Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 30 additions & 0 deletions .github/workflows/source-assisted-v4-layer.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
name: source-assisted-v4-layer

on:
pull_request:
paths:
- 'research/source-assisted-v4-layer/**'
- '.github/workflows/source-assisted-v4-layer.yml'
push:
paths:
- 'research/source-assisted-v4-layer/**'
- '.github/workflows/source-assisted-v4-layer.yml'

jobs:
verify:
runs-on: ubuntu-latest
timeout-minutes: 30
strategy:
matrix:
python-version: ['3.11', '3.13']
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with:
python-version: ${{ matrix.python-version }}
- name: Install the regeneration dependencies of the pinned pipeline
run: python -m pip install numpy==2.3.5 scipy==1.17.0
- name: Regenerate the word, admit this package's physical layer, run flow, lift, contract and assembly, compare with certificate.json
run: python3 -B research/source-assisted-v4-layer/verify.py
- name: Require an unchanged tree
run: git diff --exit-code
43 changes: 43 additions & 0 deletions research/source-assisted-v4-layer/NOTICE
Original file line number Diff line number Diff line change
@@ -0,0 +1,43 @@
Prepared by Chafik Boukhalfa with substantial Anthropic Claude and OpenAI
Codex assistance. Apache-2.0. Inherited files retain their original notices
and assistance disclosures. This package claims no exclusive priority over
concurrent work and reports no new kappa.

What this package contributes: the physical layer (operation frames and reuse
pairs) of PR #200's descent applied to the source-aligned complex word, the
admission of that layer with PR #200's checker on that word, and the
observation that PR #184's frame flow prices it at 7.0861e-4. Everything
else is retained and credited:

- icekylinx (OpenAI GPT-6 Astra and Codex assistance): PR144 paired cubes,
signed decoder, shared-core accounting, original compiler, carrier
matching (the frozen round-nine matching reused here), chronological
gauges, PR130 generalized Clifford frames; PR184 source-assisted
construction: source-parity local factorization, frame flow with
independent dirty lifts and source controls, exact lifts, contract,
global assembly (research/source-assisted, vendored byte for byte).
- ikeboy: PR191/193/194, the rebuild of PR184's construction on the PR168 v4
modules (research/source-assisted-v4: source_aligned_local_v4.py,
contract_v4.py, assemble.py, the parity-erasure-avoiding matching), and the
composition with PR189's/PR200's bit supplier.
- eumemic (Anthropic Claude and OpenAI Codex assistance): PR117 module
infrastructure; PR155/157 scalar modules, physical frames and reuse;
PR161 output merging; PR168 annealed modules, nested-prefix schedules,
local L1 circuit, fused outputs, bundle descent, both physical layers
(the physical layer replaced here is PR168's, used as the descent's seed),
pinned at 4a3c769e5c5430e7114c4d3e099ff34664677f17; PR131/143 frame/reuse
and late-read lineage; PR154 isolated snapshot verification.
- jamesyc / James Chang: PR124 compensated birth cuts, PR34 balanced
assembly, PR166 terminal-output accumulation. DaysSky: PR162 extended
carrier closure, PR192 frame-layout ceiling. GamingPuzzled: PR160.
gupt1156 PR148 and Abhinav Ramachandran/geckods PR158 (Hermes assistance).
jacklightChen / Zhihao Chen: PR23/29. Rohan Arun: PR100/103. an664: PR128.
Earlier lineage: Zhihao Chen PR97, Swapnil Jain, DaysSky PR150, hpst3r
PR147, SovereignSteak PR122, Aurel Prosz, dleen, Douglas Colkitt.
- Chafik Boukhalfa (Anthropic Claude and OpenAI Codex assistance): PR175
fast checker; PR179/181/189/196/200 local circuits, operation-frame descent
(discovery/), independent complete complex admission (complex/physical.py,
verify_fused.py, vendored from PR200 at a1175449) and the PR200 bit
supplier the assembly reads; PR202 (the pinned tree, 8d8d67b).
- OpenAI: the original complex phase network of "Integer multiplication
below n log n".
73 changes: 73 additions & 0 deletions research/source-assisted-v4-layer/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,73 @@
# A stronger complex supplier: PR200's reuse pairing on the source-assisted v4 word

