# Current validation and limits Evidence is maintained as of 2026-10-28. CI pins compiler/core stable `0.10.16+b1a0e846e`: moon/moonrun `0.1.20261016`, moonc `v0.10.16+b1a0e846e`. This is the tested reproducible release, not a claim that every newer toolchain has been checked. Older development notes are preserved in [validation-history.md](validation-history.md); their pending statements refer to those earlier revisions. ## Observed evidence | Check | Observed result | | --- | --- | | Formatting, generated interfaces, strict compilation | Passed; four targets checked with warnings denied | | Library and executable documentation tests | 178 pass per target: Wasm, Wasm-GC, JavaScript, actual Linux native | | Complete file CLI workflow | Passes all four targets, including expected finding/incomplete exits | | Packaged dependency consumer | Four public-API tests and its executable pass per target after fresh ZIP extraction | | Operating systems | GitHub Actions Linux/macOS/Windows portable execution; Linux/macOS native execution | | PostgreSQL semantics | Actual PostgreSQL 18 CI probes pass; no local database execution claimed | | SARIF output | Ten generated documents validated against the pinned OASIS 2.1.0 schema, plus static GitHub profile and negative tests | | Windows native runtime | Not verified; local MSVC/Windows SDK unavailable | | Registry publication | Release preparation passed; upload and registry consumer verification pending | Public evidence before this documentation update: - [`ad5d77b`, run 40554777347](https://github.com/cn-xjr/MigrationLens/actions/runs/40554777347): all four jobs pass, including external schema/profile validation on three OSes. A subsequent verdict regression reproduced an earlier warning hiding a later non-conflict SQLSTATE. The terminal-error parser and crash/CRLF/ambiguous-output self-checks fix this without changing library analysis or archive contracts. - [`799203c`, run 40542741103](https://github.com/cn-xjr/MigrationLens/actions/runs/40542741103): v2 provenance/compatibility checks; all four CI jobs pass. - [`7a06245`, run 40545410582](https://github.com/cn-xjr/MigrationLens/actions/runs/40545410582): full CI artifact/gate workflow and external-application execution; all four jobs pass. - [`aee009e`, run 40548135559](https://github.com/cn-xjr/MigrationLens/actions/runs/40548135559): all four jobs pass, including 68 actual PostgreSQL 18 differential comparisons. - [`7f6c91c`, run 40349695555](https://github.com/cn-xjr/MigrationLens/actions/runs/40349695555): CR-comment correctness regression and expanded timeout tests; all four jobs pass. - [`d34af8b`, run 40352191148](https://github.com/cn-xjr/MigrationLens/actions/runs/40352191148): archive/independent-consumer workflow on supported runner targets; all four jobs pass. Those revisions had 163 library/doc tests; the current documentation split preserves all 16 detailed examples and adds two quickstart tests. Consult [Actions](https://github.com/cn-xjr/MigrationLens/actions/workflows/ci.yml) for the exact revision being evaluated. The local native runs execute built processes under WSL Ubuntu 22.04; a native compilation alone is not counted. macOS evidence comes from CI. The library has no FFI; the CLI uses moonbitlang/x filesystem/process adapters and core's strict UTF-8 decoder. ## Reproduce without a database Follow the [README quickstart](../README.md), then the complete command list in [maintaining.md](maintaining.md). `MIGRATIONLENS_MOON` selects the Moon executable used by validation scripts; scripts themselves run through the Wasm host. JavaScript requires Node.js; native requires a C compiler. Consumer extraction requires Git plus Windows tar or Linux/macOS unzip. ```text moon run cmd/main -- fixtures/postgres-migrations-concurrent.sql separate moon run scripts/e2e.mbtx wasm moon run scripts/consumer.mbtx wasm ``` All three return 0. E2e also runs real upstream fixtures and authored workflows that intentionally return CLI exit 1 or 2, then checks them as expected outcomes. It covers strict UTF-8 rejection, CR-only comments, source preservation, JSON round trips, deterministic rendering, saved single/batch reports, SARIF and baseline gates. Unit/property cases cover malformed input, Unicode ranges, size/depth boundaries, duplicate conservation, timeout limits, recovery and preservation of previous defaults. The independent consumer packages the library, checks required source/licenses and absence of caches, extracts it into a new workspace, and checks/tests/runs a separate module using public imports. The README mirror and the detailed-doc package are checked in that archive. It does not publish anything or require a registry account; declared dependency downloads may still occur. ## PostgreSQL 18 execution oracle Use a disposable PostgreSQL 18 database named `migrationlens_test`, with psql on PATH. Set PGHOST/PGPORT/PGUSER/PGPASSWORD normally, PGDATABASE to `migrationlens_test`, and MIGRATIONLENS_DISPOSABLE_DB to `1`. ```text moon run scripts/postgres_probe.mbtx ``` The probe refuses other database markers/major versions. It creates and drops only its own named test objects. A failure may leave those objects behind; recreate the disposable database before retrying. CI supplies a fresh PostgreSQL 18 container and checks the real process results, not just script compilation. The oracle checks SQLSTATE 25001 for concurrent-index transaction conflicts, 3B001 for missing savepoints, session/local timeout values across transaction and savepoint transitions, and explicit versus implicit dispatch behavior. It also checks lock modes, maintenance restrictions, staged constraint enforcement, referential actions and nullability against real server behavior. CR-terminated comments and vertical-tab separators have actual server probes. Each stage is named before invoking psql; a failure includes that stage, status and stderr. Connection environment/passwords are never printed. These are semantic probes, not workload benchmarks, exhaustive PostgreSQL compatibility tests or measured proof of scan avoidance. ## Evidence boundaries Seeded generator properties cover 256 seeds in both dispatch modes on each library target. Its verdict/replay self-checks run without a database. The [differential runner](differential.md) passed 68 actual server comparisons in the linked PostgreSQL 18 job. No local database execution is claimed. Native scenario-plan generation was also executed locally, alongside native unit tests. The [SARIF validation workflow](sarif-validation.md) verifies pinned schema bytes before use, checks ten old/new output samples, rejects malformed documents and cross-reference/count violations, and proves checksum refusal/recovery. It uses pinned Python tooling only for validation; the library gains no runtime dependency. Static GitHub profile checks do not establish a successful hosted ingestion. The complete `cmd/ci` workflow is covered on all supported targets, including finding/incomplete exits, preserved full reports, baseline acceptance/rejection, strict UTF-8 failure, fresh-output protection, determinism and recovery. The JavaScript test additionally launches the built CLI from a separate application's directory. The manifest parser and its existing tests are shared internally with cmd/batch; the old command's contract is unchanged. See [CI integration](ci.md). The v2 archive adds automatic project/tool/rule/context compatibility checks. V1 encoding/decoding and legacy comparison behavior are preserved. New tests cover mismatches, strict restoration, duplicate multiplicity, independent snapshots and invalid Unicode, including a downstream executable and four consumer tests. Read [the archive contract](archive.mbt.md) for compatibility. Source and resource licenses are recorded in [THIRD_PARTY.md](../THIRD_PARTY.md). The River production fixture deliberately returns incomplete for unsupported function/ALTER statements. It must not become a clean review merely because some index operations are covered. Fixtures are not automatically executed. `complete=true` covers only the documented rules/grammar. Catalog state, data, permissions and workload are outside offline evidence. No ecosystem uniqueness, performance advantage, broad SQL compatibility or actual downstream adoption is claimed. Unimplemented future capabilities remain in [design.md](design.md).