attack-pass lane (d): F9 and F1 toward 10^6 on the class v5 freeze recorded as partials (stopped on the Igneum 2.0 drain order; 306,980 seeds 0 exhausted; 217,310 programs, two letter misses at 5.0781 percent, parity owed)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/analysis/attack-pass-2026-10.md

@@ -293,6 +293,10 @@ attack-pass) by the coordinator's exception rule; nothing touches GitHub.
 | F9 exhaustion count, the frozen tip (A GATE LINE for the 0.3.24 move) | 10^5 chain-shaped seeds on the v5 chain path (era-composed class, `--chain`) with the dn3 state at igneum-pow class-v5 1c420786, pairing e5a4ac5978462156, build-1, ten chunks of 10,000 under `lease pool 4` (chunks 1, 3, 5 to 9 under class v5; chunks 0, 2, 4 re-leased under class release on the coordinator's order; 14,477 to 14,480 s per chunk, about 1.45 s per seed); binary copied into the run dir; interim line sent at 00:55 UTC (seeds drawn, 0 exhausted, 0 panics, max 30, F1 0 failures), which cleared the move; last chunk written 02:34:54 UTC | 100,000 of 100,000 seeds drawn, 0 exhausted, 0 panics, 0 past attempt index 31, max attempt index 30; histogram by attempt index (0 = accepted on the first draw) 0: 31,454; 1: 21,460; 2: 14,660; 3: 10,263; 4: 7,047; 5: 4,701; 6: 3,297; 7: 2,256; 8: 1,532; 9: 1,027; 10: 702; 11: 509; 12: 365; 13: 216; 14: 153; 15: 103; 16: 80; 17: 56; 18: 39; 19: 20; 20: 24; 21: 9; 22: 6; 23: 9; 24: 3; 25: 4; 26: 2; 27: 1; 29: 1; 30: 1; first-draw acceptance 0.3145, mean attempt index 2.185 (3.185 draws per seed), 4,862 seeds (4.86 percent) at index 8 or above, 255 (0.255 percent) at 16 or above; the 256-attempt cap and the deterministic last resort never reached. Meaning per tier: no epoch seed in 10^5 fails to draw a program, so the liveness halt of AP-F8-2 has no observed case on the frozen tip at this count (the bound it supports is under 3e-5 per seed at 95 percent, about one epoch in 33,000 at worst; a halt a node operator would see as a stuck epoch, a miner as a dead epoch, a holder as a paused chain), and the draw cost stays at about 3.2 candidates per epoch for every node | PASS (0 of 10^5; the record `f9-grind.md`, section (d)) |
 | F1 shadow redundancy, the frozen tip (A GATE LINE for the 0.3.24 move) | 10^5 class v5 programs through the string-seed path (`generate_from_seed_bytes_program_class`, class V5, every candidate draw through the (c''') floor over 2^20) at igneum-pow class-v5 1c420786, pairing e5a4ac5978462156, build-1, `lease pool 16` class release (cores 8 to 23, re-leased 22:34:05 UTC), binary sha256 bb70bbf69a4b3223... copied into `frozen-1c420786-f1/bin`; 20,774 s of census (5 h 46 min; about 5 core-s per program, the (c''') draw cost), census.csv (sha256 4e34b669f2f5c680..., 100,000 rows) written 04:20 UTC on 8 October 2026 | 100,000 of 100,000 programs; instructions saved min 0.000, mean 0.623, max 4.688 percent (worst `attack-f1/95060` at attempt 0, 6,912 to 6,588 per iteration; `81748`, `66933`, `3006` at the same 4.688; next 4.311); chip-view ops saved mean 0.520, max 4.783; programs over 5 percent 0, over 10 percent 0; soundness: differential mismatches 0 of 100,000 (8 random states each), verifier mismatches 0 of 100,000; 0 panics; histogram of saved, 0.5 percent bins from 0: 55,241; 20,597; 11,762; 9,851; 1,484; 656; 259; 133; 13; 4; 0; 0; draw attempts per program: index 0 31,630, max 30 (the same shape as F9's). Against the v4 10^5 (max 5.078, the AP-F1-1 letter miss): the v5 tip's worst sits 0.39 points under the letter, the two top bins are empty, and the mean is unchanged (0.617 to 0.623), so the v5 leaves add no redundancy and remove the one letter miss. Meaning per tier: the shadow block of every drawn program stays within 5 percent of its naive count under the harness's rules and the honest compiler finds the same shortcuts, so no chip gets a shadow-side discount (a miner on a card pays the full block, a hypothetical ASIC gains nothing here) and the chain's verifier agrees with the harness on every program (0 mismatches), so no node disagrees with another on any drawn block. Harness gap and its fix: section 12 of the record (the running census was the old binary; the flush is in 18a9c04a for every census after it) | PASS (0 of 10^5 over the letter; AP-F1-1 FIXED-AND-PASSED on v5 at this count; the record `f1-shadow.md`, section 13) |
