layout.json (*.html.json) — audit verdict

1. Inventory — what exists

Backend (complete, tested, working):

Piece	Where	State
JSON → Jet converter	internal/layoutloader/json_layout.go (425 loc) + json_layout_test.go (617 loc)	done, well tested
.html.json load path	internal/noteloader/loader.go:202-225 (ext detection), internal/layoutloader/loader.go:91,215,426	done
Sync/push allowlist	internal/case/pushnotes/resolve.go:115-122; obsidian-sync isAlwaysPublishable (see docs/dev/local_design_iteration.md)	done
Block metadata for the editor palette	arg_type parsing in layoutloader/loader.go; GraphQL admin.layoutBlocks (internal/graph/schema.graphqls:1253-1281, resolver schema.resolvers.go:2061)	done, no consumer
JSON live-preview endpoint	POST /_system/layouts/render — internal/case/admin/renderlayoutpreview/	done, no callers; already marked "Remove (candidate)" in docs/dev/todo_remove_old_layout.md
doclint support	internal/doclint/fsenv.go:45-68	done
Original content round-trip for sync	internal/model/layout.go:17,45, internal/graph/note_path_content.go:32	done

Frontend (the actual product): nothing. LayoutBlock types appear only in generated GraphQL types (assets/ui/graphql/queries.ts:1414,2776); no component in assets/ui/ queries layoutBlocks or renders any editor. The drag-and-drop tree, property panel, palette, partial preview, and save mutation from docs/dev/json_layouts.md are all unchecked TODOs.

2. Usage reality

• Two .html.json files exist across all vaults, both demo fixtures: docs/demo/_layouts/json-test.html.json and json-include-missing.html.json (worktree/tmp copies of the same two).
• Every real layout is a plain Jet .html: docs/_layouts/mesh/* (the site chrome/landing), docs/_layouts/kanban.html, docs/demo/_layouts/* (showcase, yield-blocks demo, components), the kanban_template repo. No page anyone looks at depends on .html.json today.
• The block-assembly integration the owner remembers as "neat" is real — but it lives in the Jet .html layer ({{ block card(...) }} in docs/demo/_layouts/components/, custom/blocks.html, yield-blocks-demo.html). It survives a JSON cut untouched.
• Bug class: the live-pull layout-wipe data-loss hit _layouts/*.html.json files. Cutting the format removes that instance, though the sync empty-overwrite asymmetry itself is on the obsidian-sync side and should be fixed regardless.

3. Effort to finish the original vision

Remaining, per json_layouts.md status lists: /_system/layouts/render partial render (node_path), tree UI component, drag & drop, block palette from GraphQL, typed property panel, live preview wiring, save mutation, else branches for if/range. All frontend work in $mol, which has no existing drag-and-drop tree precedent. Realistic estimate: 10–20 person-days for a usable MVP editor, plus ongoing maintenance — for a feature with zero current demand and a product focus that has moved elsewhere (fleetbox).

And it would be built on the wrong schema: docs/dev/grid_layouts.md (the newer design) plans typed grid/flex/slot nodes that subsume magazine/wide/sidebars. An editor built now on the block-call-only JSON schema would need rework the moment grid layouts land.

4. Verdict: CUT

Why not KEEP/FINISH: the format's only value over .html Jet is machine-editability for a visual editor. No editor, no value — a human authoring .html.json by hand is strictly worse off than writing the Jet directly. Finishing costs 10–20 days against zero demand.

Why not FREEZE: freezing costs nothing today, but keeps a third template format threaded through eight subsystems (noteloader, layoutloader, pushnotes allowlist, GraphQL, doclint, obsidian-sync, writer protocol, docs) — every future change to any of them pays a small .html.json tax, and the docs keep promising an editor that will never come. For a solo-maintained codebase, surface area is the real cost.

Steelman for FREEZE (the one real tension): grid_layouts.md explicitly names JSON layouts as "the natural vehicle" for the grid DSL, and the converter is finished and tested. If the grid DSL is genuinely planned within ~6 months, FREEZE (delete only renderlayoutpreview + layoutBlocks, keep the converter) is defensible. But the grid design itself says "full plan before code," the node types (grid/flex/slot with track sizes and area matrices) share almost nothing with today's block-call schema beyond "it's JSON," and git history preserves the converter anyway. Betting kept code on an unplanned design is how zombie features accumulate.

5. What cutting removes, and migration

Delete list (roughly one day, mostly mechanical):

1. internal/layoutloader/json_layout.go + test; .html.json branches in layoutloader/loader.go (:91,215,426) and noteloader/loader.go (:202-203,225).
2. internal/case/admin/renderlayoutpreview/ + router regen (already slated by todo_remove_old_layout.md).
3. GraphQL: admin.layoutBlocks, LayoutBlock* types (schema.graphqls:1253-1281), resolver, arg_type parsing in layoutloader — unless the palette API is wanted for something else, nothing consumes it.
4. .html.json in pushnotes/resolve.go:115-122 allowlist and error message; obsidian-sync isAlwaysPublishable + extension lists; doclint/fsenv.go; note_path_content.go original-content branch (keep for .html if shared).
5. The two demo fixtures; docs: retire json_layouts.md (move to a "removed features" note), update writer_protocol.md, local_design_iteration.md, obsidian_bases.md mentions.

Migration for users: none needed — no known vault has a real .html.json. Safety valve: on push, reject .html.json with a clear "format removed, use Jet .html layouts" error rather than silently ignoring.

6. Relation to grid_layouts

Grid layouts is a different, better-scoped idea (page chrome as a configurable CSS grid) and does not depend on this format — it needs a schema designed around tracks/areas/slots. If/when it gets its full plan, start that schema clean; the JSON→Jet emitter pattern (the useful lesson from json_layout.go) is recoverable from git in minutes. Cutting now does not close that door.