plan

Hosting migration catalog digit-suffix correction

Authorized corrective bug increment, root session Codex-20261027T100516Z-plugin-session, request req-20261031T171140Z-cap10-release; own session Codex-20261031T150618Z-plugin-session, request req-20261031T185210Z-hosting-catalog-digits-fix. This is not a new feature, hosted apply, HV, or requirement satisfaction.

Requirement and observed defect

Current authoritative layer1 TR-SCHEMA-PROVISIONING-023#TR-23-C02 and TEST-SCHEMAPROVISIONING-123#TR-23-C02 require proposing every later core extension in order after core001..011, hosted scheduler and private realtime, rejecting changed, out-of-order and unknown longer history before apply. Companion C01 requires repeatable credential-redacted version-controlled provisioning. The fresh selected records are in requirements.json; criteria remain unsatisfied.

The current filename filter in scripts/HostingSchemaPlanner.psm1:9 accepts letters/underscores in suffixes and excludes the digit in existing 014_capacity10.sql. The catalog has15 records instead of16 and would omit that forward safeguard from status/apply. Local integration and browser harnesses load all *.sql, so their passing SQL cannot establish the hosted planner includes014. Catalog declared version labels are not hosted apply versions: apply generates fresh timestamp+index. No version-generation change is required.

Locked scope and contracts

Preserve all six existing history-case objectives; independently extend their catalog fixture through013 and014. Add two catalog tests: exact ordered names and bodies including014; invalid filename exclusions. A disposable Pester TestDrive holds literal fixture files, deliberately created out of order. Expected values are literal test-fixture tables independent of production enumeration and regex. Invalid filenames with wrong prefix width, missing/empty suffix, punctuation, trailing extension and hyphen remain excluded. Existing case-insensitive filename matching is preserved, not silently tightened. A nested valid-looking file is excluded because discovery is nonrecursive.

The correct mock uses the literal expected filename list rather than the production regex. Existing wrong pending and empty providers remain; an omission control returns an otherwise correct catalog minus014, and an invalid/order provider returns incorrect catalog entries. The same eight tests, inputs and assertions bind to mocks and production. Capture the all-green mock harness with named rejected controls before editing the production module. Expected control failures are retained native sensitivity evidence, not a failed acceptance suite.

The sibling tests/integration/hosting-schema.test.ts also pinned the omitted catalog:14 pending after001, ending013, and only012/013 after the hosted prefix. The hosting agent owns its three expectation changes to15 pending, ending014, and012/013/014. Its same six cases must discriminate an independent real-filename16-entry mock and named wrong providers before the production planner changes. Root explicitly requires both mock gates before the filter edit. That agent owns immediate TypeScript validation and real six-case binding; this agent does not edit its file or run overlapping tests.

Only after that gate, change the production suffix class to permit digits, preserving the three-digit prefix, full anchoring, sorting, hosted insertion point, bodies, history drift rejection and grants/SQL behavior. Existing001..014 SQL and hosting SQL files remain byte-identical. No provider, credential, configuration, application/domain/UI, test timeout, source HEAD or Git writes.

Validation and propagation

Immediately parser-check and PSSA Warning/Error-check each edited PowerShell file. First run the isolated mock gate with WhatIf before actual. Retain before-file copies/hashes and chronology. After green mocks, edit only the planner filter, validate it, run the same eight production-bound cases and the affected complete PowerShell evidence wrapper. Propagate explicit planner counts6 to8 in mock/real gates, planner producer, PowerShell wrapper and review preparer. Existing script-contract3, docs22 and provenance9 cases remain: wrapper total50 including8mock+8real. Existing review-preparation22 assertions remain unchanged unless a concrete new assertion is required; their fixture does not encode planner report totals.

Run a fresh safe actual repository catalog read, verify all16 names, normalized body/order and only012/013/014 pending against an independent synthetic thirteen-row history. This is local evidence only. Root owns the new source freeze, full source suites, review and hosted strict-TLS refresh. Historical frozen receipts and generated wiki exports remain unchanged. Sweep current source/tests/scripts plus retained historical producers by path; ignored generated wiki is an excluded generated projection, not a gate or source consumer.

Current native PowerShell mock chronology: first gate13:58:50..54, then a stronger correct-count16 invalid-name replacement control13:59:49..53. Both have8 correct cases,17 expected wrong rejections and zero skips; module SHA F002924A2F673F39A4B3260965D32486988EA71EE32F8402C27BA74725773DA0 was unchanged. Preserve both receipts. The second gate is mock-first-v2/mock-gate.json, SHA AD55ED0896E844A4A193AD8B3E0A9BB794AE26B35DF86E53941FD9F77F48AE5A. Source and test preparation is not exact-commit acceptance; subsequent root-owned frozen gates remain required.

Failure and exit handling

Any mock mismatch stops production edits. Any real mismatch retains native evidence and is corrected without weakening expectations. Report exact failures, skips, missing checks and source diffs. Preserve original native reports and prior receipts. This bounded correction ends with exact mock/real/affected-suite results and source hashes for root review; it does not mark the release, plan, requirements or HV complete.