 | F4 weak-day census, the post-freeze commit | 2^24 chain days at class-v5 8ca66afa (AP-F4-1 in the agreed form: cost A at most 205, k >= 1 or all-ROT-equal rejected, the forty redrawn), the harness on the agreed w32 convention (digits 0 to 31) and median 226; known-failed day 29,337 redrawn under the rule (cost 203 to 228), day 20,729 at 219 unchanged; build-1 `lease pool 12` class adv, 379 s, ended 22:3x UTC | 0 of 2^24 days over 1.1x on M1 (median 226; minimum cost 206 at day 27,016, 1.097x, one adder above the reject line) and 0 on M2; mean 225.79, sd 6.07 (pre-rule: 5.69e-4 over, min 203). The redraw rule removes the LUT tail by construction and the measurement agrees | PASS; AP-F4-1 FIXED-AND-PASSED against class v5 at 8ca66afa |
+| F4 day-key scan to 2^28, the strengthening line (8 October 2026, the founder's order to load the builders) | 2^28 = 268,435,456 chain days from day index 20,729 at class-v5 8ca66afa (AP-F4-1's redraw rule; the day-key draw is `MixParams::with_shape(seed_words_from_bytes(day_bytes(d)), Shape::for_class(..))`, the same mixer shape on v4 and v5, the state leaves never enter the key), the gated binary fb872a5d carried from build-1 to build-4, `lease pool 8` class measure, 379.2 s on 8 threads (11 us per day), `scale-8ca66afa-f4/census-2p28.md` written 10:15 UTC | Days with any gain over 1.1x: 0 of 2^28 on M1 (cost under 206 against the median 226), 0 against the mean 225.79 (sd 6.07), 0 on M2 (k at least 2), 0 on ROT and RC (exact 0 ops moved); M1 cost mean 225.791, sd 6.073, median 226, min 206 (day 27,016, 1.0971x, the floor the redraw rule sets: no day under 206 exists in 2^28), max 258 (day 221,075,548). The weak classes sit on their analytic expectation: ROT distinct at most 3: 8,228 against 8,237; ROT any pair sums to 32: 160,354,008 against 161,256,979; MUL popcount at most 4: 8,838 against 9,984; MUL under 256: 130 against 256; MUL two equal: 18 against 15; RC any zero: 1 against 1.0 (day 109,542,046 at cost 234, gain 0.966x, no gain); RC + rk = 0: 87 against 72 (1.8 sigma); RC two equal: 8 against 7.5; the ROT-distinct histogram matches S(8,d) 31_d / 31^8 to three figures at every d. Sixteen times the gate's count with the same floor and the same expectation lines. Meaning per tier: no chain day in the first 735,000 years of days gives a per-day datapath a 1.1x discount (the FPGA or ASIC that would bake a day's constants gains at most 1.097x on its best day, under the 1.1x gate, so no chip buys a weak day), and the day draw is uniform at 2^28 on every class the harness knows | PASS (0 of 2^28 over 1.1x; record `f4-weakday.md`, section 10) |
+| F8 hot-set gate to 256 seeds, the strengthening line (8 October 2026) | seeds p66 to p257 (192 new; with the gate's p2 to p65 that is 256) at 2^24 nonces each, chain path, class v5 on the freeze 1c420786 (the gated binary 0f5c98dc, pairing e5a4ac5978462156, the dn3 state e000824c; no kit-pairing flip reached this lane), window-model control, by-site, box 2 as three thirds under `lease pool 32` class measure, 10:07 to 11:37 UTC (4,291, 4,287 and 5,315 s), `scale-1c420786-f8/`; validation: 3 library comparison samples at 0 mismatches, `Epoch::hash_warp` agreement on 37,440 warps at 0 mismatches | 184 of 192 within 1.2x of the window model at the top 0.1 percent (mean ratio 1.023; 45 under 1.00, 114 at 1.00 to 1.02, 10 at 1.02 to 1.05, 7 at 1.05 to 1.10, 8 at 1.10 to 1.20); 8 over: p110 1.3527x, p66 1.3403x, p234 1.3156x, p145 1.2750x, p77 1.2664x, p225 1.2457x, p248 1.2188x, p89 1.2065x; every over seed's hottest item at 263 to 432 reads of 2^31 (0.0000 percent) and its hot-set verdict on the windowed control clear (largest excess X_f/f +0.60 at p110); the flat control (the auditor's first view) calls p76 and