**Claim.** The source-assisted complex word of PR #184 (icekylinx) on the PR #168 v4 modules (PR #193/#194,
ikeboy), with every input of PR #194's pipeline unchanged **except the 2,310 reuse pairs** of its physical layer
(PR #168's operation frames are kept byte for byte), certifies the complex saving

a_C = 3549537/5000000000 = 7.0990740·10⁻⁴ (PR #194: 219037/312500000 = 7.0091844·10⁻⁴, +1.28%).

Composed with PR #200's bit supplier through PR #194's unchanged 47-constraint assembly the result is
κ = 6768823/10¹⁰, the same as PR #202, because the bit branch binds (effective 6.7734·10⁻⁴). **No new κ.** The
complex cap of every composition on this lineage rises from 7.0092·10⁻⁴ to 7.0990740·10⁻⁴, and the exact-DFT
exponent saving of `research/exact-dft-transfer` (PR #225) rises with it.

## What changes

PR #194's pipeline is: regenerate the PR #168 v4 complex cache (`scripts/paired_cube_producer.py`); the
source-parity rewrite of the three query modules (`source_aligned_local_v4.py`), which also runs PR #168's physical
layer (endpoint descent and late compensated reuse pairs) and writes it to
`references/paired-cube/physical/{frames,pairs}.json` of the aligned tree; the equal-frame fresh-channel
compression (`complex_frame_flow.py`), which reads that physical layer; the exact arbitrary-dirty lifts
(`exact_complex_flow_lift.py`); the contract (`contract_v4.py`); and the assembly.

The flow's child histogram depends on the reuse pairs it receives: a pair identifies a dead ungauged donor's
register with a gauged recipient, and the flow's equal-frame compression follows those identifications. This
package replaces only the pairing. PR #168's pairing is its maximum matching of donors to recipients in a fixed
order; the pairing in `witness/physical-pairs.json` is PR #200's late compensated pairing (`discovery/`,
`physical_opt.py:pairs`) with the maximum-weight pairing of `pairs_mw.py` (transversal-matroid greedy with augmenting paths, donor weight the gain heuristic of PR #200): the same legality (donor ungauged, non-root, dead before the recipient's
first operation, donor end frame inside the recipient gauge), a maximum matching, but a different selection among
the maximum matchings. The operation frames are PR #168's own (0 of 17,789 changed); the matching arcs, modules,
local circuit, gauges, aligned rewrite and flow options are PR #194's.

Discovery record (`discovery/README.md`, float flow roots): PR #168's pairing 7.00918·10⁻⁴; PR #200's default
pairing 7.09765·10⁻⁴; the maximum-weight pairing of `pairs_mw.py` (transversal-matroid greedy with augmenting paths, donor weight the gain heuristic of PR #200) 7.0990740·10⁻⁴. Random pairings fall to 6.6·10⁻⁴, and single operation-frame moves on
top of PR #168's frames never raise the flow root (a screen over every endpoint move), so the frames are left
untouched.

