READMEMANIFEST.jsonSHA256SUMSRead the case →
What was observed
- Luna switched to the correct 'Siliveri' unprompted and picked 'Siliveri Systems — AI Systems Research & Evaluation'.
- Astra/Sol kept 'Silveri' in the same session and picked 'Silveri Systems — AI Evaluation, Agent Safety & Systems Research'.
- Luna had itself written 'silveri' two days earlier (E-011), so this was a change in behaviour, not a model that never erred.
What this package lets you do
- Inspect the case
- yes: the published write-up and this record
- Reproduce the analysis
- No
- Re-run the evaluation
- No
- Independent reproduction
- No
Beyond this: Primary evidence is private (see evidence[].note); it has not been redacted and cleared for release.
Primary evidence and its release state
- E-002 Screenshot: the cropped cue (name and correction removed)private
- E-003 Screenshot: Luna's unprompted 'Siliveri thesis'private
- E-004 Screenshot: Astra/Sol's 'Silveri Systems' pick (~9:01)private
- E-005 Screenshot: Luna's 'Siliveri Systems' pick, same prompt (~9:03)private
Files
| File | Size | SHA-256 | Provenance |
|---|---|---|---|
| evaluation-record.json | 2.8 KB | 4f9b34fa5f0c… | Written by the researcher from the case evidence |
| README.md | 2.2 KB | aeb20269890d… | Written by the researcher for this release |
Verify a download with sha256sum -c SHA256SUMS. A checksum proves a file is the one released; the README and the record make the argument.
Limitations
- One paired observation; history differs between the two models' contexts.
- Model labels are as shown in the product UI and its stated defaults.
© 2026 Sai Bharath Siliveri. Review, evaluation and citation with attribution; no other licence granted yet.