p88 a hot set at X_f/f +1.07, which the window model explains (both within 1.2x on the windowed control); the 6-sigma largest-bucket line flags 57 of 192 (p110 at +61.67 sigma on buckets of 64), the AP-F8-1 tail mechanism at its rate; predicted sources: 190 none, 2 one-one-bit through a load (p212 at 1.1915x, attempt 4; p225 at 1.2457x, attempt 0), routed to the hash lane for by-site attribution on the frozen tip. Over all 256 seeds: 245 within 1.2x (95.7 percent), 11 over, the rate at 256 (4.3 percent) the gate's rate at 64 (4.7 percent), max 1.3527x against the gate's 1.5047x (p10). Meaning per tier: the top 0.1 percent of items carries at most 0.23 percent of reads on any seed (the window model's own 0.17), so no card or chip gains a cacheable hot set on any of 256 epochs; the excess class is the named quarter-bit bucket at one narrow-window site, at the same rate at four times the count | PASS at the gate's reading (the over-1.2x rate stable at 256; no hot set on any seed; p212 and p225 to the hash lane); record `f8-uniform.md`, section (e) |
+| F9 exhaustion count toward 10^6, PARTIAL (stopped on the founder's drain order, 16:38 BST 8 October 2026: Igneum 2.0 moves the baseline to the class v6 generator, docs/plans/igneum-2.0.md, so the class v5 freeze runs serve no pin) | seeds 100,000 to 999,999 on the v5 chain path with the dn3 state at 1c420786 (pairing e5a4ac5978462156), twelve chunks across build-4 (six lower ranges of 85,000 at 8 threads, the gated binary cadce754) and box 2 (six upper ranges of 65,000 at 6 threads), every lease class measure, 11:08 to 16:38 BST, rows keyed by seed in `scale-1c420786-f9/par*.tsv` on both boxes | 206,980 distinct seeds drawn (build-4 142,138, box 2 64,842), 0 exhausted, 0 panics, max attempt index 37 (seed 421302, build-4 chunk 2; seed 718097 at 32 is the only other past 31); tail over the partial: 20: 48, 21: 18, 22: 16, 23: 13, 24: 8, 25: 5, 26: 4, 27: 3, 28: 1, 29: 2, 30: 1, 32: 1, 37: 1; the geometric tail at its rate (about 0.69 per index), the 256-attempt cap and the deterministic last resort never reached. With the record's 10^5: 306,980 seeds on the frozen tip, 0 exhausted. Meaning per tier: unchanged from the 10^5 record (no epoch seed fails to draw; the bound under 1e-5 per seed at 95 percent on 3e5); the index-37 seed costs that epoch 38 candidate draws, about a minute of CPU on one core for every node, once | PARTIAL, 0 of 306,980 exhausted; the bound re-runs on the class v6 generator from the served kit (the D1 harness), where it belongs |
+| F1 shadow redundancy toward 10^6, PARTIAL (stopped on the same drain order) | class v5 programs on 1c420786 through the string-seed path with the (c''') floor, indices 0 to 349,999 on build-4 (40 threads, 8 from 13:46 to 15:30 BST on the coordinator's yield to the class v6 census lane) and 350,000 to 999,999 on box 2 (one 80-thread lease, then 52, then four 13-core parts after two release-class pre-emptions), the flush-tested harness (binaries 14180ef4 and 42ee04c7 with --start), rows in `scale-1c420786-f1/out*/census.csv` on both boxes, deduped by idx | 217,310 distinct programs (build-4 83,000 of the lower range, box 2 134,310 of the upper), differential mismatches 0, verifier mismatches 0, 0 panics; saved mean 0.623 percent, max 5.0781: TWO programs over the 5 percent letter, `attack-f1/392513` and `attack-f1/865158`, both at attempt 0, 6,912 to 6,561 per iteration (13 of 256 per pass), the exact saving of the v4 10^5 letter miss `attack-f1/37341` (AP-F1-1); next 369298 at 4.6875. The parity reading (section 7.3 explain, 7.4 emit-c and clang -O3) was queued at class measure on both boxes from 15:21 BST and never got a core (release-class waiters ahead on box 2, build-4 full), so under AP-F1-1's ruling the two read as letter misses at honest-compiler parity, PASS, THE PARITY VERDICT OWED and not made; the rate at 2.2e5 (2 in 217,310, 9e-6) is the v4 rate at 10^5 (1 in 100,000). Meaning per tier: unchanged from the 10^5 record if parity holds (the compiler takes the same 13 and no chip gains); if the D1 harness reads these two on the class v6 generator and the compiler does not find the 13, the shadow redundancy bound is the finding to carry there | PARTIAL, 2 letter misses of 217,310, parity unread; the bound and the two programs' parity re-run on the class v6 generator (the D1 harness) |
