Repository navigation
Research: the complex delivery layer is rigid — target hosting and raw-face sharing closed exactly - #651
Research: the complex delivery layer is rigid — target hosting and raw-face sharing closed exactly#651higherbeing wants to merge 1 commit into
Conversation
…d raw-face sharing closed exactly) Exact negative results on the K-triangle word: no target can host a face or 8-sum for other targets (cap = consumption, consistent ratios only across tetrads, own-row frame only in same-tetrad caps, own read cannot be deferred), and no read order of face registers removes the slot copies (pollution-tolerance census over all 168 incidences). Scripts run the pinned constructor unchanged and reproduce every number. Prepared with Claude Code (Anthropic) assistance.
|
Thank you @higherbeing. This contribution is now retained on main in the scoped 100-PR review and contributor credits. Contribution: Delivery-layer rigidity in the K-triangle word. Review and limits: Target hosting and raw-face-sharing exclusions retain the fixed monotone-frame, own-read-first model. CP-SAT bounds and unsuccessful searches are not a general coding lower bound. The complete original source, notices and relevant execution evidence are retained through the pinned source archive. The review distinguishes fresh local runs, source-bound historical Linux CI and prior exact-head review; it does not claim 100 fresh full mathematical audits. Closing this historical PR as reconciled. This preserves parallel and cumulative credit without replacing the selected frontier or promoting priced research targets to completed constructions. Reviewed head: |
2 participants
Research note; no κ claim. Adds only
research/complex-delivery-rigidity/(README, scripts, results, manifest). No supplier, certificate, verifier or existing file changes. Based onmainat17ce0c7.Two exact negative results on the K-triangle word (#622 lineage)
Slot copies and the in-place mixing of the 8-sums are about 15 % of the complex local rank mass. Two proposed ways to remove them, "group delivery" (my #520 §3) and target-hosted / output-borrowed registers (#571's open list), are closed on this word family:
No target can host a face or 8-sum for other targets. Checked on all 14 blocks from the pinned constructor, executed unchanged with a role dump: a target's cap contains an 8-sum's span iff it consumes it (112/112); two consumers have consistent coefficient ratios only in opposite tetrads (ratio
σ_x[p0]σ_z[p0], constant on their 3 shared keys), while same-tetrad pairs share 2 keys with opposite ratios; a target's own-row frame lies only in same-tetrad caps; and the own read cannot be deferred (own registers are never written after their last target read, but the module's next read lifts the frame out of the cap in all 112 cases). Hence every cross-tetrad target-to-target read is frame-illegal and same-tetrad sharing is the existing butterfly, once per target. A CP-SAT read-ordering model with exact atom flow confirms the 8-sum-level floor: pure registers need 20 copies per block (bound 12) against the 8 the word uses with 2×2 mixes.No read order of face registers removes the slot copies. In the raw word no register is read by two targets of the same block; all 120 multi-reader registers are Direct delivery with 504 helpers for conditional κ = 0.001517760977820331 #531's cross-block raw reads. The pollution-tolerance census over all 168 (key, consumer-block) incidences shows a target's frame after the butterfly, or after its own row plus the selected faces, is inside the caps of no consumer group (0/1344 each); only the bare own-row frame is tolerated: 160/448 foreign cases for selected incidences (56 are used by Direct delivery with 504 helpers for conditional κ = 0.001517760977820331 #531), 64/896 for non-selected ones, and 96 of the 112 non-selected incidences have no tolerated foreign target at any stage.
So under the hub's rules (monotone frames, target frames inside
u_t^⊥, helpers never reading targets, own row first) the delivery layer is at its floor; #646's measured loss from more skeleton atoms and #645's gain in the factory layer point the same way. The README states what a rule-level change would have to touch.Reproduction
Standard library only;
order_model.pyneedsortoolsand is skipped otherwise (its recorded runs are inresults/). The constructor fromresearch/complex-571-cold14-k492and the inputs fromresearch/direct-delivery504-globalare read in place and not modified.Credits and disclosure
icekylinx (#425/#433/#531 construction, SAME504 factory), jeremykothe2 (#622 constructor and schedule, hyperplane ends), DreamingOfClouds (#604 cold ends), DominikScholz (#571 factory floor and open list), chafreaky, eumemic and the #446 authors. Prepared by higherbeing with Claude Code (Anthropic) assistance; the analysis code was written with that assistance and is Apache-2.0.