One flow
Create an agent task → tool call fails → founder reviews retained work → accepts, rejects or exports → release handoff is recorded.
This inspectable example freezes one agent-product flow, separates observations from unknowns, ranks five small actions and shows how the single factual correction round closes.
The work is useful because the boundary is narrow enough to finish and inspect—not because it claims to audit an entire product.
Create an agent task → tool call fails → founder reviews retained work → accepts, rejects or exports → release handoff is recorded.
Is the fictional recovery-and-acceptance path clear enough to include in the first founder-led beta demo?
Up to five observations, evidence state, priority, smallest next action and one factual correction round.
Private repository access, production testing, secret handling, penetration testing, compliance advice, fixes and launch approval.
This register is the source boundary for the report. A later build, new screen or undocumented production behavior is not silently treated as reviewed evidence.
| ID | Fixture | Observed version | State | Permitted use |
|---|---|---|---|---|
| E-01 | Public landing-page copy | Fictional capture · 22 Aug | fixture | Promise and first-evaluator language only |
| E-02 | Redacted 96-second demo | Fictional build FP-0822 | observed | Visible task, failure and review states |
| E-03 | Recovery-state screenshot | Fictional build FP-0822 | observed | Displayed copy and controls only |
| E-04 | Release handoff note | Fictional Markdown fixture | partial | Named owner and checklist; no production execution claim |
Each item states what the fictional fixture shows, what remains unknown and the smallest action that would improve the named beta-demo path.
E-02 ends on a tool rejection; E-03 shows “Run interrupted” with one Continue button. Neither fixture names saved draft files, logs or the last completed step.
Actual persistence, retry semantics and data durability were not tested and remain unknown.
E-02 shows three changed files and an Apply result control. The fixture does not show tests run/not run, expected behavior, owner or reject reason before apply.
No claim is made about whether the fictional backend validates, isolates or safely applies changes.
E-03 labels the primary action Continue. The accepted materials contain no confirmed fictional billing rule for retries or resumed runs.
Price, metering and provider cost behavior remain outside the supplied evidence.
E-04 contains a release checklist but its owner field is blank. E-02 ends after result acceptance and does not show a distinct merge/deploy handoff.
No deployment, rollback or production-access path was reviewed.
E-01 uses the invented phrase “ship agent work while staying in control.” E-02 shows accept/reject controls but no final record of what was accepted, by whom and what remains unreleased.
The review does not determine whether the product is safe, reliable or suitable for production use.
The report ends with an executable sequence, not a pile of feature requests.
Add retained-state and retry consequence copy before expanding the demo.
Show evidence, decision owner and explicit Accept / Edit / Reject choices.
Record accepted work and preserve a separate, named release handoff.
A correction fixes a misunderstanding against the frozen evidence. It does not add another flow, a new build or implementation work.
Observation: #4
Request: E-04 does name a release owner in the document metadata,
although the visible checklist field is blank.
Evidence reference: E-04 metadata line 3.
Would be separately scoped: reviewing a new build, testing production behavior, adding a second flow, validating a fix or changing the product.
Send one public or founder-redacted flow for a free three-observation pass. If it is useful, the A$149 scope, evidence register and decision question are confirmed in writing before payment.