# Example 03: reusable libraries checked together

This is a real current-workflow capture, not a before/after repair or a customer integration. Two synthetic libraries supply seven processing nodes; the GraphML also displays the OrderRequest event vertex (eight authored vertices, eighteen including framework scaffolding).

## What the records show

The native runner compiled the included generated processor against two separately built JARs and the pinned runtime. Seven scenarios pass: normal/equality accepted; over-limit and stock shortage rejected; simultaneous refusal uses declared limit-first priority; a deliberately unavailable risk check throws with no output; the next valid request succeeds. Six outcomes are recorded overall. The independent reviewer reran this on the same machine in a fresh output directory.

Released Analyser 1.40.0 actually read 12 records and found all 7 logged node IDs in the supplied graph. Record 3 accepts a value of 6000 pence against a 10000-pence limit. Record 4 rejects 12000 pence against that limit; record 5 refuses stock; record 6 checks refusal priority; record 7 accepts equality; record 8 logs risk UNAVAILABLE; and record 9 accepts the next request with outputCount=6. Missing audit output alone is not proof of non-execution: the native harness separately asserts the exception and unchanged count for the failed request.

The five-step walk explains accepted, over-limit, stock-refused, failed-risk and recovered requests. The HTML reel is a genuine Analyser export with a presentation-only responsive CSS adaptation. It includes embedded settled-step screenshots, title/context and a finish page linking the adjacent .fexp. It does not execute the application. It presents current expected results; no earlier defect or corrected revision was invented.

## How it was captured and checked

Codex used the isolated localhost REST action socket (the same supported verbs as MCP, not an MCP-transport capture). Inference was hosted by OpenAI; no local-model, human-approval or measured-savings claim. The published transcript contains actual tool requests/results, with local roots replaced by placeholders and unrelated context omitted, not private reasoning or credentials.

Jar: released Analyser 1.40.0, SHA-256 b886f2d4ab2475c68e588a06c8dd25ca9608428ae7356d232ce9b22a6a87c290.

Bundle identity: sha256:252f7c5fad7e7049f59aa3cb244a20c44ae4528c0039659e151649640cd64180.

The running Analyser exported the bundle using report.bundle. Its CLI verified 4 members. The contained log and graph match the canonical captured bytes exactly. Reopening in the same isolated app created a disposable workspace: 12 records, the new graph, saved 5-step walk and both final-step targets CURRENT. This is not a second-machine verification.

Open library-workflow-investigation.fexp in Analyser, then play “Reusable libraries checked together”. Keep library-workflow-reel.html and the .fexp beside each other when downloading so its relative link resolves. The source package is supplied separately.

## Limits

The bundle contains no source roots or replay payload. Flags do not travel; saved walk, narrative and notes preserve observations. Source/run relationship and the declared processor identity are not signed deployment provenance. Ordinary unsigned member hashes detect changes, not sender authenticity or universal correctness. Test clock switches to 1000 ms after init; the application truthfully shows out-of-order timestamps and unknown stream completeness. Use record indexes, not wall-time analysis. This synchronous synthetic exception test establishes neither transaction rollback nor general outage recovery, production readiness, arbitrary vendor compatibility or authoring-cost savings. The diagram's branches are dependency paths, not parallel threads.

The published reel only changes the metadata CSS to wrap long class names on narrow screens. The original export is retained privately; reel-presentation-receipt.json records both hashes and exact CSS substitutions. Screenshots, captions, recorded facts, bundle identity and link are unchanged.

## 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.