## What is checked (`verify.py`, about 80 seconds, Python 3.11+ with numpy/scipy for the regeneration steps)

1. **Pins.** `baseline-pr202.tar.gz` holds the 308 files of PR #202 at `8d8d67b` that the pipeline runs (PR #194's
package and PR #184's decision/global scripts, the PR #168 v4 scripts, references and inherited witnesses, the
Clifford frames note, and PR #200's certificate, arithmetic and complex checker), each pinned by sha256 in
`SOURCE.json`; the three witness files are pinned too. Reviewers can confirm with `git show 8d8d67b:<path> |
sha256sum`.
2. **Regeneration.** The producer and the aligned rewrite run unchanged, including PR #168's physical layer and its
own physical scalar/frame checks on that layer; the frames this package installs are byte for byte that layer's.
3. **Admission of this pairing.** With the package's pairs installed, PR #200's complete complex checker
(`complex/physical.py` with `verify_fused.py`) re-executes the aligned word: every frame contains its node span
and is nondegenerate, every role chain nests (entrance, operations, root frame, full space), every alias is read
at its deadline after its donor's last operation with the donor's end frame inside the recipient gauge, the
target read chronology holds, the exact adjoint is rederived, and every formal source, target and dirty column
is replayed in both shear signs on an injective packed basis with an independently derived coefficient bound.
4. **Flow, lift, contract.** PR #184's flow runs on this layer with the package's kernel pairs; the exact lift
verifies every local map and inverse gate program over the rationals (all coefficients dyadic) and the monotone,
acyclic once-only reuse DAG with frame containment; the contract certifies the fresh columns, read chronology,
target chains and the normalized profile (13 checks), with the package's physical and kernel pairs as the frozen
witnesses that PR #194's contract compares against.
5. **Assembly.** PR #194's `assemble.py` composes the certified complex profile with PR #200's bit certificate and
reproduces `certificate.json` exactly (exact fields only; the full run record is `report.json`): complex saving
3549537/5000000000, κ 6768823/10¹⁰, the next grid point rejected by the unchanged assembly.

The complex profile keeps PR #194's m = 66, W = 12,052 roles per vertex, rank mass 794,112 and deficit 1,320; only
the child histogram changes (the flow compresses the same 13,372 physical roles to 9,412 but splits the remaining
increments differently).

## Scope and limits

- Conditional exactly as PR #194: the retained analytic, uniform-recursion, precision, weighted bit frames,
fixed-tape routing, restored-row and all-size semantic interfaces remain hypotheses; the scalar replays and exact
lifts are finite checks. PR #184's independent complex scalar transcript is not re-exported here; the one-column
identity is replayed by PR #200's checker on the pre-flow word and by the lift's inverse gate programs.
- Nothing here changes κ; the bit supplier binds. It raises the complex cap for every bit-side construction on
this lineage and the exact-DFT exponent.
- The pairing is a discovery heuristic; no optimality among maximum matchings is claimed.
30 changes: 30 additions & 0 deletions research/source-assisted-v4-layer/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
# source-assisted-v4-layer

**Complex saving 7.0990740·10⁻⁴ (no new κ).** PR #194's source-assisted complex word with PR #200's reuse pairing in
place of PR #168's (operation frames unchanged) certifies

a_C = 3549537/5000000000 = 7.0990740·10⁻⁴ (PR #194: 7.0091844·10⁻⁴, +1.28%),

through PR #184's flow, exact lift and contract and PR #194's assembly, which with PR #200's bit supplier gives
κ = 6768823/10¹⁰ (the bit branch binds, as in PR #202). The complex cap of this lineage and the exact-DFT exponent
saving (`research/exact-dft-transfer`) rise to 7.0990740·10⁻⁴. [PROOF.md](PROOF.md) states what changes and what is
checked.

## Verify

python -m pip install numpy==2.3.5 scipy==1.17.0
python3 -B research/source-assisted-v4-layer/verify.py # about 80 seconds; -O is refused

The script rebuilds the 308 pinned files of PR #202 (`8d8d67b`) from `baseline-pr202.tar.gz` after checking every
sha256 in `SOURCE.json`, regenerates the PR #168 v4 cache and the aligned word, installs the physical layer of
`witness/` (PR #168's frames, this package's pairs), admits it with PR #200's complete complex checker, runs the
frame flow with the package's kernel pairs, the exact lift, the contract (with the package's physical and kernel
pairs as the frozen witnesses) and the assembly, and compares with `certificate.json`. `--write` regenerates the
certificate; `--temp-root` chooses scratch storage.

## Files

`verify.py`, `certificate.json` (exact fields), `report.json` (full run record), `SOURCE.json`,
`baseline-pr202.tar.gz`, `witness/physical-frames.json`, `witness/physical-pairs.json`, `witness/kernel-pairs.json`,
`discovery/` (the pairing and the sweep that found it; not run by the verifier), `PROOF.md`, `NOTICE`.
`.github/workflows/source-assisted-v4-layer.yml` runs the verifier on Python 3.11 and 3.13.
Loading
Loading