irrevon
Record
ADR-0000
Status
accepted — superseded by ADR-0026 (sanitization supersession, decision content unchanged; ADR-0026 countersign pending)
Date
2026-07-20
Supersedes
—

Freeze project scope — benchmark-first, C2-scoped, with binding non-goals

Context

The master doc (../master-doc.md) — the human’s own authoritative document — already froze the technical question (§1.2), declared the non-goals binding (§5.4), and fixed the benchmark-first, C2-scoped framing after adversarial revalidation (§1.1, §1.3, ADR-009). M1 requires this freeze to be recorded as ADR-000. This ADR is a faithful record of that existing decision, not a new one; it is recorded accepted on the master doc’s authority, with an explicit human countersign requested in ../review-queue.md section 3.

Decision

The frozen scope is, verbatim by reference:

  • Technical question: master doc §1.2 — duplicate/orphaned/lost effects under ambiguous irreversible actions, and how much a deterministic reconciler eliminates or surfaces without the agent’s cooperation.
  • Framing: benchmark-first; the reference engine is a baseline within the benchmark, not a product (§1.1, ADR-009). Scoped to C2; C1 null pre-committed; C3 demonstrated as an impossibility boundary.
  • Non-goals (binding, §5.4): not a durable-execution runtime, auth layer, approval gateway, dashboard, or hosted SaaS; no billing; no universal exactly-once claim; compensation is not rollback; the LLM never holds sole authority over an irreversible action; not a payments/clearing protocol; no revenue-bearing activity until written immigration guidance permits it.

This freeze deliberately does not freeze open implementation choices: language (ADR-0013), C2 sandbox (ADR-0012), Stripe version (ADR-0010), statistics margins (preregistration §5), licensing (ADR-0014).

Alternatives

  • Leave the freeze implicit in the master doc — rejected: M1 explicitly requires ADR-000, and the freeze needs a countersignable artifact.
  • Re-litigate scope at scaffold time — rejected: the master doc survived adversarial revalidation; reopening without new evidence burns the ~10 hrs/week on relitigation.

Consequences

Agents may not add product requirements or expand scope; anything scope-shaped goes to the review queue. Tasks citing this ADR can rely on §5.4 as a hard boundary.

Risks

A frozen scope can ossify against genuinely new evidence. Mitigated by the reopen trigger below and the quarterly competitor review (§14.2, O-3).

Reopen trigger

An adopted competitor covers the C2 reconcile-by-query wedge (master doc §4.3 kill rule), or the pre-committed falsification criterion fires at M7 (§8.6 → §14.3 reframe).