Skip to content

odinfree/storm-ecdsa-harness / skills/beachhead-ratchet

beachhead-ratchet

Verified

Take a left Pareto move even when it raises Toffoli, under a T budget written down first, then win the cost back on your own tree, attributing every lever to the component the width cut made dearer. Learned from bulengerk's ecdsa.fail record (9 left moves, 61 down rows). Use when a width cut works but costs T, when you hold the leftmost row of a band, or when a rival posts rows built on your tree.

FileKindSize
EVIDENCE.mdmarkdown3.3 KB
SKILL.mdskill md5.7 KB

EVIDENCE.md

View on GitHub
# Evidence: beachhead-ratchet

Source data: an ecdsa.fail API snapshot of 25 Oct 2026 (81 bulengerk submissions, 74 validated; 9 left moves and 61
down moves at submission time, computed against the frontier as it stood). Every claim rests on his public submission
notes: "notes.md:N" is line N of his 81 notes concatenated in submission order (fetch them with `ecdsafail
submissions`), and submission ids identify each row. The `report-*` names refer to our working reports, which are not
published. Most rows also give the public note line or submission id they rest on; the rest summarize his notes
as a whole. [V] = checked in his source code;
otherwise his own statement.

| claim | evidence |
|---|---|
| Q770 beachhead: 774 -> 772 -> 771 -> 770 at +20.4%, +10.8%, +22.2%; 186.6M -> 304,134,670 (+63%); stated target 300,000,000 T with an explicitly permitted overrun of 1.37822% | report-descent770 "Entry"; notes.md:9080, 9234, 9405, 9410-9414; rows c9089985, b2b994fd, 4f55785a |
| 13 down rows to 115,382,810 (-62%) in ~31 h; 62% of the cut recovered the purchase overhead; half the cut in the first three rows (11.6 h) | report-descent770 "Why so fast" 1; submissions.json |
| Q792: landed +31% (959.7M -> 1,257.8M); 38 rows in all (37 reductions after the landing row) to 574,285,626 (-54.3%) | report-leftmoves table; report-descent792 scope and row table |
| Q794 / Q793 landed at +38% / +44%, then 5c42a5be -17.5% at +2.7 h | report-leftmoves table and pattern 6 |
| misses consistent with no frontier re-read: 13ba6c73 (Q774) and 38e6918e (Q695) landed after a better rival row | submission timestamps on the board |
| invoice model within ~100 T: 196,250,688 vs 196,250,710 measured | report-descent770 pattern 3; notes.md:10019 |
| gates: ">= 5% below that parent's executed cost" (d181b19f notes.md:10888); "10 x new_sum <= 9 x baseline_sum" (9a8bbcae notes.md:11034); the user's "strict four-percent threshold" on the final Q792 row (notes.md:8847-8850; the penultimate reduction was 3.51%); ">= 5%" (notes.md:9739) vs strictly "> 5%" (notes.md:10021) on Q770 rows | report-ops "Cadence"; report-descent792 cadence D |
| Q770 carry-row ladder (5-9% per multi-change row, not per isolated kernel): row_add.rs:55, frogdrop.rs:1456, row_add.rs:223, signed_batch.rs:4 | report-descent770 pattern 5 [V] |
| 13-line diff for -9.63% (d8e2a8d4) because overlays keep call signatures | report-descent770 rows; notes.md:9941 |
| pins in source: CANDIDATE_TOFFOLI 670139450 -> 656602578 in q792_lifecycle_r01.rs | report-descent792 "Ordering" [V] |
| submit-while-validating on a frozen parent; "score did not improve" as the correctness proxy (22 notes) | report-ops "Cadence and pipeline", "Bases and credit" |
| re-absorb: nasqret's 39227bab = his row-10 tree + 8 edited files; row 11 re-imported 5 byte-identical, beat nasqret by 4.1% within 5 h | report-descent792 "Keeping the leftmost point" [V] |
| left Q792 for gnuchev's frogdrop tree the same day the 776-774 band opened (9 Oct) | report-descent792; notes.md:8970-8974 |

Limits: lambda of his rows is not exposed by the board; the percent gates may be the user's or the model's choice; the
per-lever split inside multi-change rows is his own invoice.

- Left moves that raised T: seven of nine (Q665 and Q660 lowered T: report-leftmoves rows Q665, Q660). Observed increases about +11% to +122%.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 323kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall