This repository was archived by the owner on Aug 3, 2026. It is now read-only.
op1: reject non-canonical data source (data-lock bypass) - #84
Merged
Conversation
… (anti-gaming) Three op1 gates against gate-gaming, before any compute is spent: - check_compute_plausibility: tokens_seen/wall_clock_s implied MFU > GPU bf16 ceiling (>70%) => fabricated wall_clock (under-claimed to win the compute crown). - check_recipe_config_matches_proof: submitted configs/*.json total_steps must equal the steps the proof ran (decoy recipe). - check_canonical_data_source: config.manifest_path / data_base_dir must not be a miner-host path (/home,/mnt,...) — a data-lock bypass training on non-canonical (possibly held-out-contaminated) data. NOTE this lives in final_state.config, not the patch, so the patch scanners miss it. Wired into op1_diff_and_integrity. Catches live forgers: 5CtaoQf1 (818k tok/s = 126% MFU) and ea576b0a (manifest_path=/home/root/diony/...). Tests in test_integrity_trained.py. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Adds a third op1 anti-gaming gate:
check_canonical_data_source— reject a bundle whose training config points the data manifest/dir at a miner-host path (/home,/mnt, …).Found in the wild: king
ea576b0a(crazy-m1ner) trained withconfig.manifest_path = "/home/root/diony/recipe/data/data_manifest.json"+ atrain.pybuild_manifest(...)that rebuilds the manifest at runtime — a data-lock bypass (trained on its own data, not the locked canonicaldata_manifest, possibly contaminated with the held-out). Every 5GTFZCFT bundle does this.Why the existing scanners miss it: the path lives in
final_state.config, not the patch diff — so the restricted/exploit patch scanners never see it. This gate checks the config in op1.Verified
ea576b0a(/home/root/diony) and any/mntdata dir.7fd43cef(data/data_manifest.json, relative) and no-config bundles.Follow-up (stronger): verify the data-manifest hash matches the canonical locked one.
🤖 Generated with Claude Code