Challenge Ledger · Open · Unfunded
Run it. Break it.
Tell us what you find.
The canonical claims are public. The ledger is open to every replication, positive or negative. No cash bounties. Co-authorship instead.
Canonical claims
| # | Claim | Value | Whitepaper | SHA-256 |
|---|---|---|---|---|
| 1 | DRAM working set multiplier | 4.1445× | WP-1 | e435686a5d9da80ef3c9edf5f569b457a688619aa55ae10211896b5e3349d2f1 |
| 2 | Water conservation | 97.6% | WP-2 | fbf0ef6c2510059d29a0d6079e0fd1edc2fdba3e6a091d53664b2349027badea |
| 3 | Zero-copy IPC · 4-hop nominal | 65.9 µs | WP-3 | 1743f9fcb97d741d9eb29958a7f42630516637592a3dcec85f2d76e0d59c4428 |
| 4 | Sub-cycle grid defense | 45 MW / 16.7 ms | WP-4 | 0f5f6572f15d3fd7d610ddd24ab6560af1af77831ed7f60cbd73c8febdb0c7c8 |
Honesty constraint. Every value above is formula, derivation, or model — per manymoats.com/papers. Not a meter reading. Replication is the test.
Replication protocol
1
Run
Open
./battlegrounds.html in any modern browser. No install. No account.2
Compare
Within
±10% → replicated. Outside ±10% → refuted or configuration difference.3
Submit
Email
ledger@manymoats.com with hardware config, raw output, methodology, interpretation.4
Publish
Replicated → co-author next revision. Refuted → co-author correction. Inconclusive → co-design better test.
Ledger
| ID | Target | Challenger | Objection | Status | Artifact |
|---|---|---|---|---|---|
CHAL-001 |
Claim 01 (DRAM) | Claude Opus 5.5 Pass 5 | 5.14× multiplier neglected 10 GB macOS reservation | REVISED | Clamped to 4.14× (174.1 GB managed equivalent) |
CHAL-002 |
Claim 02 (Water) | Kimi V4 Audit | Conflated total campus water draw with evaporative tower loss | REVISED | Reconciled latent ceiling (0.94M gal/day) and blowdown |
CHAL-003 |
Claim 03 (IPC) | Kimi V4 Audit | Thunderbolt 4 cannot provide cache-coherent shared memory | REVISED | Re-anchored to PCIe Direct-DMA NTB with virtual channels |
CHAL-004 |
Claim 04 (Grid) | Claude Opus 5.5 Pass 5 | 50 MW shed broke 3.0 MW/ms ramp cap; flyback at 63.7V | CONFIRMED | Clamped to 45 MW (2.70 MW/ms), 57.1 V flyback, 1 GVA model |
CHAL-005 |
Claims 01–04 (Fleet) | DeepSeek Math Audit | Cross-claim arithmetic & dimensional precision stress-test | CONFIRMED | Reconciled 42GB arena, 71% misting duty, 4-hop IPC ack, 113.0 kA/s |
Negative results published with equal prominence. Living document.
What we will not do
No cash bountiesCash attracts bounty farmers. Co-authorship attracts people who care about the truth.
No suppressed negativesFalsify a claim → we publish it and update the canonical value.
No attacks on criticsFind a flaw → you are doing us a favor. Try to break it now.