Repository navigation
Execute actual Full tensor programs through physical H and exact diagonals (no new κ) - #6894
Merged
Mister-Meeseeks merged 1 commit intoNov 10, 2026
Conversation
Bind heterogeneous Full programs to a common global grid and an encoded physical caller. Include all source, scalar, admission and cleanup costs; retain native and outer analytic premises. No new kappa. Prepared for @CubeJerry with substantial OpenAI Codex assistance.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Connect the actual forward and stored-inverse Full programs of #4253 to a complete global tensor field and an encoded physical consumer. Complete compiler-generated identity layers synchronize coefficient-dependent depths across every axis and coarse label; a new common-grid execution includes all diagonal roundings.
For side N=2^m and maximum logical depth L, the word has L(8m+2) global H calls and L(8m+3) diagonals. The exact reference is the direct sum of the original tensor programs. Explicit source and arithmetic errors are at most 2^-t/8 and 2^-t/4. This extends #4178's generic caller to actual Full sources; it does not tensor nonlinear rounded trajectories.
The physical caller combines #3627's H service with #3709's exact WordProduct at every width. Four full real products precede each complex truncation. Input admission, source preparation, encoded controls, movement, owned returns and cleanup are charged. The scalar term remains explicit; it is absorbed in the H tariff when beta*.99827101>log2(3)-1, subject to all stated H-domain conditions. This is not an outer retuning.
Base: 7c49fe3. SOURCES.json and PHYSICAL-SOURCES.json pin every source by Git blob and SHA256, including byte-identical boundary files from 271f1d4.
Reproduce (Python 3.12, sympy==1.13.3):
Expected: PASS_REGISTERED_RESEARCH_PACKAGE research/full-global-tensor-caller outcome=completed.
Executed: all six registered commands, independently rerun on the frozen joint commit. Source checks cover 11 trajectories and 1,351 nonzero roundings. The actual stored inverse of (2+i)I2 executes all 21 stages at unchanged F407/W608, including 92 exact word products, in normal and optimized modes. It has source error but zero nonzero stage roundings; a separate three-axis/two-label physical service fixture has 40. Twelve source controls, six endpoint/radius mutants, fourteen private-port guards, warm-module rejection and observer-erasure checks pass. Complete remote tree matches the tested 21-file addition.
No new κ. Native, stock, exact-integer and ownership contracts remain inherited. The physical H witness uses disclosed reduced q2 bindings and finite certified metadata folds; it is not production native execution. Outer Gaussian/cohort admission, selected-rate dispatch and full integer recovery remain open. Full-history preflight was not run in the shallow checkout.
Prepared for @CubeJerry with substantial OpenAI Codex assistance and coordinated independent AI review. Credit Jose Manuel/jmbarrancoml (#4205/#4253/#4178/#3627/#3709), the retained native/compiler/ownership authors and the source-owned boundary contribution. Apache-2.0.
Proofs, exact costs, receipts, source pins and open questions:
https://github.com/CubeJerry/integer-mult-bounds/tree/d51859e7816771f627033b23db0f9786ada863f8/research/full-global-tensor-caller