 | F9 exhaustion count | 10^5 chain-shaped seeds on the v5 chain path with the dn3 state, box 1, ten parallel chunks (the chain draw costs about 2.2 s per candidate through (c''), so 10^6 is about fifty hours); killed before its first chunk closed, re-queued through `lease pool` | pending | pending the re-queue |
 | F1 shadow redundancy | 10^5 class v5 programs through the string-seed path, box 1; the known firings fire under v5 (planted 50 of 256: 19.53 percent; the real block 0.000; the must-not-fire 1.157); the census killed before its end, re-queued through `lease pool` | pending | pending the re-queue |
 

docs/analysis/attack-pass/f4-weakday.md

@@ -337,3 +337,40 @@ and measured; the agreed numbers above replace this record's M1 figures wherever
 v5 list takes adv-mixer-2's form. Harness note: `attack-f4 plant` and `day` default `--median` to 221 when the flag is
 absent (the plant lines of the class v5 run this evening read "median 221" for that reason); every census line in
 this record was run with `--median 231` and is restated against 226 above.
+
+## 10. The 2^28 day-key scan on 8ca66afa (8 October 2026, 11:15 UK; the strengthening line)
+
+Run: class-v5 8ca66afa (AP-F4-1's redraw rule), the gated binary (sha256 fb872a5d...) carried from build-1 to build-4,
+`lease pool 8 --class measure`, `census --from 20729 --count 268435456 --threads 8 --dedupe --class v5`, 379.2 s
+(about 11 us per day per thread), `scale-8ca66afa-f4/census-2p28.md`, 10:15 UTC. The day-key draw is
+`MixParams::with_shape(seed_words_from_bytes(day_bytes(d)), Shape::for_class(..))`: the mixer shape is the same on
+class v4 and v5 (x8, cache 2^26) and the state leaves never enter the key, so the census header's "class v4 shape"
+is the v5 draw by construction, as section 9's byte-identical reading found at 2^24.
+
+| Metric | Reference | Days over 1.1x | Against 2^-20 |
+|---|---|---|---|
+| M1 per-day LUT datapath | median cost 226 (gain over 1.1x = cost under 206) | 0 of 268,435,456 | under |
+| M1 against the mean 225.79 (sd 6.07) | cost under 205.26 | 0 | under |
+| M2 DSP-bound datapath, 16/(16-k) | k at least 2 | 0 | under |
+| ROT value (wiring) and RC value (inverters) | exact 0 ops moved on every day | 0 | under |
+
+M1 cost: mean 225.791, sd 6.073, median 226, min 206 (day 27,016, 1.0971x; sixteen days share the floor, all with
+NAF sum 158 and M2 k 0), max 258 (day 221,075,548). No day under 206 exists in 2^28, which is the redraw rule's floor
+holding at sixteen times the gate's count.
+
+Classes against the analytic expectation over 2^28 days: ROT all equal 0 (0.010 expected); ROT distinct at most 3
+8,228 (8,237); ROT distinct at most 4 412,703 (412,524); ROT max multiplicity at least 4 568,392 (630,744); ROT
+same-word pair sums to 32 32,997,408 (32,996,609); ROT any pair sums to 32 160,354,008 (161,256,979); ROT all 8 in
+{1,2,30,31} 19 (20.6); MUL any = 1 or all-ones 0 (2.0 each); MUL popcount at most 2 0 (64); MUL popcount at most 4
+8,838 (9,984); MUL under 256 130 (256); MUL two equal 18 (15); RC any zero 1 (1.0; day 109,542,046 at cost 234,
+gain 0.966x); RC popcount at most 4 or at least 28 82,815 (82,886); RC + rk = 0 for any of the 72 keys 87 (72;
+1.8 sigma); RC two equal 8 (7.5). The ROT-distinct histogram matches S(8,d) 31_d / 31^8 to three figures at every d
+(d = 8: 100,095,148 against 0.37294 x 2^28 = 100,110,000). Two cells sit under expectation by more than Poisson
+would give (MUL popcount at most 2: 0 against 64; MUL under 256: 130 against 256): both are the redraw rule's
+refusals (a MUL word under 256 has NAF weight under 9 and a popcount under 9, which the rule's cost-A bound rejects
+on most of its days), so the deficit is the rule working, not a bias.
