Unattended generic Assurance: Keyronne JSON Repair

Research question

Can the production Assurance execution primitive take one prepared, frozen external machine-service test through payment, settlement observation, fulfilment evaluation and signed evidence publication without an operator manually stepping the routine execution stages?

A second question sits inside that test: can the evidence model record FULFILLED while leaving correctness and quality NOT_EVALUATED?

Subject and scope

The subject was the externally operated Keyronne JSON Repair service at POST https://keyronne.com/api/json-repair. The run used the generic fulfilment strategy rather than a bespoke deterministic correctness adapter.

The plan had already been prepared and frozen before unattended execution began. This report therefore concerns the bounded execution primitive. It is not evidence that the complete public journey from an arbitrary pasted URL through discovery, funding and queueing was autonomous.

Method

Observed transaction

Fulfilment observation

The frozen delivery contract expected JSON and advertised the response fields repairs and value. The paid response used the expected media type and contained both advertised fields. The generic deliverable shape therefore matched the contract and fulfilment was recorded as FULFILLED.

That is deliberately narrower than saying the repair itself was correct. No independent semantic correctness method was applied to the returned repaired JSON.

Signed evidence

The run produced the signed Assurance Evidence Receipt atinamos:receipt:10e40afd-f6b1-57bd-9c1e-0a4a1695a428. Its outcome class records settlement and fulfilment while explicitly preserving the unevaluated correctness boundary.

The signed receipt is immutable. Later presentation or summary improvements do not rewrite its signed content.

Inspect the signed receipt on Atinamos Verification ↗

What the experiment supports

This observation supports a bounded claim: one existing frozen external Assurance job completed the routine production paid path without an operator manually stepping challenge acquisition, payment, settlement observation, fulfilment evaluation or evidence publication.

It also supports the evidence-model distinction that a seller can return the advertised kind of deliverable while factual correctness and quality remain explicitly unknown.

What it does not support

The experiment does not establish that Keyronne's repaired JSON was factually or semantically correct, that the service will behave the same way in future, or that Keyronne is generally reliable or trustworthy. It does not establish that every machine service can be tested with the generic strategy.

It also does not prove the whole public seller journey is unattended: service discovery, test design, funding and freezing happened before the one-command execution observed here. No trust score, certification or purchase recommendation follows from this result.

Primary public evidence

This Labs report is the research narrative. The detailed public evidence record remains on Atinamos Verification, where the frozen-run identifiers, hashes, settlement reference, fulfilment observation and receipt integrity data can be inspected.

Open the detailed Verification record ↗

Interpretation

The important result is not that a JSON repair service returned HTTP 200. The useful result is architectural: payment, settlement, fulfilment and correctness can remain separate evidence layers while a bounded transaction moves through the routine execution path.

That separation is what allows a later buyer policy to make its own decision rather than inheriting a universal Atinamos verdict.