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

attack-pass: AP-F8-1 tail attributed (p4, p8, p10, p34 as per-site bucket concentration at a narrow-window site; the largest-bucket bound named for a next class)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-08 08:48:03 +00:00
parent d8459e7918
commit 7ac25503ed
1 changed files with 15 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
docs/analysis/attack-pass-2026-10.md
View file

@ -506,10 +506,10 @@ control, box 2; stream 07a809a7 / 8bdcbdd8, pairing id a788661687db4bb3): FAIL.
| p15 | 2.57x | 2, 12 | 4.49 percent | 0.982 | zero through rotl at 0 |
| p18 | 2.50x | 6, 30 | 5.55 percent | 0.937 | all-ones through a load |
| p56 | 2.01x | 2, 10 | 3.34 percent | 0.994 | unattributed (new over sub-version 1) |
| p10 | 1.50x | 8, 28 | 2.04 percent | 0.979 | unattributed (identical to sub-version 1) |
| p8 | 1.38x | 14, 51 | 1.42 percent | 0.980 | unattributed |
| p34 | 1.25x | 1, 13 | 1.35 percent | 0.997 | one-bit value through sub |
| p4 | 1.22x | 1, 8 | 1.45 percent | 0.981 | unattributed (1.57x on sub-version 1) |
| p10 | 1.50x | 8, 28 | 2.04 percent | 0.979 | bucket concentration at a narrow-window site (r0, window 2^22, offset 1, last writer mad at 20): largest 256-item bucket 5.6x window expectation, index entropy 13.71 of 14 bits, hottest 0x4004da at 362 reads, source none (identical to sub-version 1) |
| p8 | 1.38x | 14, 51 | 1.42 percent | 0.980 | bucket concentration at a narrow-window site (r7, window 2^22, offset 2, last writer xor at 44): largest bucket 3.1x, entropy 13.72 of 14; plus site 6 (instr 33, r3, window 2^23, mad at 30) at 0.834 percent, bucket 3.5x; hottest 0x837de4 at 420 reads, source none |
| p34 | 1.25x | 1, 13 | 1.35 percent | 0.997 | one-bit value through sub (r3, window 2^23, offset 1, last writer sub at 5): largest bucket 3.5x, entropy 14.96 of 15, hottest 0x800010 at 541 reads, saturated source 0.0001 percent |
| p4 | 1.22x | 1, 8 | 1.45 percent | 0.981 | bucket concentration at a narrow-window site (r2, window 2^22, offset 1, last writer mad at 4): largest bucket 4.5x, entropy 13.74 of 14, hottest 0x4000e7 at 355 reads, source none (1.57x on sub-version 1) |
Every failing seed is one low-entropy load site. Clean-seed spread of the per-site bucket entropy (848 site rows of
sub-version 1's 53 clean seeds): min 0.9865, p1 0.9961, p5 0.9999, so bucket entropy separates only the strong four;
@ -525,7 +525,17 @@ word indices over 2^20 evaluations with the shadow executed against the uniform
site ratio 0.9960, p1 0.9990, median 1.0000; the strong five p23 0.8361, p18 0.9274, p19 0.9335, p15 0.9432, p56
0.9654; floor 0.98 sits 0.015 from each side. At 2^24 the weak four (p34 0.9181, p4 0.9614, p8 0.9630, p10 0.9612)
share their value with two clean seeds (p44 0.9612, p52 0.9613), so the 2^24 stage is not taken and p4, p8, p10 and
p34 stay the open tail, unattributed. The ratio refuses about 4 percent of candidates that pass every other test
p34 are the tail. The tail attributed (hash lane, 8 October 2026, 09:40 to 09:46 UK, attack-f8 census at 2^24 with
the window-model control and by-site attribution, tree b38b4af6 on the frozen 017e7037; the gate ratios reproduced to
four places, p4 1.2169x, p8 1.3774x, p10 1.5036x, p34 1.2501x, the hot-set verdict clear on the windowed control for
all four): each is a per-site bucket concentration of about a quarter bit (0.26 to 0.29 bits short of 14 on a 2^22
window; p34 0.04 of 15) at one narrow-window load site whose last writer is mad, xor or sub, with every other site at
its flat share; the ratio tracks the largest-256-item-bucket excess (5.6x gives 1.50x, 3.1x to 4.5x give 1.22x to
1.38x); (c'') passes them at 0.9927 to 0.9963 because distinctness does not see a bucket; the check that would catch
all four is a per-site largest-bucket bound (about 2x window expectation at the 2^20 units), a generator change for a
next class, never for the frozen ones. Meaning per tier: a quarter bit at one site is under the window model's own
spread (the gate line's 61 of 64 stands), so no card or chip gains a cacheable hot set from it; the bound is the next
class's item, not a change to 017e7037 or 1c420786. The ratio refuses about 4 percent of candidates that pass every other test
(4,099-program census at 017e7037: mean attempts 2.086 against 1.998, max 17, 0 lossy-sourced load sites of 65,584,
0 exhaustions; suite 103 of 103; devnet epoch-0 at attempt 1, id a785001687d8688a, pairing verified by this lane).
RE-GATE VERDICT on sub-version 3 (017e70376489251e18564c0abce7e466e606c8b3; pairing id a785001687d8688a verified;