11 — Build order
Summary
The v3 order, chosen by two rules: the scarce components are editorial (the maps' finding), so writing them is the critical path; and every new screen gets a synthetic-reader round as a fixed mockup before implementation, because a defect found there costs a file edit rather than a refactor. Phase 1 pays the library's debts (per-node dates, the five scenarios, the entry checker as a script). Phase 2 writes and tabletop-tests level one before building it, and publishes the runs including abandonments. Phase 3 ships fragment sharing. Phase 4 ships the three sets on the visitor's claims, with the transcript import waiting for its decision rather than being prototyped around. Phase 5 — only after synthetic rounds cleared the levels — spends the few dozen real readers on visual variants first, qualitatively.
Key concepts
- Tabletop before build — the programme's highest-value use, scheduled rather than admired
- Phase 4's gate — the one feature that could breach P1 by accident waits for its decision
- The standing checks — same-origin, model suite, the no-free-text grep, the wording rules — every phase
Key ideas
- A published tabletop against an unbuilt page is the estate's publish-before-build move one layer deeper.
- The persona-simulation service is unscheduled because building it before a calibration record exists is building the unfalsifiable version.
- Phase 2 is done when an abandonment produced a change and the changed page's re-run is published beside it.