Step-2 admission records: schemas now, command enforcement next cycle

Summary

spc-67 delivers itd-189's record shapes for the widening reading's step 2: every admission into the candidate set carries recorded grounds, every declined proposal carries a disposition, and a surprise is its own entry rather than a field on either.

This cycle delivers schemas, not commands. No reading has run, so there is nothing to write yet, and the intent scopes command enforcement to Iteration 2. That leaves an obvious trap: a schema no code reads is dead scaffolding. It is avoided by wiring the schemas to the gate that reads committed records rather than to the command that will one day write them. A hand-written admission record is validated by record_schema the moment it is committed, so make preflight exercises every shape this spec declares from the day it lands. The writing verb is the next cycle's; the refusal is this cycle's.

Scope

In: the admission-grounds record shape and its store; the statement that a declined proposal is spc-58's disposition record in its declined state and not a second record type; the surprise entry's store and keying; the record-lint wiring that makes all three refusable; the outstanding report's admission leg.

Out: everything under ## Out of scope.

Approach

Three shapes, two of which already exist

The reuse rule does most of the work here. Of the three records itd-189 names, only one is new.

The admission-grounds record is new. Family adm, minted through recordid.Minter.Mint like every other record family in this workstream (adr-45). It carries schema_version, id, run, proposal (the widening item's rdi-N), and grounds (free text, non-empty). It is filed at .abcd/work/issues/admissions/<run-id>/adm-N.md, bucketed by run id exactly as the reading store is, because an admission is meaningful only against the run whose proposals it admits.

The declined-proposal disposition is not new. Ruling (19) (2026-08-28) already reserves declined for the widening position, and spc-58 already delivers the disposition record, its disposition_grounds field and its state-availability check. A second declined-proposal record type would be a parallel store for a state the disposition vocabulary already holds. This spec therefore declares no disposition schema at all; it states that the declined proposal's record sits in dispositions/rdi-N/ with state: declined, and it adds the one thing that is genuinely missing: the report that notices a widening proposal which is neither admitted nor declined.

The surprise entry is one record, declared here. itd-180's own scope reserves the surprise entry's schema "in this family now", populated in Iteration 2, and itd-189 also lists it. The two intents describe one record. spc-58 reserves its shape, keyed by occasioned_by, and leaves it unpopulated; this spec declares the entry itself — its family, its store, its required keys and its allow-list (the schemas below). A surprise entry is srp-N, filed at .abcd/work/issues/surprises/, carrying occasioned_by naming whatever occasioned it (an rdi-N detection, an adm-N admission, a consequence). It is never a field on a disposition and never shares a key with one: the reading's output, the researcher's response and the surprise that occasions abduction are three acts and three records.

Wiring the schemas to the gate, not to a verb

internal/core/issueschema gains AdmissionRequired, AdmissionKnown, SurpriseRequired and SurpriseKnown, in the same schema-as-data idiom the issue record already uses: one definition, read by every gate that asks the question. internal/core/lint/schema.go gains two store entries, adm at .abcd/work/issues/admissions (bucketed by the run-id grammar spc-58 introduces) and srp at .abcd/work/issues/surprises (flat), each declaring its requiredFields from issueschema rather than from a hand-copied list.

That is the whole of the enforcement this cycle, and it is real: an admission record with a blank grounds is a blocker finding today, because checkRecordRequiredFields treats an absent value and an empty one alike. The [HAND] part is confined to who writes the file, not to whether anything checks it.

The outstanding report gains a leg, not a twin

spc-58 delivers reading_outstanding, a report-only rule pinned at info that names every reading item with no keyed disposition. The widening position's question is the same question with a wider answer set: a widening proposal is answered by an adm-N record or by a declined disposition, and one with neither is outstanding. That is one more branch inside the existing rule, not a second rule. Adding a parallel admission_outstanding would put the same judgement in two places, and the first divergence between them would be silent.

Its severity stays info and stays pinned in code for spc-58's reason: an unanswered proposal must never block an unrelated push.

Acceptance criteria mapping

Tests

Every case below is watched to fail before its change lands.

• internal/core/issueschema/admission_test.go : TestAdmissionRequiredIsTheOneList and TestSurpriseRequiredIsTheOneList assert the lint's required-field sets are the shared values, not restatements.
• internal/core/lint/schema_test.go : TestAdmissionRecordRequiresGrounds (blank and absent grounds both flagged), TestAdmissionRecordRequiresProposal, TestAdmissionStoreBucketsByRun (a run-id directory is a declared bucket; a directory that is not run-shaped is still reported), TestSurpriseRecordIsNotADisposition (a surprise id in the disposition store, and a disposition id in the surprise store, are each refused), TestSurpriseOccasionedByResolves (an occasioned_by naming no record in the corpus is a finding).
• internal/core/lint/reading_outstanding_test.go : TestWideningProposalWithoutAdmissionOrDeclineIsOutstanding, TestDeclinedDispositionSatisfiesTheAdmissionLeg, TestAdmissionRecordSatisfiesTheAdmissionLeg, TestAdmissionLegSeverityIsInfoNotBlocker.

The corpus fixtures are one well-formed admission record, one well-formed surprise entry and one declined disposition, each of which the gate must pass: a rule watched only failing is a rule that might refuse everything.

Out of scope

• The commands that write these records. Iteration 2, by itd-189's own words; this cycle's records are hand-written and gate-checked.
• The disposition record, its vocabulary and its state-availability rule, all of which are spc-58's.
• The surprise entry's schema declaration, which is spc-58's reservation; this spec states its keying and separateness only.
• The widening reading's definition, its candidate set and the criteria by which a proposal is characterised. Those belong to the definitions bundle and to the selection-criteria discipline, and the criteria never arrive at invocation.
• Populating any of these records. No reading has run in this cycle.
• Dispatching abcd <id> on adm-N or srp-N, which shares spc-58's residual: the cited-id grammar covers the four id-bearing families only.