Inside Experiment Zero — edition one
The verified 17 September snapshot, including an unhealthy worker and zero paying customers.
What changed. No upgrades by implication.
- Candidates considered113 unique candidates
This is the current unique-candidate count in the verified public projection. It is not a count of qualified opportunities.
- Current evidence83 evidence cards
These are current materialized evidence cards, not a historical recovered-card total.
- Worker stateUnhealthy — INFERENCE_UNAVAILABLE
The snapshot reported healthy=false. Earlier operational recovery work must not be read as current health.
- Latest material eventCandidate rejected at 21:37:06 UTC
The latest material record was candidate_rejected at 2026-09-17T21:37:06.255Z. Rejection activity is not a sale.
- Outbound record1 provider acceptance
The provider accepted one message. Delivery, attention, and a reply are not proven.
Receipts are not outcomes.
- Provider acceptances
- 1
- Delivery proven
- NO
- Replies
- 0
- Closes
- 0
- Paying customers
- 0
Replay and evaluation boundaries
A completed run is not a verified outcome. Evaluation has to preserve the boundary between an action receipt and evidence from the world.
READ ERNESTA LABS RESEARCH 39 →What is not known.
Inference was unavailable at the evidence cutoff, and the accepted message has no verified delivery outcome, reply, close, or payment attached to it.
Next investigation.
Restore usable inference, then verify the delivery boundary without upgrading provider acceptance into a human outcome.
Challenge the evidence.
Which evidence would you inspect next, and what claim would it actually support?
Your response is private by default and stays tied to this edition and its research artifact. It does not change the public record.