+
+Verdict: PASS (0 of 2^28 days over 1.1x on every datapath; the floor 206 and the expectation lines unchanged from
+2^24). Consequences per tier: no chain day in the first 735,000 years of days gives a per-day datapath a 1.1x
+discount, so an FPGA or ASIC that bakes a day's constants gains at most 1.097x on its best day and no chip buys a
+weak day; a GPU miner is untouched; the day draw is uniform at 2^28.

docs/analysis/attack-pass/f8-uniform.md

@@ -257,3 +257,33 @@ Sample sizes reached: 2^28 derivations (lines); 10^6 nonces on three programs (d
 2^26 nonces (cross-hash histogram); 64 seeds at 2^18 nonces (the census).
 
 Times UTC in the logs; the runs ran 08:13 to 09:2x UTC on 7 October 2026 (09:13 to 10:2x UK).
+
+### (e) The hot-set gate to 256 seeds on the frozen class v5 tip (8 October 2026, 11:07 to 12:37 UK; the strengthening line)
+
+Run: seeds p66 to p257 (192 new; with the gate's p2 to p65, 256) at 2^24 nonces each, chain path, class v5 on the
+freeze 1c420786 (binary sha256 0f5c98dc..., pairing e5a4ac5978462156, the dn3 state e000824c...), window-model
+control, by-site, box 2 as three thirds under `lease pool 32 --class measure` (10:07:14Z to 11:37:13Z; 4,291, 4,287
+and 5,315 s). Validation: three library comparison samples at 0 mismatches; `Epoch::hash_warp` agreement on 37,440
+warps at 0 mismatches. Ratios in `scale-1c420786-f8/ratios.tsv`, logs in `log/third-{A,B,C}.log`.
+
+| Reading | 192 new seeds | With the gate's 64 (256) |
+|---|---|---|
+| Within 1.2x of the window model at the top 0.1 percent | 184 (95.8 percent) | 245 (95.7 percent) |
+| Over 1.2x | 8: p110 1.3527x, p66 1.3403x, p234 1.3156x, p145 1.2750x, p77 1.2664x, p225 1.2457x, p248 1.2188x, p89 1.2065x | 11 (the gate's p10 1.5047x, p8 1.3787x, p4 1.2166x) |
+| Mean ratio; bands | 1.023; under 1.00: 45; 1.00 to 1.02: 114; 1.02 to 1.05: 10; 1.05 to 1.10: 7; 1.10 to 1.20: 8 | |
+| Hot-set verdict on the windowed control | clear on all 192 (largest X_f/f +0.60 at p110, +0.52 p234, +0.47 p66) | clear on all 256 |
+| Hot-set verdict on the flat control (the auditor's first view) | p76 and p88 at X_f/f +1.07 (both within 1.2x on the windowed control: the window model explains them) | |
+| 6-sigma largest 64-item bucket against the window density | flagged on 57 of 192 (p110 +61.67 sigma): the AP-F8-1 tail mechanism | |
+| Hottest item on the over seeds | 263 to 432 reads of 2^31 (0.0000 percent) | |
+| Predicted sources | 190 none; 2 one-one-bit through a load: p212 (1.1915x, attempt 4, site instr 7, r5, last writer load at 2), p225 (1.2457x, attempt 0, site instr 37, r5, last writer load at 36) | |
+
+The over-1.2x rate at 256 seeds (4.3 percent) is the gate's rate at 64 (4.7 percent); the worst ratio fell (1.3527x
+against 1.5047x) and every over seed has the tail's shape: one narrow-window site a quarter bit short, the hottest
+item at a few hundred reads, no hot set. p212 and p225 are the only seeds with a predicted source (one-one-bit
+through a load), both passed by (c''') on the frozen tip; they go to the hash lane for by-site attribution as the
+gate's tail did.
+
+Verdict: PASS at the gate's reading. Consequences per tier: on any of 256 epochs the top 0.1 percent of items carries
+at most 0.23 percent of reads (the window model's own 0.17), so no card or chip gains a cacheable hot set; the excess
+class is the named bucket concentration, at the same rate at four times the count, and the per-site largest-bucket
+bound stays a next class's item.