Directory location is the source of truth for lifecycle state

Context

After ADR-2 introduced three intent kinds, two field-vs-directory inconsistencies surfaced:

1. Disciplines initially carried status: active in frontmatter. But the disciplines/ directory already encoded that — there is no inactive discipline state. Field and directory said the same thing twice.
2. For standalone/bundle-member intents, status was historically a positional state in a sequence (draft / planned / shipped). For disciplines, status: active was a type predicate doubling for one of two directories (disciplines/ vs superseded/). Different jobs, same field name — a smell.
3. Superseded intents initially recorded only superseded_by: <itd-M>. But "superseded" means different things depending on what the intent was: a superseded standalone is a retired capability; a superseded discipline is a retired rule. Without recording the original kind, future readers (and the shape-classification auditor) had to infer it from context.

Two failure modes followed: field-vs-directory drift (an intent could in principle have status: active in the wrong directory and internal/core/lint had no clear contract to assert) and lossy supersession (kind context erased on retirement).

Decision

Directory location is the single source of truth for lifecycle state across all intent kinds.

• Standalone and bundle-member intents derive lifecycle state from drafts/ / planned/ / shipped/.
• Disciplines derive state from disciplines/ / superseded/.
• No intent carries a status: frontmatter field. Disciplines drop it entirely (it was redundant); standalone/bundle never had a status: field that disagreed with directory.
• Superseded intents preserve kind_at_supersession: <original-kind> so the shape the intent had when retired stays legible.

internal/core/lint enforces:

• No file in drafts/ / planned/ / shipped/ carries a kind: discipline value.
• No file in disciplines/ / superseded/ carries a status: field.
• Every file in superseded/ carries both superseded_by: and kind_at_supersession:.

Alternatives Considered

1. Keep status: field as the source of truth; make directory advisory. Rejected: filesystem moves are cheap, atomic, and self-documenting in git log; field edits are easy to forget and create silent drift. Directory-first matches the existing intent lifecycle (/abcd:intent plan already moved files between directories) — making the field authoritative would have required a second mechanism.
2. Keep both status: and directory; require lint to verify they agree. Rejected: redundancy is a smell; the lint rule would only ever catch discrepancies that a single source of truth prevents structurally. The two-source design encourages each writer to update one and forget the other.
3. Drop status: only from disciplines; keep it for standalone/bundle. Rejected: the principle is now non-uniform. Two kinds use directory-as-truth, one kind uses field-as-truth. Future readers and tooling must know which kind they're looking at to know which to trust. Uniformity is the gain; partial application loses it.

Consequences

Gains:

• One source of truth per concern. git log shows lifecycle transitions as file moves; no parallel field history to reconcile.
• The principle is uniform across kinds, applies consistently to future kinds, and generalises beyond intents (disembark snapshots, lifeboat artefacts, etc.).
• internal/core/lint contract is simpler (assert directory matches kind, not directory matches field matches kind).

Costs / obligations:

• Manual mv operations outside /abcd:intent verbs can put a file in the wrong directory; lint catches at pre-commit but the moment-of-error is the moment of move, not the moment of verb. Mitigated by /abcd:intent reclassify being the only sanctioned move path, with reclassification_history written in frontmatter.
• The status: field is reserved vocabulary for non-intent surfaces (e.g., RFCs still use status: open|resolved-yes|...). Naming registry must distinguish — clarified in 02-constraints/04-naming.md.

Downstream consequences:

• ADR-5 (brief is current state) extends the same principle to the brief itself: the live brief/ directory IS the brief; no parallel version: field, no parallel archive/<NN>/ directories duplicating git history.
• The issue ledger (spc-20, itd-4) applies the same pattern: the .abcd/work/issues/{open,resolved,wontfix}/iss-N-<slug>.md ledger carries no status: field; the sub-directory IS the lifecycle state. A wontfix_reason: or resolution: field is required in the corresponding directory (enforced by _issue_lib._validate_invariants, not by schema location alone), keeping per-state context legible.

Worked examples