Skip to content

Research: atomic triple-sum connector does not beat the universal budget (no new kappa) - #1518

Merged
Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
hevlyo:research/atomic-trio-connector-obstruction
Oct 16, 2026
Merged

Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
hevlyo:research/atomic-trio-connector-obstruction

Conversation

@hevlyo

@hevlyo hevlyo commented Oct 14, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Adds one charged alphabet symbol to Ross Richardson's universal dirty connector: an atomic TRIO_{j,k,l} that adds three read-only sources into z in one frame.
  • On the universal 7-mask and 31-mask batches, every positive source-map price leaves this symbol at least two cheapest applications above the known 6-toggle and 12-toggle repairs. No frame saving survives the counting bound and the ten-call census.
  • No selected-frontier, supplier, or kappa change.

Scope

Task connector-alphabet-change, priority revision 2, seed 7073f19eda1a91682f62a8206ecc4836, scope key ff4f5113ed2256ab19ed. Base ef5f39403610d231a0c27e8a6b4f08c10209bcd6. Scope status: refined.

The new symbol pays λ_j + λ_k + λ_l + f with every λ_j > 0 and frame price f ≥ 0. It does not expose partial sums. Weight-2 DUO (#1403), two-connector transfers (#1412, #1486, PR #1438), mutable sources (PR #1340 / #1370) and the same-alphabet six-source extension (PR #1185) are different objects and are not rerun.

Coordination: discussion #1504. Overlap refresh at submission covered 759 open PRs and 193 discussions, complete pagination. The matches are the predecessor, this claim, and those neighboring alphabets.

Evidence

Python 3.12.10 registered replay:

PASS_ATOMIC_TRIO_CONNECTOR three_source_orders=24 misses={3: 8, 5: 8, 6: 8} five_source_words=6048000 full_covers=0 best_coverage=30
PASS_REGISTERED_RESEARCH_PACKAGE research/atomic-trio-connector outcome=negative-result

python3 -B -m unittest discover -s tests -p 'test_research_*.py' passed (20 tests, 1 skipped). The predecessor control PASS_UNIVERSAL_CONNECTOR_BOUNDARY pairings=945 max_coverage=31 queries=74 leaves=182 was rerun. A full selected-frontier replay was not run.

Test plan

  • python3 -B scripts/research_package.py --package research/atomic-trio-connector --check-only
  • Python 3.12 scripts/research_package.py --package research/atomic-trio-connector
  • Research onboarding unit tests
  • Maintainer review of the price model and the census domain

Limits

Finite exclusion of this one symbol on two batch sizes. Not a proof-assistant theorem, not a bound for every other connector alphabet, and not a new multiplication exponent.

Prepared by Riquelme Soares (@hevlyo) with substantial Cursor agent assistance. Predecessor credit remains with Ross Richardson (universal connector boundary / PR934). Apache-2.0.

Optional workflow note: task connector-alphabet-change / revision 2, negative result. The largest detour was a full clone of this repository, which is too large for a one-hour session; a sparse checkout was enough. One concrete improvement would be a blob-filtered contributor bootstrap in the package guide.

Made with Cursor

…budget.

The charged weight-3 symbol stays at least two source applications above the known three- and five-source repairs. No new kappa.

Co-authored-by: Cursor <cursoragent@cursor.com>
@Mister-Meeseeks

Copy link
Copy Markdown
Contributor

Batch review at 0c6a58aaf22e0fc1b34497f8f6f80db6212b4621.

This remains open after scoped triage. Next review step: The complete census excludes only ten-call schedules with one TRIO and twelve source applications. The README claims no improvement for arbitrary positive application/frame prices and any number of TRIOs, but ten-call schedules with two or more TRIOs have not been excluded by that census; extra applications could trade against fewer frame charges. Narrow the tariff claim to the proved application-budget window, or supply a bound covering higher-application ten-call words before whole-head admission.

This is a specific continuation point, not a rejection or a request to redo already recorded checks. Completed local evidence is retained separately from outstanding whole-head obligations.

Batch review, attribution and evidence. Thank you for the contribution.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants