The build-remote stamps read earlier were UTC and were labelled BST. Internal adversarial pass, not an independent review.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Internal adversarial pass, not an independent review. Q1 (this lane, the
algebraic structure of M_r): the fold probe and the integral cube-sum degree
test on the frozen-identical binary show no composition cheaper than 8x. No
affine fold, no word separability, no key-order commutation; algebraic degree
>= 16 after one application and saturated after two. The 9,360 ops per item
stand. Q2 margin (full diffusion at 2 applications) and Q3 census (best day a
1.17x FPGA multiply datapath, 1 in 2^24, zero wall-time gain, ROT-all-equal
never seen) kept as courtesy runs attributed to adv-mixer-3 and adv-mixer-2.
igneum-pow merged to build/master and re-proven byte-identical to c3d32437;
pre-merge runs re-run on the merged tree and matched. Logs copied into the
branch so the report paths survive; run-box.sh writes outside the mirror.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Internal adversarial pass, not an independent review. Adds the cube-sum
(higher-order differential) command to bound the algebraic degree of the
applications: a low degree would admit a cheap polynomial batching of the 8
applications, so degree saturation at small d is the no-shortcut bound. Plan
section 8 records main's three-lane re-scope: this lane is the algebraic
structure of M_r.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Internal adversarial pass, not an independent review. Q3 census over 2^24 day
keys: largest per-day M1 FPGA-datapath gain 1.1726x on one day, 3.26e-4 of days
over 1.1x, zero days with any DSP or wall-time gain, ROT-all-equal never seen.
Q1 fold probe: 1e6/1e6 affinity violations, 0 dead word pairs, 0 key-order
agreements, so no cheap composition of the 8 keyed applications. Q2 diffusion
sweep and the f4 analytic tail are running; numbers land as they finish.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Internal adversarial pass, not an independent review. The plan restates M_r
from the frozen crate and spec 1.8, ranks Q3/Q2/Q1/Q4, gives the method, the
planted known-fail shape and the box-hour estimate per step, and lists every
file opened. Records the byte-identity result: build/master differs from the
frozen commit in accept.rs, emit.rs, generator.rs, packcheck.rs and two test
files, but is byte-identical over memhard.rs, seed.rs, bind.rs, derive.rs and
the Cargo pin, which define M_r.
Harnesses: adv-mixer (new: diffusion margin for Q2, the fold probe for Q1,
with a planted-weak-day hook), f4-weakday (copied read-only from
build/attack-pass for the Q3 census), f8-uniform (copied from the regate
worktree).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Home row 03, the litepaper's chip section at /litepaper#chip-model (paragraph and table; the class v3 row last, "never the launch
state"), the miner page's line and evidence.md row 17 in its eight columns: 2.1x to 3.9x under class v4 from the first block,
class v5 removing the stored-dataset chip as a category, the 5x to 9x only as the class v3 baseline, the devnet's activation
height as a devnet fact. The litepaper's abstract carried the same claim in the old form ("5x to 9x today; class v4, now on
the vote") and now reads launch-first too. tools/ci/ledger-text-check.mjs follows the sentences (X35 on the home page and the
litepaper). No disclosure prize anywhere.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Suite at c3d32437 on box 2: 103 passed, 0 failed (lib 140.8 s: every class v4 draw in the tests pays the 2^20 ratio pass). CI 37639406567 success. Census: 4,099 programs, 0 lossy sites of 65,584, 0 exhaustions, mean attempt 2.086 (af1b7c46 1.998), max 17; the ratio refuses about 4 percent of candidates that pass every other test. tools/ca3-v4-uniform draws the seeds across the available cores (the serial run became a four-hour job under the ratio pass); the analysis and the row order are unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The first 16 GB card and the first Thunderbolt card. The app's first poll listed it (vendor nvidia, 16,311 MiB,
driver 610.47 shared with the 5090, kind discrete: the app does not know it is an eGPU) and started a miner on
it by itself. Job run-5060ti-bench-20261007-b (tools/bench-5060ti/pc2-5060ti-bench.ps1, the card alone under
the runner's --cards-off, the 5090 mining, no power limit written: 180 W default read back unchanged, PC 2 lost
power twice that day): G1 PASS with 867dbc45cfb36b4d (sub-version 1) and 90f794dd556f7a3b equal to the Mac's;
the window 1,105 dispatches of 2^24 in 602 s at 30.882 MH/s, 114.8 W mean at utilisation 90 percent and over,
0.269 MH/W, PCIe 4.0 x4 through the enclosure (the hash is latency-bound, the link costs nothing measurable).
Owed: the efficient point (the app's Ember Tune; PC 2's Power Helper refuses every request since the restart)
and the prove-beside read (the shipped WSL2 host has no IGNEUM_CUDA_DEVICE selector; the floor server is on
PC 2; the prover-floor host is the cut). The kit (tools/bench-5060ti/make-kit.sh) takes the sub-version 1 packs
from ca3-v4-amend 0a62293a, since master's packs-ca3-v4 is the 6 October stream.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
