THE RECORDS · No. I · VOCABULARY

THE LEXICON

Every trade develops a shorthand. This is the Bureau's: the labels that number a spec's claims, the bracketed markers that stop a run, and the verdict tokens agents use to pass or refuse each other's work.

A STANDING RECORD  —  REVISED 2026 · VII · 5

The scheme exists for one reason: so that anything downstream can cite an item without re-quoting it. A build prompt says it satisfies FR-7; a run pauses "blocked on OQ-3"; a critic checks AC-11 by number. A label is a type prefix plus a sequence number, numbering restarts per section within each spec, and the reference stays exact no matter how many hands the work passes through.

The deliverable labels

FR-N

Functional Requirement

One numbered claim about what the thing must do. Build prompts cite the FR they satisfy, so no line of work exists without a requirement to answer for it.

AC-N

Acceptance Criteria

The checkable conditions of done. The Challenger verifies these by number — not "looks good," but "AC 11 holds, AC 12 does not."

EC-N

Edge Case & Failure Mode

What happens at the margins, named before the build instead of discovered after it. Bureau-specific vocabulary — you won't find EC in a standard PM handbook.

OQ-N

Open Question

A decision that needs a human, or research that hasn't happened. Work can be formally blocked on an OQ by name, which is the point — the unknown gets a handle instead of a shrug.

AS-N

Assumption

A thing taken as true, declared and numbered so it can be revisited when it stops being true.

ARCH-N

Architecture item

A design element or risk in the Architect's section of the spec.

ADR-NNN

Architecture Decision Record

A decision plus its rationale and the alternatives it beat — three digits, built to outlive the person who made it.

API-1 …

Domain tags

Scoped requirement ids inside one spec — API-1, DB-2, AUTH-1, UI-3. Local shorthand, never framework-wide.

Just as telling is what the Bureau never writes. The standard PM alphabet — NFR, REQ, US, UC, TASK, TC, BUG, RISK — does not appear. The concepts exist; they simply never earn an XX-N code. And a checkpoint is never CP-1 — checkpoints are markers, not labels, which is the next section.

The checkpoint markers

A checkpoint is a bracketed literal in the run narrative. When one appears, the run stops and waits for a human:

[CHECKPOINT]                   generic stop — a decision belongs to a human
[DESIGN-MODEL CHECKPOINT]      the design's shape needs a human eye
[DESIGN HANDOFF]               design leaves the workshop for review
[DEV-VERIFIED CHECKPOINT]      dev build green — nothing ships without a word
[EXTERNAL-ACTION CHECKPOINT]   something irreversible wants to fire

Clearing one requires the literal word go. Silence does not count. "Continue" does not count. "Looks good" does not count. The rule is blunt on purpose: a system that proceeds on ambiguity is a system that ships on ambiguity.

The verdict tokens

Each reviewing role has a fixed vocabulary for passing or stopping work — small on purpose, so a verdict can be adjudicated mechanically:

The Challenger (critic)BLOCKER · WARNING · CHECKPOINT

A blocker routes a fix; a warning is logged and survived; a checkpoint summons a human. The Challenger rates — it never decides. Its verdict block closes go / block / accept.

The Delegate (automated gate)proceed · revise · escalate

Exactly three outcomes, schema-checked. A missing or malformed verdict escalates — it never auto-proceeds. The full argument is the dispatch Gate the Flow, Not the Judgment.

The Witness (observer)go · block

Reads across runs and distinguishes blocked — needs a human — from merely in progress.

The Notary (external attestor)FLAGGED

Advisory only. A FLAG means the cold-review seal was broken — the findings are discarded and the review re-issued, because a warm review is worth nothing.

The states of a run

A phase is pending, in_progress, complete, or blocked. The run as a whole derives from its phases: not_started → in_progress → complete → archived, with blocked outranking everything. A run untouched for four hours is stale, and the Ministry of Flow says so out loud. How these live in the state file is Record No. IV — Anatomy of a Run.

The cast that speaks this vocabulary — who rates, who decides, who fetches — is introduced on the main floor, and their model assignments are Record No. II — Cast & Routing.

END OF RECORD No. I