Prove standalone recording with an unmodified third-party library

[dprada]

Purpose

Define and prove a minimal standalone Recorda substrate that works with a library outside MOLI, including a library whose source cannot be edited. Keep component-owned scientific results and existing provenance authoritative. MOLI integration follows only after the standalone behavior is demonstrated.

First experiment

• Use a small, deterministic third-party library operation through an explicit boundary owned by the caller. Starting a session alone must not claim to capture uninstrumented calls.
• Exercise an opt-in decorator on a separate example to check the inactive path and nested correlation.
• Persist an operation start before execution and a terminal outcome afterward. An interrupted run must remain visibly incomplete.
• Record implementation identity, safe input/output descriptors, status, timestamps, and explicit omissions. Do not serialize opaque objects, secrets, or large payloads by default.
• Inspect the resulting record using Recorda alone; no MOLI dependency or platform context.

Acceptance evidence

Show success, exception, interruption, inactive behavior, nested operations, and a sensitive input. Demonstrate that an existing native record can be linked by stable reference without copying or replacing it. State which calls were instrumented and which were outside coverage.

Follow-on

After this experiment, test one Sabueso retrieval and one entity-resolution boundary. Evaluate MolSysSuite's native records before designing any integration or extra recording there. Then test MOLI context, EventLedger, and composed ProjectRecord as a separate step.

Design note: devguide/FIRST_SLICE.md. This issue concerns Recorda's standalone behavior; any later change to the shared MOLI provenance contract belongs in uibcdf/moli.

[dprada]

Architecture review of the first standalone slice: three acceptance details should be settled before implementing the wrapper.

1. Keep the scientific purpose/type of an operation separate from the callable's package/module/function identity. A record named only external_library.analyze can become a call log without saying which scientific action was declared. The first fixture should inspect both fields, even if the semantic vocabulary remains provisional.
2. Specify behavior when the persisted started entry cannot be written before calling the target, and when the terminal write fails after the target returns. For this small prototype, a strict fail-before-call rule is a reasonable default for the first case. The second must never produce a record presented as complete; preserve the target outcome and expose the provenance failure. This is the local version of MOLI's recording-reliability boundary; a broader strict/buffered policy can follow evidence.
3. Include one physical-quantity input or output in the acceptance fixture. If Recorda persists a quantity, value, unit, and scientific meaning must travel together, or the value should be omitted with an explicit reason. This checks compatibility with MOLI's quantity integrity policy without making Recorda depend on MOLI.

These checks keep the first experiment small while testing whether its record is scientifically interpretable and honest under failure.