docs(corpus): standard for status (#1323)

[serina-mcfall]

Summary

Adds launchpad/docs/corpus/standards/status.md, the canonical governance node for the corpus node schema's five-value status enum. It defines draft/active/deprecated semantics for the first time in this repository, and links to AGENTS.md's retirement procedure and ADR-0029's flagged-conflict escalation rather than duplicating either.

Related issue

Closes #1323

Issue type

Task

---

Agent provenance

Field	Value
Harness / provider	Claude Code
Model	claude-sonnet-5
Session reference	N/A
Initiating human	@serina-mcfall

Objective

Create launchpad/docs/corpus/standards/status.md as the single canonical policy node for the corpus status field.

Impacted components

• launchpad/docs/corpus/standards/status.md

Approach and rejected alternatives

Branched from origin/task/636-corpus-agents-md (not launchpad) per the batch dispatch instructions, because that branch carries an unmerged fix to launchpad/docs/corpus/AGENTS.md this task's governing procedure depends on; the diff below therefore legitimately includes those unmerged AGENTS.md changes.

For flagged and retired, the meaning is already fully owned elsewhere (ADR-0029 + node.schema.json's own description for flagged; AGENTS.md's "Retiring a node" section for retired) — this document names what they mean and links out rather than restating the procedures, per the corpus AGENTS.md's "authoritative sources this file duplicates none of" pattern already used by the two merged sibling standards (confidence.md, decision-references.md).

For deprecated, I found launchpad-26/buzz#1311 ("task: document corpus standard for deprecation", parented to the same PRD #605, targeting launchpad/docs/corpus/standards/deprecation.md) already open and unlanded. Rejected alternative: writing the full deprecation policy into this node. That would violate the one-idea-per-node rule and directly duplicate #1311's assigned scope, so this document instead names deprecated's place in the status lifecycle only, defers the procedure to #1311, and adds a MUST (requirement 6) bridging the interim gap: until #1311 lands, an author moving a node to deprecated must record what replaces it in the node's own body.

For draft and active, no document in this repository defines operational semantics beyond the bare enum member names (verified: node.schema.json's status description discusses only flagged; PRD #602, issue #622, and the schema's own plan document were all checked and none define the other four values). This document supplies that meaning as original guidance and says so explicitly in "Scope and authority", rather than presenting an invented convention as if it were restated from an existing authority — mirroring how corpus-standard-confidence disclosed its own invented confidence bands.

Rejected alternative: leaving relationships empty by default. Per the batch dispatch's corrected guidance, checked git ls-tree -r --name-only origin/launchpad -- launchpad/docs/corpus immediately before finalizing front matter (re-run after all body edits) and found four loaded nodes: corpus-agents, corpus-readme, corpus-standard-confidence, corpus-standard-decision-references. Added two references edges — to corpus-agents (this document leans on its "Retiring a node" section rather than restating it) and to corpus-standard-confidence (the flagged-vs-low-confidence distinction this document's MUST 2 states is the same distinction that node already makes on the confidence side). Did not add an edge to corpus-standard-decision-references, despite citing it repeatedly in prose, because nothing in this node depends on it being true or supersedes it — a bare references edge would say less than the existing prose cross-references already do. Did not target any of the other 4 documents in this same 5-way batch (generated-content, identifiers, naming, normative-language), per the dispatch instruction that none of them will have landed on launchpad by the time this PR does, regardless of what any single worktree shows.

Verification

Command run:

python3 launchpad/project-intelligence/corpus/validate.py

Raw output:

UNVERIFIED  corpus-agents: evidence entry 1, citation 1: is a commit reference, which names no openable file
UNVERIFIED  corpus-agents: evidence entry 20, citation 1: is a graph-edge or tool-result citation, which names no openable file
UNVERIFIED  corpus-readme: evidence entry 1, citation 1: is a commit reference, which names no openable file
UNVERIFIED  corpus-standard-confidence: evidence entry 1, citation 1: is a commit reference, which names no openable file
UNVERIFIED  corpus-standard-decision-references: evidence entry 1, citation 1: is a commit reference, which names no openable file
UNVERIFIED  corpus-standard-status: evidence entry 1, citation 1: is a commit reference, which names no openable file
PASS  corpus validation found no errors; 6 item(s) reported unverified

Exit status: 0 (confirmed via separate re-run; the UNVERIFIED lines are non-fatal notices for commit citations, expected and unavoidable per AGENTS.md's own documented behavior — every existing node's provenance entry produces the same notice).

Also ran, because this worktree's local verify-gate pre-commit hook does not recognize python3 launchpad/project-intelligence/corpus/validate.py as a matched test command (a known, already-tracked gap in that hook's TEST_PATTERN, unrelated to this repository) and a piped invocation is explicitly refused by that hook as unattributable, so the commit's verification stamp was earned by this separate, genuinely-run, unpiped command rather than by validate.py itself:

python3 -m unittest discover -s launchpad/docs/corpus/schema/tests -p "test_*.py" -v

Raw output:

Ran 27 tests in 0.035s

OK

• Tests or checks were run and the raw output is pasted above
• The diff is confined to the scope of the linked issue
• No secrets, keys, tokens or hostnames were added to tracked files

Not verified

• No node anywhere in this corpus has ever carried deprecated, retired or flagged, so every claim in the new document about how those values behave is drawn from the schema, AGENTS.md and ADR-0029, never from a worked instance — noted in the node's own "Expected but not verified" section.
• Whether any generated view (index, knowledge-crate projection) treats the five status values differently (e.g. excluding retired/deprecated from a default listing) was not checked — no generator exists yet for this corpus.
• Whether reversing a status in place (e.g. retired back to active on the same file) is intended to be possible was not settled by asking anyone; it is named as an enforcement gap in the new document because nothing in node.schema.json or validate.py restricts it, not because reversal is expected to happen.
• Did not run the full just ci gate (out of scope for a docs-only corpus node per this repo's own guidance, and this fork's AGENTS.md governs docs/deployment work rather than Rust/desktop/mobile changes this PR does not touch).

