# Validation evidence

2026-10-31, uniqueness-aware review: Windows ran 80 core test blocks on all four
targets, four additional Wasm CLI tests (84 total there), and six JavaScript
bridge blocks (86 total there). All passed. Twelve new core blocks include a
64-case exact-value matrix, compound tuple collision safety, empty/space/zero/
Unicode preservation, input-versus-merge violations, input symmetry, deterministic
repair order and bytes, invalid/key/unrelated edits, new duplicate-pair refusal,
partial repair/replacement/revocation, ordinary-decision dependency guards,
strict/duplicate/versioned JSON, defensive ownership, declared header/row lineage,
64/65 rules, text and complete-evidence overflow refusal/recovery. Existing ordinary
APIs and legacy wire regression tests passed unchanged.

The full pinned verification ran all fifteen Wasm workflow helpers, six packaged
external-consumer tests plus the catalog app on every target, formatting, generated
interfaces and strict all-target checks. `moon doc` also passed. The new helper
actually executed the release native CLI on Ubuntu 22.04/WSL, checking simulated
identifier edits on all 19 columns of the licensed OurAirports rows, complete-only
file export, exact CSV/TSV replay, full audit, revocation, invalid-source and stale
refusal/recovery. This is runtime evidence; Windows native async CLI stays excluded.

Actual offline desktop Chrome QA passed the new uniqueness workflow and every
existing browser workflow: full group evidence, explicit field repair, exact CSV/
TSV and audit equality with the release CLI, saved-envelope reopen, revocation and
whole-record deletion. No HTTP requests/console errors were observed. Browser
limits remain explicit; no other browser-engine compatibility is claimed. CI
includes the new helper/consumer/bridge tests and Linux UI gate; assess its own
head. No Mooncakes publication or registry consumption has been performed.

2026-10-31, offline browser review: Windows actually ran the existing 68 core
tests on Wasm/Wasm-GC/JavaScript/native, four additional Wasm CLI tests (72 total
there), five new JavaScript bridge test blocks (73 total there), fourteen Wasm
end-to-end helpers and the five-test packaged external consumer on each of the
four targets. All passed. Bridge tests cover exact 20,000-row/200,000-cell limits,
overflow refusal/retry, strict/deep/malformed input, deterministic serialization,
compound keys/quoted TSV/Unicode, empty values, two review phases, explicit
dependent revocation, audit v2 and stale snapshot binding. The actual offline
Chrome file-URL check exercised uploads, decision application, downloads,
bundle reopen/replay, exact CSV/TSV equality against the release Wasm CLI,
before/after audit equality, stale/invalid JSON/invalid UTF-8 refusal, cancellation,
state recovery, 13-conflict pagination and cross-page drafts. Desktop and 390px
responsive screenshots were inspected; no HTTP requests or console errors were
observed. This establishes desktop Chrome evidence, not Firefox/Safari or actual
mobile-device compatibility. Format/interfaces/strict all-target checks and
self-contained HTML generation passed. Root public signatures are unchanged;
the JS-only bridge has a compiler-generated internal transport interface.
Linux UI QA plus three-OS bridge/build and existing backend checks are configured
in CI; assess the corresponding head. Windows native CLI remains excluded.
Instructions and all limits are in [the browser workflow](browser.md).

Scale/snapshot-sharing commit `1b47e0a26fc9e662c279f4150aa922e9072bec13`
passed [CI 41161857569](https://github.com/cn-hww/mntablemerge/actions/runs/41161857569)
on Linux, Windows and macOS. That run does not validate subsequent browser work.

