Skip to content
Merged
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
175 changes: 175 additions & 0 deletions research/dirty-fanout-batch/CERTIFICATE.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,175 @@
{
"base": "f567884aeb8851b6e794af1184119a3bfbe1e631",
"binary_optimum_domain": "all ordered binary splits, unweighted n=1..64",
"controls": [
"omit_final_leaf",
"reverse_factor_order",
"change_common_source_during_batch",
"omit_common_restoration",
"negate_without_reverse",
"omit_one_output",
"empty_common",
"empty_output_list",
"insufficient_scratch",
"ancestor_alias",
"target_source_alias",
"coefficient_alias",
"boolean_sign",
"nonlinear_left_map_has_dirty_cross_term",
"rounded_left_map_is_not_exact_linear_map"
],
"dense_cases": 795,
"rational_cases": 1,
"rounded_cases": 27,
"scope": "Exact linear multi-output interface; no instantiated supplier, rounded inverse theorem, or new kappa",
"seed": 58,
"seed_case": {
"common_factors": 8,
"dirty_arrays": 4,
"invoices": {
"batch": {
"additions": 2952,
"calls_by_factor": {
"0": 16,
"1": 16,
"10": 4,
"11": 4,
"12": 8,
"13": 8,
"14": 8,
"15": 8,
"16": 8,
"17": 8,
"18": 8,
"19": 8,
"2": 16,
"20": 8,
"21": 16,
"22": 16,
"3": 16,
"4": 16,
"5": 16,
"6": 16,
"7": 16,
"8": 2,
"9": 4
},
"coefficient_reads": 2214,
"coefficient_storage_words": 207,
"data_storage_words": 30,
"head_travel_word_slots": 458208,
"instruction_records": 246,
"leaf_calls": 246,
"multiplications": 2214,
"negations": 369,
"preflight_array_length_checks": 10,
"preflight_coefficient_alias_pairs": 7380,
"preflight_matrix_row_length_checks": 738,
"preflight_role_alias_entries": 10,
"source_reads": 2214,
"target_read_writes": 738
},
"grouped": {
"additions": 9096,
"calls_by_factor": {
"0": 80,
"1": 80,
"10": 4,
"11": 4,
"12": 8,
"13": 8,
"14": 8,
"15": 8,
"16": 8,
"17": 8,
"18": 8,
"19": 8,
"2": 80,
"20": 8,
"21": 16,
"22": 16,
"3": 80,
"4": 80,
"5": 80,
"6": 80,
"7": 80,
"8": 2,
"9": 4
},
"coefficient_reads": 6822,
"coefficient_storage_words": 207,
"data_storage_words": 30,
"head_travel_word_slots": 885344,
"instruction_records": 758,
"leaf_calls": 758,
"multiplications": 6822,
"negations": 1137,
"preflight_array_length_checks": 10,
"preflight_coefficient_alias_pairs": 22740,
"preflight_matrix_row_length_checks": 2274,
"preflight_role_alias_entries": 10,
"source_reads": 6822,
"target_read_writes": 2274
},
"separate": {
"additions": 8160,
"calls_by_factor": {
"0": 40,
"1": 56,
"10": 16,
"11": 8,
"12": 16,
"13": 16,
"14": 16,
"15": 8,
"16": 16,
"17": 16,
"18": 16,
"19": 16,
"2": 56,
"20": 16,
"21": 16,
"22": 16,
"3": 56,
"4": 72,
"5": 56,
"6": 48,
"7": 72,
"8": 16,
"9": 16
},
"coefficient_reads": 6120,
"coefficient_storage_words": 207,
"data_storage_words": 30,
"head_travel_word_slots": 998624,
"instruction_records": 680,
"leaf_calls": 680,
"multiplications": 6120,
"negations": 1020,
"preflight_array_length_checks": 10,
"preflight_coefficient_alias_pairs": 20400,
"preflight_matrix_row_length_checks": 2040,
"preflight_role_alias_entries": 10,
"source_reads": 6120,
"target_read_writes": 2040
}
},
"left_factors": [
1,
2,
3,
4,
5
],
"word_sha256": {
"batch": "a8a8a350ea43fec5769f393de1eb077c51216a5aa2cbec7046511f58fd8fa970",
"grouped": "de2b43e1f8d00c93a1a6a8c90378713fdc8d59d42339f40a7a59a23b4cba7818",
"separate": "50f414e6ea161dc6ab1f54c90bf100ebe3ec1433d9b8fb84c8e9b2832f636828"
}
},
"short_star_words_rejected": 9331,
"status": "PASS_DIRTY_FANOUT_BATCH",
"symbolic_cases": 128,
"tape_cases": 15,
"weighted_cases": 384
}
29 changes: 29 additions & 0 deletions research/dirty-fanout-batch/CLAIM.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# Sprint claim: dirty common-factor fanout

