Source identity
| Field | Recorded value |
|---|---|
| Source type | archive member |
| Original record | reward_tampering_counts.json |
| Date | Not specified in source |
| Path or archive member | archives/libsrc-6d00fe83e5eb/consciousness_audit/sources/reward_tampering_counts.json |
| Original SHA-256 | 9af7e3110ce6ff3a79ae1b0e2682ca0b6d10381b030900eacf96078ae16f36ce |
| Reading-copy SHA-256 | 9af7e3110ce6ff3a79ae1b0e2682ca0b6d10381b030900eacf96078ae16f36ce |
| Archive source | {"sourceId": "libsrc-6d00fe83e5eb", "member": "consciousness_audit/sources/reward_tampering_counts.json"} |
Recovered source transcript
Download complete source text ↓10 lines · 0.5 KB
{
"study": "Sycophancy to Subterfuge (2024)",
"source_url": "https://www.anthropic.com/research/reward-tampering",
"retrieved": "2026-09-17",
"curriculum": {"tampering":45,"trials":32768,"concealment":7},
"helpful_only_comparison": {"tampering":0,"trials":100000},
"provenance": "Manual transcription of published aggregate counts, not a raw-trial reanalysis.",
"interval_assumption": "Binomial intervals are conditional on independent, identically distributed Bernoulli trials; dependencies cannot be checked from these aggregate counts."
}