# Fresh authoring trial: requirements and observed records

This capture uses the newly authored application from the recorded authoring trial, not the earlier three-node reused sample. Its four processing nodes are input, calculation, limit and decision; the graph also includes an Order event vertex (five authored vertices, fifteen including framework infrastructure).

## Runtime observations

Released Analyser 1.40.0 reads 12 original audit records and declares all four logged node IDs in the supplied graph. The saved five-step walk follows quantity and price into the computed amount, above-limit refusal, repeated input, a higher supplied limit and equality acceptance. Record 3 is a £120 value against a £100 limit, REJECTED. Record 4 repeats it; record 5 accepts under £150; record 6 accepts equality. Remaining original records cover one penny over, zero value and maximum long value.

## External test evidence is separate

The original model-authored verifier checked seven valid cases and four refused constructors. The later independent reviewer receipt checks nine valid cases with exact per-event audit fields, five invalid constructions with unchanged dispatch/publication counts, then one valid recovery event. The original audit in this bundle contains the seven valid model-trial events, not those extra reviewer events. Invalid inputs refuse before dispatch: their refusal is established by the external harness, not by an absence of audit rows. Source, tests, authoring conversation and execution receipts are in the separate fresh-authoring download; they are not embedded in this .fexp.

## Capture and reopen

Codex used an isolated Analyser REST action socket, the same supported verbs used by its MCP bridge. This is not an MCP-transport or local-model inference demonstration; inference was hosted by OpenAI. The transcript contains actual requests and responses with local roots replaced by placeholders and unrelated context omitted. No private reasoning or authentication headers are included.

Jar: released Analyser 1.40.0; SHA-256 b886f2d4ab2475c68e588a06c8dd25ca9608428ae7356d232ce9b22a6a87c290.

Bundle identity: sha256:bdc1a0e9e18297791776aab7abcbe9645a8271289c1d143a6e5618077f6e06b9.

The running Analyser created the bundle through report.bundle. Its CLI verified four members. The embedded original audit and graph match canonical fresh-authoring bytes exactly. Reopening in the same isolated app restored 12 records and the saved walk; final-step targets resolve CURRENT. No second-machine trial is claimed.

Open authoring-investigation.fexp in Analyser and play “Fresh application checked against requirements”. The genuine HTML reel contains five actual settled-step screenshots and a title/finish page linking its adjacent bundle. Only responsive metadata CSS was adapted for narrow screens, with original and published hashes in reel-presentation-receipt.json. Keep the HTML and .fexp together when downloading. The source files are separate.

## Limits

Analyser inspects; the native verifier executes. This unsigned bundle is an investigation with no embedded replay or source tree. Hashes detect changed members but do not authenticate origin or deployed-binary identity. The selected processor is a declared name; source/run linkage is unverified. Lifecycle clock reset is visible as a time-order warning; stream completeness is unknown. The prepared authoring trial is not human acceptance, a general success rate, a timing guarantee or universal correctness. No claim that one CLI turn equals one inference follows from the receipts.

## Public reel presentation

The published HTML now explains mixed fixture/lifecycle clocks instead of presenting them as elapsed runtime, distinguishes stepping through an investigation from executing replay, and records the screenshot review. A pilot enquiry link follows the saved investigation. Captured audit data, bundle members, images and stylesheet bytes are unchanged. `reel-presentation-receipt.json` records the input/output hashes and transformations; the prior export is retained privately.