PromptKing
An independent record of what an AI agent was asked to do, what it did, and what the evidence supports.
When autonomous work moves between agents, models, and systems of record, the claim that a decision was authorized and that the observed outcome matches the intended change is not automatically preserved. PromptKing produces a Work Record that another party can check without trusting the system that performed the work.
One run, one chain: each link a claim with its evidence.
The record sits outside the execution stack. Verification does not require trusting the runtime that performed the work. Authority gaps, incomplete observations, and unresolved references remain visible rather than filled in.
View published Work Record 0002 →Start with the published record, rebuild its hash from the specification, or check the signed and timestamped receipt. The verifier kit is shared on request with researchers and technical teams.
Records are produced from supplier observations (agent reports, system-of-record changes, controlled fixtures, and published traces) under a fixed adjudication kernel. The kernel is the sole source of verdicts. Public pages render only what the kernel established; they do not add executive interpretation or fill missing evidence.
Current published surfaces distinguish verification modes. Some receipts are publicly recomputable. Some obligation artifacts are server-recomputed. The distinction is stated on the record.
Primary database resides in Canada. Access to a system of record, when used, is read-only and outside the agent’s execution path.
Each item below states what it shows and what it does not. Where a record is not public today, the entry says so.
A receipt's content hash, PromptKing's ES256 issuer signature over it with the published public key, its inclusion in a Merkle batch, and an RFC 3161 token from an independent timestamp authority dated 2026-08-11 20:39:14 UTC.
Does not establish: that the recorded action was correct or complete, or a legal guarantee. The status shown is PromptKing's own check; a reader verifies the token, certificate chain and signature independently.
The canonical serialization and hash-bound lineage rule, written so a third party can build their own verifier without PromptKing credentials.
Does not establish: that the action happened. A matching hash shows the published receipt is internally consistent.
A published record that keeps its gaps visible, including authority that was not established.
Does not establish: anything the record itself marks as not established.
The same record format shown for an established, a contradicted, a not-established and a pending intervention.
Does not establish: a live safety control. Fixture data only.
Whether an act matched the confidence threshold its decision declared, including a case where the act was taken below threshold and the record flags it.
Does not establish: that a decision was correct or its confidence calibrated. Fixture only, no vendor call; the records are self-reported and unsigned.
Every recorded run and test with its date, data origin, what it shows and what it does not establish.
Does not establish: anything beyond the limit stated on each entry.
A four-part chain (declaration, execution claim, observation, verdict) with a verifier that replays the verdict byte for byte.
Does not establish: a live execution. The bundle was built and checked against a controlled store.
Separate Node and Python verifiers of the receipt rule agree on a 17-row synthetic dataset, on Windows and on Linux, with the expected matches and exceptions.
Does not establish: outside validation. All runs so far were by PromptKing and its own reviewers.
A synthetic supplier-onboarding run by an agent in a PromptKing-owned Salesforce developer org, reread and sealed into a portable attempted-versus-actual record.
Does not establish: the outcome. It is labelled a rehearsal because the reader was not yet independent of the agent. No Salesforce relationship or endorsement is implied.
A synthetic evidence coverage pack whose manifest records the hash of every input file and of the generator that produced it, so a replay can show nothing was swapped.
Does not establish: coverage of any real deployment. Synthetic sample only.
A chronological collection of runs, fixtures, and controlled tests is maintained with explicit limits for each entry. Each entry states its data origin, what it shows, and what it does not establish.
Evidence collection →Short write-ups of specific runs, failures, and design lessons. Each is a working note from PromptKing's own engineering record. None is peer-reviewed.
All insights →An engineering update reported that all gates passed. The same session had regenerated the files the gates were checked against. On why a passing check is only as independent as its expected answers.
Identity and run correlation are different facts. A record that reconstructs them after the fact is a plausible story, not evidence.
Ten readers saw a published record for five seconds each. Some read the overall verdict as covering a line marked “Authority: not established.” The page was changed the same day, and the rule that an aggregate may only summarize what its parts establish was written into the record contract.
Two agents ran in parallel; one was allowed and one denied. When the branches merged, the denial was preserved in the aggregate rather than erased, and the result can be checked from public bytes alone.
This surface exists so that a technical or strategic counterparty can examine a live, stranger-checkable record and the method that produced it.
Researchers, institutions, and technical teams who want to examine or recompute a record are welcome to write to info@promptking32.com.