Completed the sourced adversarial research report. It covers seven falsifiable directions across all six tracks, ranks the top three, and gives a first 24-hour plan.

I independently confirmed that the public repository HEAD matches the requested commit and reproduced the selected secp256k1 control’s stale hashes and retired cost divisor. I checked the report’s structure locally. No production attack or official score was reproduced.

Created artifacts/prior-art.md with six falsifiable directions, all six tracks, primary-source citations, cost equations, experiments, rankings, and a 24-hour plan.

Disclosed live-page status inconsistencies and source-access limits. Structural and mathematical sanity checks passed. No submission code executed or Yukon submission made.

Created artifacts/mechanisms.md with six distinct directions, all six tracks’ rules, primary sources, charged cost equations, falsification experiments, top-three ranking, and a 24-hour plan.

Structural and small algebra checks passed. Hosted-page verification limits are disclosed. No submission code or production attack was run. The file remains untracked.

I've written artifacts/curve-directions.md. It merges the three input reports, settles where they disagree, rejects the weak proposals, and ranks seven directions. Only my own integer calculations were run; none of the proposed experiments were.

What I checked myself:

• Rules unchanged: the GitHub commit list still shows the pinned commit 6423c27 as the newest on master, so nothing has changed since the pin.
• Live Yukon page: secp256k1 is "open", with a "current record" of 127.08 and 14 rows down to 146, all marked "Awaiting review". The "paused" status in one input report did not appear at my fetch.
• Integer facts: for BLS12-381, 2^32 is the largest power of two dividing r−1, gcd(E,r)=1, log2 E ≈ 4313.55, and p has order exactly 12 mod r. I also computed the step counts each method needs to reach 0.39 success.

What I could not check: this machine's proxy blocked git and curl, so I did not re-hash the old BSGS control's protected files. Its stale-digest finding rests on two input reports that agree with each other.

Disagreements between the input reports, settled:

• 127.08 is the real benchmark. It sits close to what a known secp256k1 rho walk costs (about 2^126.7 steps at 0.39 success), so new secp256k1 ideas must beat roughly 2^127, not the 2^129 or 2^146 rows.
• P-256 has no hidden speed-up. Its curve constants rule out any symmetry beyond negation (a standard textbook fact), which is stronger than the input reports said.
• Old BSGS re-audit demoted. It is folded into a 30-minute sub-check; two reports already reproduce the finding, and BSGS cannot compete with 127.

Top 3, ranked by evidence, information gain and how cheaply each can be killed:

1. Rho cost floor versus 127.08. A displayed 127.08 allows about 1,300 instructions per walk step in total. Counting instructions in real secp256k1 kernels either shows that budget is impossible for this method, or confirms 127.08 as the bar to beat. This is the first experiment, about 6 hours.
2. Cheon cost ledger for the powers track at n=2^20. This is a published algorithm, not a new one. Because memory is not scored, I predict a score of about 120 ± 1, roughly 6 bits below the generic baseline. That track has no result posted yet.
3. Rho in the pairing target group, using its order-12 symmetry. This would give both pairing-inversion tracks their first concrete baseline, roughly 2^125.6 steps. It is a known technique; none of the three inputs proposed it.

The report also covers a pairing-inversion claim audit (the rigorous negative-result route), a full-cost finite-field (NFS) comparison, walk tuning as calibration, and a low-prior algebraic (index-calculus) route. It ends with a 24-hour plan and explicit rejection criteria.

The live record's supporting dossier was not inspected, and the requester's measurements remain unreproduced. The file is left untracked for upload.