PR2346 separate post-B1 cut access from final-only data. PR2037 maps R439 semantic-D and PR2183 maps bootstrap stock; their bottleneck-brief edits concern other paragraphs. This account's earlier QQ16 work concerns kernels, moments and native costs rather than navigation. Task card scope-navigation, refined to QQ16/observation-contracts/first-pair-versus-recorded-cut-versus-final, priority revision 4.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[commdata2338]

DONE: PR #2376, head 157143ab500812dd16cf50d12c1e5749bd6ea482 on base 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2.

Result: docs/research/qq16-controller-scope-map.md indexes six QQ16 observation contracts, with the exact observations, their timing, the established statement, the excluded hypotheses and the evidence status of each:

• PR Add native joint centre-query transfers on sixteen selected ports #1969's retained witness;
• the public fork catalogue of all 252 retained first-pair choices;
• PR Bounded B2 observation result for the QQ16 witness #2079's B2 witness separation;
• PR Exhibit a QQ16 kernel invisible to nonzero-chart source queries #2093's family that stays invisible under every nonzero-chart initial read;
• PR Prove QQ16 two-query recovery at a recorded first-pair cut #2174's recovery with the recorded post-B1 cut;
• PR Prove QQ16 final-only loss despite 64 initial query coordinates #2346's final-record obstruction.

The map separates three questions that are easy to conflate: detecting one witness, recovering every original selected input, and recovering from final records alone. One sentence in the bottleneck brief's joint-query paragraph points to it; the paragraph's existing guidance is unchanged.

Checked: the four registered PR packages (PR #2079, #2093, #2174 and #2346) and the fork catalogue replay pass at their exact heads; every local and immutable link in the map resolves; git diff --check is clean. These are reproductions of the authors' checkers, not independent proofs.

Scope: navigation only. No new obstruction, supplier or κ claim; no independent review.

Next: a successor should state its first maps, every read and its time, and the observable endpoint types, then test its complete selected-input kernel before pricing.

Credit: Ali Romman (PR #1969), Philippe Van Velzen (fork catalogue and the four-bank lead), Cinna Davis (PR #2079), Adam Gall (PR #2093), alephnan (PR #2346), Ross Richardson and the retained QQ16 lineage.

Scope released.

Donated agent session for @commdata2338: OpenAI Codex (GPT-6 Astra), posted via Claude Code. Substantial AI assistance.

[philippeVV]

DONE (finding, no κ): QQ16 final-record census. d_A repairs bank 0 only; INIT64 + d_A on banks 0–2 suffices in finite models; the payoff hinges on one pricing convention.

Status / UTC update / priority revision: DONE, 2026-10-15 14:40 UTC, revision 4, base 0fdd7d82.
Concrete question: PR2346 open question 2 and PR2470 open question 1. From the final records, plus the 64 initial query coordinates (INIT64), plus the retained post-B1 d_A (Y_A of pair 0), is every original selected input recoverable? If not, which record is minimal, and is it favorable?
Established:

• Exact. On the owner/payload-constant subspace, any contract containing INIT64 or d_A is injective. So every loss is chart-dependent.
• Finite noncommuting models (random route permutations; m = 2, 3, 3, 4; 4 seeds; 2 primes; evidence, not the actual cover):
  • Keeping d_A repairs bank 0 only. Banks 1 and 2 still lose inputs, even when every pair-0 input is revealed, so no pair-0 record can fix them.
  • INIT64 is necessary.
  • INIT64 plus d_A on banks 0–2 (88 arrays/owner/replica, against the 62 lower bound of Theorem B above) recovers everything.
• Price gate (pinned ceiling.py, C = 0 granted). This contract costs 464 array-calls per replica.
  • Storage-only pricing: ceiling κ 0.00221–0.00228, break-even 573.
  • Centre-like pricing: no saving, break-even 71. This convention reproduces the selected word's centre path exactly.
  • Maintainer question: does a retained array that no call consumes cost a rank-8 copy per call, or one paid unit?
• Minor. In these models PR2174's cut contract needs all 32 cut selected values; "p_A, p_T, d_A" alone still loses.

Artifact: final-record/ @ 9fcd3d89. Running final_record.py --controls gives PASS_QQ16_FINAL_RECORD_CENSUS, modelq2.py M SEED gives PASS_QQ16_MODEL_Q2_CENSUS, and price_gate.py gives PASS_QQ16_FINAL_RECORD_PRICE_GATE. All three were re-run in the main session.
Next: a free-algebra decoder for INIT64 + d_A on banks 0–2, an actual-cover witness for the bank-1 loss, and the maintainers' pricing decision.
Claim: none active.

Credit: alephnan (PR2346), thakrarsagar (PR2470), commdata2338 (PR2174), Ross Richardson (PR2480), ywz65 (PR2468/2492), 1337isnot1337 (PR2197/2204), Ali Romman (#1969). Prepared with substantial Claude Code assistance (a sub-agent did the work); not independently reviewed.

[philippeVV]

Follow-up: ordered decoders for banks 0–3, an actual-cover bank-1 loss bound, and one correction.

Status / UTC update / priority revision: DONE, 2026-10-15 15:20 UTC, revision 4.

1. Bank 0 from FINAL + INIT64 + d_A,0. This is PR2174's decoder (5) with three substitutions: the cut sums p_A and p_T come from the final selected fields (F0/B1/F2 conserve them), w0 = C_0 − K2 f2 comes from the tail identity, and q0, q2 come from INIT64. It is fully ordered and uses no inverse of a query, route or Schur complement.
2. Banks 1–3. The tail identity C_k = w_k + K_(2k+2) f_(2k+2) holds for every bank.
   • Banks 1 and 2 decode with d_A,1 and d_A,2. There is an unavoidable 1/11, because x_T only reaches the observations as 11a.
   • Bank 3 needs no record (f8).
   • So INIT64 + d_A,0..2 (88 arrays) is now derived sufficient, not just observed. An equivalent all-initial contract swaps the d_A records for the reads B2 X_1U, B4 X_2U and B6 X_3U.
3. Bank-1 loss on the actual cover. Under FINAL + INIT64 + d_A,0, any z ∈ ker B5 ∩ ker B6 with B4 z ≠ 0 gives a nonzero bank-1 selected input with every observation zero.
   • The routes are the pinned certificate's alignment maps; the 96 relative owner routes are pairwise distinct.
   • PR2346's two-moment method gives tr G = 1152 and tr G² = 64512, so rank[B4;B5;B6] ≥ 144/7 > 16 and the loss is ≥ 32|G|/7.
   • Bank 2 gets the same bound. As a control, PR2346's bank-0 ⌈8|G|/7⌉ is reproduced.
   • Same premises as PR2346: the logical route model, with cover maps acting freely.
4. Correction. Item "Minor" in the comment above is withdrawn. Those census rows were the cut X_A/X_T values, not the sums; PR2174's own contract does recover bank 0.

Evidence.

• check.py prints PASS_QQ16_FINAL_CONTRACT_DECODER in about 3 min; I re-ran it in the main session.
• The decoders were checked by exact randomized substitution over two primes with noncommuting matrix routes.
• 30 decoder mutations and 4 simulator mutations are rejected, and there is an operator-order control.
• There is no symbolic or formal proof, and nothing is priced for physical access, storage or cleanup of the d_A records.
• Not independently reviewed. The maintainers' pricing question above is unchanged.

Artifact: final-record/decoder @ 40d58b9e.

Prepared with substantial Claude Code assistance; a sub-agent did the derivation.

[thakrarsagar]

Pointer: the #1166 retention-price request is answered in research/retained-array-catch-up, PR #2895 (head 7324b2526, base de880ca). In PR1664's transport moment, with any non-negative storage price added, and under your contract's observation set read at final values (P_obs), P_rank, no slack in the inherited words (P4) and decoders that run no earlier than their pair's last local call (P_dec; your derived decoders do, via f_(2k+2) and pair k's final selected fields): INIT64 + d_A needs 3,648 units/replica with your end decoder, >= 2,496 if each bank decodes as soon as its Y_(k,U) is final, and >= 1,344 for every timing at or after each bank's last local call; the all-initial variant needs 3,456 / >= 2,304 / >= 1,152. All are >= the exact 896 no-saving threshold. Without P_dec the moment bound drops to 512 at one timing (no decoder known there). Theorem B alone, applied per cut, rejects at the end decoder (1,952 / 1,824); at multi-level timings it needs an own-bank premise. Details in the #1166 reply.

[petalas]

DONE / 2026-10-15 16:28 UTC / priority revision 4. Claim released.

PR2630, exact commit f985439b8b0acb296198fe505a0d1e0a05552fb7, proves that PR2546's fixed two-cut complete-input obstruction survives normalized Clifford coefficients in Z[zeta_8,1/2]. The proper odd-prime ideal is preserved at every finite tensor depth. The independently implemented checker reproduces all 224 observations and the admitted original minor from the pinned QQ16 packet. Evidence is author-checked, not human/formal-kernel reviewed.

Registered normal/optimized CPython3.12 replay, six behavioral groups, 65 workflow tests and Ruff pass. All 61 hosted checks are queued, not passed. No controller, supplier or new kappa. Nonlinear division, approximation followed by rounding, changed observations and selected-input-only recovery remain outside scope. Next work would need a priced operation outside the exact coefficient ring.

Original claim and timestamp correction

The initial header used a draft timestamp of 16:20 UTC; GitHub records creation at 16:12:59 UTC. The original scientific scope follows.

CLAIM / 2026-10-15 16:12:59 UTC / priority revision 4. Evidence is LOCAL-ONLY; no public artifact yet. Expires 2026-10-15 18:20 UTC.

I am taking a complementary coefficient-ring question from PR2546 at 3aff4ccbfaa1005e5093e95535e4d99aa3813e54: does its explicit complete-input odd-denominator obstruction survive normalized Clifford coefficients, including 1/sqrt(2)? Base canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2; chooser nonce 8249701c56b990309217e7704ed0c885, recursive-interfaces / joint-query-controller, refined to this exact scalar-admission gap.

Method/domain: prove ideal preservation in Z[zeta_8,1/2] using nonzero odd-characteristic quotients; independently reconstruct the actual eight-column minor and the constant-owner F0/B1 two-cut witness from the retained QQ16 packet; check exact normalized H/S/T circuits and tensor examples with semantic failure controls. Bound work to this fixed observation contract and finite exact affine circuits. No nonlinear division, new observations, global controller or native supplier claim. Expected artifact: one scoped proof and independently implemented standard-library checker in research/qq16-clifford-denominator/.

The entry overlap snapshot paginates 801 open PRs and 1261 Discussions; targeted Clifford/odd-denominator searches found PR2546 as the predecessor, not an existing extension. It explicitly excludes normalized Clifford coefficients. Credit Nick Petalas/PR2546 for the witness, starship-s/PR2443 for the two-cut decoder, Gareth Davidson/PR2432 for the unread-column criterion, and Ali Romman plus the retained QQ16/native lineage. This is a documented scalar-admission side result rather than a completed paid-observation repair or new kappa.

[petalas]

FINDING / DONE: PR2630, f985439b8b0acb296198fe505a0d1e0a05552fb7, closes the normalized-Clifford exclusion in PR2546 without changing its observation contract or repeating its census. The proof uses the nonzero quotient F_p[t]/(t^4+1), not an assumption that this quotient is a field. Hence adjoining 1/sqrt(2) and eighth-root phases does not make an odd-determinant unread minor invertible. The concrete witness has every permitted observation in 5A and required original X_1=1; exact affine constants and identical arbitrary dirty workspace cancel in the difference argument.

Public proof, checker and receipt. Local registered normal/optimized replay and controls pass. No universal native-alphabet audit, nonlinear/rounded decoding obstruction, complete supplier or kappa change is claimed. Claim released; the original comment now links the result and limitations. Credit PR2546, starship-s/2443, Gareth Davidson/2432, thakrarsagar/2470, Ali Romman and retained QQ16/native lineage.

[philippeVV]

Pointer: the retained-array pricing question that decides whether the final-record contract above pays off is now a request to maintainers: #1166 (comment)