HardYards/SPRINT14.md

6.7 KiB

SPRINT 14 — THE YARD EDITOR: SITES ARE DATA, NOW MAKE THEM MAKEABLE

Sprint 13 ended with the game looking like itself: the garden matters, the night looks like a night, the front door onboards a stranger, and the audit tells the truth. 362/0/0, live at partly.party/hardyards. Sprint 14 builds the thing every future branch needs: an in-browser editor that turns "sites are data" into "John builds a yard the same day he builds its assets." Every piece already exists — site loading, raycast picking, validateSite, the audit and garden engines, the shared wind builder. This sprint is assembly, not invention. Read docs/MANUAL.md first if you haven't — it's new, and it's the cookbook this editor automates.

Baseline: selftest 362/0/0. Keep it green; mutation-check anything you'd call coverage. The editor is a TOOL with a UI — its internals don't need (dt,t) determinism, but its EXPORT does: same yard in, byte-identical JSON out (stable key order, pretty-printed), or site diffs become unreviewable.

Gate 1 — THE EDITOR CORE (A owns)

web/world/editor.html (+ js/editor.js) — it ships with the game (it's web/, it deploys; a public "build a yard" toy is a feature, not a leak — no secrets live in site JSON).

  1. Load: start from any existing site or an empty template (flat yard + fence ring + garden bed at origin). Import = the same loadSite/createWorld path the game uses — the editor renders the REAL dressed world, not a diagram.
  2. Place / drag / delete on the ground plane via raycast: posts, trees (species from the existing GLB set), structures (house, shed, carport), the garden bed, the gnome and props. Placement writes the SAME named-key site-JSON shapes the loader reads — no parallel schema. Drag = live re-dress of that entry (cheap: re-dress the one object, not the world).
  3. Export: canonical JSON to clipboard/download — stable key order, comments preserved via _design/_why fields the author fills in a side panel (the repo expects levels to carry their design notes; make the field visible, not buried).
  4. Validation is LIVE: validateSite runs on every edit; errors render in the panel, not the console. An invalid yard exports with a loud banner in the JSON ("_INVALID" key) so it can't sneak into data/sites clean.
  5. Round-trip assert (a.test): template → place minimal rig-able yard → export → loadSite the export → world builds, anchors adopt, enum passes. Editor bugs must not be able to produce a yard the game can't boot.
  6. Night hookup stays manual and documented: a site becomes playable when a week.js NIGHTS entry names it (MANUAL.md site cookbook step 6) — the editor's export panel SAYS this so nobody thinks export = shipped.

Gate 2 — THE LIVE SCORE (B + C; the editor tells you if the yard is a game)

  1. B — "SCORE IT": one button runs the audit gauntlet against the edited yard in-page: winnable lines at $80 (funnel ON, header says so), cheapest honest line, held-vs-bare garden separation (gardenfly), margin flags on knife-edge corners (the 15% rule). Results as a card, not console text. It's the same engines — site_audit/gardenfly already run in a browser; feed them the in-memory site object instead of a fetched JSON. Slow is fine (seconds, on demand); wrong is not: the score must go through windForSite() and the real commit→attach chain — the editor must never grow its own private harness (Sprint 13's whole lesson).
  2. C — wind authoring: place/rotate/drag the venturi (axis drawn as a LINE through the gap, both ends — it's mod π and the editor should show that), gain/radius sliders, tree-shelter cones rendered as translucent volumes. Plus a wind-field overlay: arrows/streaks sampling speedAt across the yard at a chosen storm+time, so an author SEES where the funnel screams before D has to feel it. Deterministic sampling, same seed path as the game.
  3. B + C shared assert: the editor's scored wind and the game's played wind are the same numbers — one probe point, one storm second, exact match pinned. Two harnesses, one number, by construction this time.

Gate 3 — THE PALETTE PROVES IT (E builds, D authors)

  1. E — the placeable set, audited: every GLB the editor offers must carry its extras honestly (rating_hint, collateral, collateral_key) — audit the current set, fix gaps in the factory, and add TWO new props built for temptation: pick from the trampoline (typed honestly in the enum this time), the kids' swing set (two honest posts + a crossbar that is NOT an anchor), or a second tree species with a different branch ladder. Each ships with baked reasoning, a wreck if priceable, and your negative-control assert discipline.
  2. D — author a yard COLD, editor only: no JSON hand-editing, no console. Build a complete third yard (your design instincts, any theme the palette supports), score it with B's button until it's winnable-but-mean, name it, export it, wire its night in a scratch branch, play it. File the entire experience in THREADS: every place the editor fought you is gate-1/2 feedback; every place it sang is proof. The yard you make is site_03 CANDIDATE — it ships only if gate 4 picks the more-sites branch, but its existence makes that branch nearly free.
  3. D — pool nit while you're in: the phantom-sail seam (A took the view half and filed it — re-test at gate-3 time; if A's fix landed, verify; if not, it's your repro that keeps it honest).

Gate 4 — JOHN: PLAY, THEN CHOOSE (eight sprints standing)

partly.party/hardyards — five nights, three sentences in THREADS. Then, with the editor in hand, the branch decision Sprint 12 deferred:

  • Reputation / warranty (DESIGN.md's consequence loop; the invoice exists to hang it off)
  • More sites (D's site_03 candidate is sitting right there; the editor makes a fourth and fifth cheap)
  • Heatwave days (the slow-pressure half of the design; sun-arc coverage, a different puzzle entirely) Nobody builds any of the three this sprint. The editor serves all of them.

Pool, spread by ownership: B's invoice line for the beyond-saving night (offer accepted S13); bench-vs-UI residual cause (open; margin rule stands in); bare-beds-win-nights-2-3 (gate-4 material — do NOT tune it this sprint); dev-line visibility default (A, one look).

gate 1: editor loads/places/drags/exports; round-trip pinned; live validation
gate 2: SCORE IT runs the real engines through the real chain; wind authoring
        visual; editor-vs-game wind pinned exact
gate 3: palette extras audited + two temptation props; D authors site_03
        candidate cold and files the experience
gate 4: John's three sentences + the branch, chosen in THREADS