Security implications

None believed. This is an addition to a public, docs-only corpus node under launchpad/docs/corpus/; it adds no secrets, no executable code, and no new HTTP/API surface, and does not alter runtime product behavior.

Escalations

• Filed launchpad-26/buzz#1516 ("task: decide whether corpus node status transitions should be restricted (e.g. one-way retirement)") — a gap found while executing this task, not this node's to fix: nothing in the schema or validator prevents moving a node's status backward (e.g. retired -> active) on the same file, and no document in the repository states whether that is intended. Out of scope for this PR per its own "creating/editing a second hand-authored canonical corpus document" boundary and per AGENTS.md's "stable knowledge belongs in a document, active work becomes an issue" rule — this is unsettled, not yet documented, and therefore a task, not something to resolve inside this node's body.
• The deprecated status's full policy is intentionally left to #1311, already filed and open; this PR does not duplicate or preempt it.
• Whether the corpus-standard-decision-references cross-references in this document's prose should eventually become a typed references edge (once this node's claims start depending on that document being true, rather than merely citing it) is left as a judgment call for whoever authors the next sibling standard, per AGENTS.md's "the first sibling node is the moment to revisit it" guidance.

[tucktuck101]

Review — standard for status (#1323)

The call that mattered most here is right: flagged is defined as ADR-0029 defines it — "two
same-claim-type authoritative sources contradict each other and a human has not resolved it.
Not a lifecycle stage and not a confidence marker" (status.md:153 table row, and the FACT at
:22). That is the definition that could have collided with confidence.md and it does not.
The status vocabulary is the exact five-value enum, no additions. Two accuracy defects.

High — retired is asserted terminal, sourced to a file that does not say so, and contradicted twice in this same document

The value table's retired row (status.md:153) reads "The terminal lifecycle state. ...
the id is spent permanently" and names launchpad/docs/corpus/AGENTS.md in its "Who defines
it" column.

AGENTS.md does not define that. Grepping the merged corpus AGENTS.md for
terminal|one-way|irreversib|stays spent returns exactly one line — :413, "Never reuse or
rename the id. A retired id stays spent" — which is a claim about the id, not about the
status being terminal.

And this document disclaims terminality twice elsewhere:

• :254 (enforcement table): "a node can be moved from retired back to active on the same
  file without any check objecting"
• :307 (scope and omissions): "Whether a status transition should ever be restricted (for
  example, one-way retirement) | Not decided anywhere in this repository today"

So the normative meaning table asserts terminality, attributes it to a file that doesn't assert
it, and the same document denies it is settled. The rest of :113-144 is scrupulous about
separating derived authority from original judgement — this row is the one slip. Either drop
"terminal" and keep the AGENTS.md attribution for the file-stays/id-spent half, or keep it and
change the definer column to "This document (guidance, not schema-enforced)" as the draft and
active rows already do.

Medium — the H1 ships the exact form being reconciled away

status.md:83 is # Standard: \status`— topic in a code span. #1508, in this same batch, is landing precisely that de-code-formatting change toconfidence.md(verified in its diff:# Standard: `confidence`-># Standard: confidence`). A newly authored standard shipping
the form that is actively being removed will need the identical follow-up edit immediately after
merge. One pair of backticks.

Low — one FACT overclaims exhaustiveness about validate.py

The evidence entry at :40 ends: "The only occurrences of the string "status" in the file
belong to an unrelated per-citation ValidationReport.status field". Grepping
launchpad/project-intelligence/corpus/validate.py for status returns four hits: :592 (the
field), :790 and :796 (verdict.status, i.e. that same field), and :11 — a module
docstring line "Exit status is 1", which belongs to neither. The load-bearing half of the claim
— that no logic reads or branches on a node's status — is true, and I checked all four hits to
confirm it. Only the "only occurrences" clause is false, and it is declared FACT. Narrow it to
"the only occurrences that refer to a node's status".

Batch correction — the AGENTS.md hunk is not an index registration

Worth recording because it looks like one: launchpad/docs/corpus/AGENTS.md at this PR's head
is byte-identical to the launchpad tip (verified with git diff --quiet), and the same is
true across the sibling standards PRs. The hunk is the whole task/636 change inherited through a
stale merge base (919886b41), not a registration this PR adds. Merging produces no net
AGENTS.md change and there is no cross-PR conflict there. There is also no standards index to
register in — the corpus AGENTS.md forbids a hand-authored index.

What is correct

• flagged matches ADR-0029 and node.schema.json exactly, and the standard explicitly
  forbids the low-confidence reading — no collision with confidence.md.
• Status vocabulary is the exact five-value closed enum.
• Honest about what is not settled: the omissions table records the transition-ordering gap
  rather than inventing a rule.
• Node validates clean.
• CI green at head (latest run per check).

Reviewed by tucktuck101's review lane. Every claim above was checked against the files at this
PR head and the merged corpus on launchpad.