The Consolidation Program

One engine. The v2 family (crates/n0, crates/n0-model, crates/n0_dev) is the chassis; the legacy family (crates/grida and its satellites) is the know-how. The program ends when the chassis carries all the know-how, the conformance bar has flipped, and the legacy engine has contracted to nothing.

Status: Active. Owner: universe@grida.co. Tracking: gridaco/nothing#43 (umbrella), gridaco/nothing#9 (the v2 model program), gridaco/nothing#1 (the migration anchor).

Genre: program record — doctrine, destination, and method for the people and agents doing the work. Not a domain spec. The domain specs live in the sibling wg clusters (canvas/, format/, feat-*) and always win on conflict; a conflict with one of them is a finding to file, never something to paper over here.

Read in this order

A zero-context session becomes a working session by reading four short docs, in order:

1. The goal of n0 — what the consolidated engine becomes, and what it refuses to be.
2. End-state topology — chassis vs know-how: who absorbs whom, per domain, and the render-mode taxonomy.
3. The method — how one consolidation step is executed, with a worked example.
4. The charter — the ratified route: phases with entry/exit gates, the first PRs, the owner decision registry, and the known unknowns.

Amendment (ratified 2026-07-21): the Web-First Amendment revises the topology to one engine kernel, source-native semantic models, one provisional resolved render contract, many hosts, and moves HTML/CSS + SVG to the front. Read it with the goal and topology — it supersedes their "one model" reading (the engine kernel converges downstream, not at one authored model) and reorders the charter's phases where they conflict. It leaves every standing rule below intact.

Reference, consulted as needed: the glossary — the program vocabulary. Use its terms; do not mint synonyms.

Decision proposal, not yet ratified: the conformance-bar flip rule — the pre-score threshold, coverage, and oracle-discipline proposal tracked by gridaco/nothing#49.

Findings (grounded evidence for open decisions): SVG paint in the shared cascade — why SVG fill/stroke cannot come from the shared cascade under the compiled Stylo engine, and the options that gates.

The charter is the only doc of the four that sequences work — and it records the route; the current position (active phase, landed PRs, taken decisions) is tracked on gridaco/nothing#43, which a session reads before starting work. The other three docs are direction: stable in shape, expected to gain detail as decisions in the registry land.

Standing rules (bind every session)

Stated once, owned by the charter's doctrine section:

• Patrol-before-drop. No deletion or replacement without a patrol triage and a captured-essence ledger first.
• Migration by extraction. A module becomes a shared crate when its second consumer appears in the workspace — never before.
• Absorption direction. The chassis absorbs the engine role by consuming extracted crates — never by copying; the legacy engine is decomposed, not absorbed.
• Two gate classes. Zero-behavior moves gate on byte-identical sweeps; capability grants gate on the scoreboard. Nothing lands unmeasured.
• Oracle discipline. The conformance bar is the Chromium/consensus oracle — never the legacy engine itself.
• Frozen surfaces. The v1 schema (format/grida.fbs) and the published wasm freeze contract stay untouched until Phase 6 says otherwise.
• Owner gates. Merges to main and every registry decision need the owner's explicit GO. All-green CI is necessary, not sufficient.

Working conventions

• Plans are *.plan.md files — gitignored working scratch, never committed knowledge. Durable knowledge lands in this directory or the sibling wg clusters; work items land in issues.
• Skills to load per task: docs-wg and naming before authoring or renaming anything; links for cross-repo references; fixtures, render-reftest, render-perf, io-svg, io-grida for their engine loops.
• Concluded programs are archived under repo-root archive/ — the frozen v2 workbench record at archive/model-v2/ is the precedent, and this directory retires the same way when the program concludes.