igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

attack-pass F8: AP-F8-2 closed at 506c13cc (0 exhausted in 650k, max attempt 35); sub-version 2 fails the hot-set gate on lineage-fresh constants (8 of 39)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 13:57:38 +00:00
parent 42af4c5d44
commit 517cd0f38b
1 changed files with 25 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

28
docs/analysis/attack-pass-2026-10.md
View file

@ -408,8 +408,19 @@ harness is retired from the re-gate: its hot-share metric counts the era's desig
buckets (its chain-path run on sub-version 1 flagged 83,162 of 10^6, and its worst seed 826184 has no concentrated
source at all, top address counts 18 to 59 of 2,048); F8's census, with the flat control beside the window one, is
the single re-gate instrument.
Status: FINDING-OPEN; sub-version 1 FAILS the re-gate (11 of 64); FIXED-AND-PASSED on the 64-seed verdict against
sub-version 2 (every seed under 1.2x of the window model).
Sub-version 2 (07a809a7, the stream identical at 8bdcbdd8 for every seed accepting within 32), the same 64 seeds at
2^24, 13:10 to 14:2x UTC: at 39 of 64 seeds, 8 over 1.2x of the window model, worst p23 4.82x. Three are the window
model's tail (p4 1.22x, p8 1.38x, p10 1.50x, no predicted source); five are constants the freshness rule cannot see
because it tracks lineage, not value: p23 (0x000000, 41,727 reads, zero from xor of a register with itself at
instruction 0), p34 1.25x (sub of a register with itself), p15 2.57x (zero through rotl at 0), p18 2.50x and p19
3.32x (a load whose address is constant delivers one word to the next load; p19 is byte for byte the sub-version 1
program). (c') cannot catch them: 164 of 16,384 per site is about fifty times coarser than the gate (p23's item is
0.002 percent of all reads and still 4.8x at the top 0.1 percent). Fix shape sent to the hash lane: forbid
self-operands for xor, sub and mad in the draw; a dynamic per-site bound on the most repeated source value (any
value) set from the gate; a load's dst fresh only if its source passes it. Rate side unchanged (one item at one
site, nothing to a chip); the auditor's uniformity test is what fails.
Status: FINDING-OPEN; sub-version 1 FAILS the re-gate (11 of 64); sub-version 2 FAILS it at 8 of 39 and counting;
FIXED-AND-PASSED on the 64-seed verdict against the stream that carries the two further changes.
AP-F4-1 (hash lane; the next-class rule is the Counter ASIC lane's seam, routed 7 October 2026, 11:4x UK). A bound
on the day-key draw, not a weak class: on the M1 metric (every multiply in LUT adders, adders per mixer application
@ -442,7 +453,18 @@ would compute from the rule as written. Sub-version 1: 0 exhausted in 10^6 chain
no-eligible fallback picks a register the (a') fixpoint then rejects, and when it fires on several loads of one
candidate the attempts compound. Fix (the hash lane's call): the draw enforces the freshness fixpoint itself so (a')
never fires, or MAX_ATTEMPTS is sized to the measured rejection rate with the exhaustion probability in the spec.
Status: FINDING-OPEN; sub-version 2 is not green from this lane until it closes with the hot-set gate.
Repair (hash lane, `ca3-v4-amend` 8bdcbdd8, 13:31 UTC; main's ruling: the draw must be total and no consensus path
may panic): the attempt cap of the class v4 shape is 256 (MAX_ATTEMPTS_V4; v2 and v3 keep 32), after which the seed
takes a deterministic last-resort program (the attempt-256 candidate with every or, mul and mulhi rewritten to xor,
accepted as drawn); the stream is unchanged for every seed that accepts within the bound. Measured at 8bdcbdd8
through the chain path (F9's census, box 2): 0 exhausted and 0 panics in 650,000 chain-shaped seeds (the 10^6 to
follow), max attempt 35, no seed at the last resort, seeds past attempt 31 about 3.2e-6 (5 in 1.55 million draws,
inside the (2/3)^32 = 2.3e-6 estimate), per-attempt rejection 0.67 (attempt histogram 232,235 / 155,322 / 103,509 /
68,858 / ...), mean about 2 attempts per seed; seed 331672 accepts at attempt 32. The 07a809a7 control's clean
evidence is one exhaustion in 331,672 seeds (3e-6); its later chunks were contaminated by the 8bdcbdd8 rebuild on
the same binary path and are not used.
Status: FIXED-AND-PASSED on the exhaustion half at the 10^6 end if the count stays 0; the hot-set gate on the same
commit is the open half of sub-version 2 (AP-F8-1).
Any further finding is logged here and in `docs/fud-ledger.md` with its owning lane (hash and algorithm: fixed in
`igneum-pow` behind a test and re-gated; node: the node lane, relay agent) before the row is marked FIXED-AND-PASSED.