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

adv-cache-2: line shard s0 (internal adversarial pass, not an independent review)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-07 20:33:29 +00:00
parent 2d0104af5e
commit 2157f553cc
1 changed files with 8 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
docs/analysis/cryptanalysis/report-chained-cache-2.md
View file

@ -76,6 +76,13 @@ Log: `logs/adv-cache-2/lines-2e32.log`. 4,294,967,296 item derivations, 34,359,7
The binary's verdict line says FLAGGED on one per-day, per-round test: day 20823, round 2, one 64-line segment at 362 reads against a mean of 256 (+6.62 sigma). That is one bucket in 256 days x 8 rounds x 65,536 segments = 1.34 x 10^8 Poisson(256) draws; the chance of a 6.6-sigma bucket somewhere in that many is about 1 in 70 by the Gaussian tail and more by the Poisson skew, so it is a tail event, not a structure (the day's pooled rounds and every pooled statistic sit at the control's values). A chip that stored that segment for that day would serve 64 lines at 1.4x their uniform share for one round of one day: nothing. The bound: at 2^35 reads over 272 chain days (16 + 256) the line index is uniform to the control's precision in every statistic that can price a store.
### 1.2 Further line-census shards (lease, 32 cores)
| Shard | Days | Reads | Lines: largest / smallest (sigma), chi2/dof | Segments: largest / smallest | Top 1 percent of lines vs control | Verdict |
|---|---|---|---|---|---|---|
| s0 (20:24:04Z to 20:31:19Z, 261 s of run on 27 to 32 cores) | 20746 to 20809 (a duplicate of 64 days inside the 256-day census above: an oversight in the shard plan, kept as an independent re-read) | 2^33 | +5.26 / -4.97, 0.99924 | +4.06 / -4.55 | 1.05937 vs 1.05938 percent (1.0000x) | PASS |
| s2b, s3b, s4 | 21002 to 21193 (fresh) | 3 x 2^33 | queued | | | |
## 2. Q2: the hot set across the hashes of an epoch
### 2.1 Devnet 3, epoch 0 (program `fce15bf61030be57`, day 20733), 2^24 nonces
@ -281,6 +288,7 @@ The first load of a hash (site 0 in iteration 0) depends only on the nonce, the
| Q2 Devnet 3 v3 diagnostic 2^24 (lease, 48 asked) | 2 | about 5 min | 48 | 0.1 |
| Q2b era-fixed 12 to 23 (lease, 40 cores, killed) | 2 | 416 s | 40 | 0.05 |
| Q2b era-drawn 1 to 2 (lease, 40 cores, killed) | 2 | about 3 min | 40 | 0.02 |
| Q1 shard s0 (lease, 27 to 32 cores) | 1 | 435 s | 32 | 0.04 |
| Pod-hours | none (no GPU, no rented pod) | | | 0 |
Running total at 20:3xZ: about 1.7 box-hours by wall x threads / 96 (the boxes ran at load 400 to 600 for the first two hours, so the CPU actually consumed is under that). The 8-hour reading is not near. The 8-hour reading is not near.