- Track: recursive interfaces; OS-random seed **58** (Python secrets.randbelow(100)).
- Exact base: `f567884aeb8851b6e794af1184119a3bfbe1e631`.
- Object: simultaneous independent output shears `y_i += A_i B x` with a shared, unchanged input and entirely dirty workspace.
- Method: one batched dirty commutator, exact noncommutative replay, a restricted-query lower bound, and complete weighted recursive cost/storage accounting.
- Seed choices: fanout `2 + 58 % 5 = 5`, common factor length `1 + 58 % 17 = 8`; left lengths 1 through 5.
- Deliverable: scoped interface construction and standalone exact verifier/certificate; no new kappa.
- Finish target: 2026-10-13 08:29 UTC (19:29 Australia/Melbourne). Final 10 minutes reserved for verification and handoff.

Three candidates were considered: cancellation of repeated transparent calls; deferred sibling cleanup; shared common-right-factor output batching. The first overlaps PR840's native dyadic pair; the second overlaps PR822's dirty connector reuse and PR714/788 clean checkpoints. The third changes the multi-output contract and retains conservative private scratch.

At claim time the public issues page listed only #650, with no pinned sprint coordination issue. GitHub API read access returned `Forbidden`; no claim was posted. Public open-PR metadata was checked through #856, plus searches for boundary, echo/checkpoint/batch/prefix, common right factor, fanout and multiple outputs. Relevant context: #628, #663, #673, #689, #714, #730, #788, #822, #840. Existing echo work changes decomposition/scratch for one product output; this work batches distinct outputs sharing B. No claim of general algebraic novelty or first discovery is made.

Selected state read: #799/#589, kappa `1661634653669567/10^18`, its actual X4 proof note and maintainer review. This package does not change that construction.

## Result update

Completed one scoped construction: common-factor multi-output batching with
246 versus 680 complete binary factor-shear calls and four dirty arrays.
The proof, full weighted/storage/routing/precision bill, exact certificate,
independent checks and controls are in [README.md](README.md) and
[VALIDATION.md](VALIDATION.md). There is no new κ or admitted supplier.

The overlap refresh reached #859 and explicitly distinguished #793's scalar
coefficient batching. No coordination issue was found. API access is denied,
so the completed contribution is handed off as one local commit and clean patch;
no external claim or PR was posted. Existing source and the selected frontier
remain unchanged.
31 changes: 31 additions & 0 deletions research/dirty-fanout-batch/NOTICE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
# Attribution and scope

Prepared for the requesting contributor with substantial OpenAI Codex assistance
in the mathematical argument, implementation, tests and prose. The user's human
identity is not asserted. This package uses the repository's Apache-2.0 license.

Chafik Boukhalfa / chafreaky's PR628 supplies the dirty product commutator,
literal-inverse convention, exact restoration contract and rounded-error
analysis. Its reference source, proof and notice are retained byte for byte.
Ali Romman's PR621 and OpenAI's exact-DFT/Newton framework are inherited context;
we do not use their analytic kernel claims to certify the abstract matrices here.
The commutator and batching algebra are classical; no literature-priority claim
is made.

Relevant concurrent work inspected: androolloyd's PR673 (multiway dirty echo),
commdata2338's PR677 (native DATA-query contract), Dillon Amburgey's PR689
(ternary placement), FanPUYen's PR714 (clean root checkpoint), Sagar Thakrar's
PR730 (fixed scratch optimization), nightangelflowerwin-ops's PR788 (reusable
clean checkpoint), PR822 (dirty sibling connector reuse), and Wentao Li's PR840
(dyadic ADDRESS-query pair). These concern different contracts or one-output
schedules. Their code is not consumed; no relative optimality against their
broader classes is claimed. Varun Mittal's PR793 batches scalar coefficient
preparation in historical JP8 compensation and preserves its recursive profile;
this package instead batches full product-query calls under a different abstract
multi-output contract. No PR793 code or claimed savings are consumed. PR663's nonlinear-restoration distinction is
acknowledged and explicitly respected by the rounded controls.

The accepted frontier's authors and source lineage retain all attribution.
This local multi-output result supplies neither a new κ nor an instantiated
current supplier, physical tape compiler, full theorem formalization, human
peer review or measured performance result.
Loading