Two kinds of done

Moving the project into a repository produced visible progress: structured records, exports, queues, scripts and a workbook. All useful. None a substitute for reading the work.

The next mistake was procedural. We aimed for a focused production sprint, then repeatedly worked in tiny research batches and stopped. Short completion messages made activity look more continuous than it was. Calling the queue “ready” did not process the queue.

That lesson belongs here too. A system designed to make evidence auditable must make its own progress auditable.

The two ledgers

1. System: records, states, matrices and exports
2. Evidence: inspected work, reasoning and limits
3. A working system can contain incomplete evidence

System completion means the records, fields, matrices and exports exist and reconcile. Research completion means substantive sources support the intellectual claims being made. One can advance much faster than the other.

Stage 4A names the minimum evidence packet: correct identity, current activity, a substantive primary artifact, a demonstrated capability, a relevant IoV question, reasoning, limits, confidence and an eligibility decision. A scoped review can check something useful without completing that packet.

Stage 4B asks for deeper work across three distinct artifacts and closer examination of contribution, criticism and limitations. A profile page is not the third artifact because the first two were difficult to obtain.

Automate what is deterministic

Scripts can reconcile IDs, preserve snapshots, generate exports and detect missing fields. They cannot make an uninspected paper inspected. The revised workflow uses those strengths without pretending they are interchangeable.

A completed batch must leave structured records and reconciled counts. A web search without a saved finding is an attempt, not a finished research packet. This is less glamorous than declaring a breakthrough every minute. It is also rather more likely to produce one.