2026-10-31, scale baseline/snapshot sharing: 68 core tests ran on all four targets;
four additional Wasm CLI tests (72 total), all fourteen existing Wasm workflow
helpers and the five-test packaged consumer on each target passed. New aliasing
tests cover shared input/result/review lifetimes, exported-array/JSON mutation,
exact report/CSV bytes and failure/replay recovery. The measured runner also
actually ran on Wasm, Wasm-GC, JavaScript and Linux native. Native reference
measurement compared seven cases with a warmup and three fresh-process samples
per variant (42 recorded), checking every output cell and equal fingerprints.
The complete raw timings/RSS and host/toolchain details are in
[the reference](benchmarks/native-linux.json); the [measurement contract](performance.md)
states mixed timing outcomes and scope. Format/interface/strict checks passed;
the root public interface is unchanged. This does not establish browser heap or
universal speed/scale guarantees. Assess remote CI on the corresponding head.

2026-10-31, column rename/delete evolution adds eight black-box test blocks
(66 core total). Windows actually ran all core tests on Wasm/Wasm-GC/JavaScript/
native, all four Wasm CLI tests (70 total there), fourteen Wasm end-to-end
helpers, five packaged external-consumer tests and the catalog executable on
each of the four library targets. All passed. New checks cover presence/value
combinations, branch symmetry, unchanged values/zero records, two review phases,
explicit dependency revocation, complete discarded-value evidence, bound full/
assigned sheets, combination/replay, old lineage source compatibility, composed
row/key lineage, stale/policy rejection, Unicode, 512/513 declarations, raw and
escaped wire limits, deterministic output, defensive arrays and failure/retry.
The real-data evolution helper also actually executed the native release CLI on
Ubuntu 22.04/WSL, checking full 19/18-column OurAirports outcomes, audit/revocation,
CSV/TSV replay and driver byte preservation. Public interfaces were regenerated;
format/script/strict all-target checks passed on the pinned 0.10.14 toolchain.
Windows native CLI remains excluded. The new capability's CI must be assessed on
its own head. Prior adoption CI [41102027572](https://github.com/cn-hww/mntablemerge/actions/runs/41102027572)
completed successfully for `717d260cd6890443d5c6e352a44121f4b7892375` on Linux,
Windows and macOS, including the packaged consumer gate; it does not validate
the subsequent evolution change.

2026-10-31, downstream adoption/package gate: the real `moon package` ZIP is
inspected and extracted outside the checkout, then consumed by a separate module
through `moon.work` and only public imports. Windows actually ran four external
test blocks and the catalog executable on Wasm/Wasm-GC/JavaScript/native,
including a generated full-record OurAirports structural-review test read from
the archive. All passed without project warnings; only terminal println output
normalizes Windows CRLF, while library CSV/TSV equality stays exact. Full pinned
verification also passed all 58 core tests on four targets, four Wasm CLI tests,
thirteen Wasm end-to-end helpers, format/interface/strict checks and archive
filter assertions. The template uses a non-main test package and a separate main
entrypoint. `.moonignore` behavior was verified on the pinned toolchain. This is
local package consumption evidence, not Mooncakes upload or registry installation.
Linux/macOS CI is configured for the same consumer gate; assess its own run.

2026-10-31, complete-table file export adds four CLI white-box test blocks:
exact UTF-8 bytes (Unicode/quotes/tabs/newlines), no-replace publication and
failed-rename cleanup, exclusive sidecar collision preservation/retry, and two
competing exporters with exactly one complete winner. Windows ran all 58 core
tests on Wasm/Wasm-GC/JavaScript/native, all four new CLI tests on Wasm (62 tests
total there), and all thirteen Wasm end-to-end helpers. The updated output
helper also passed against the release Wasm CLI with two concurrent processes.
Ubuntu 22.04/WSL actually ran the four new native CLI tests and both debug and
release CLI output workflows; the release helper checks real process competition,
regular/dangling symlink destinations and a symlink sidecar in addition to exact
CSV/TSV output, input preservation, structural conflict, stale review, malformed
input, invalid arguments and recovery. This is runtime evidence, not only a
native build. Format/script checks, strict all-target checks and regenerated
interfaces passed on the pinned 0.10.14 toolchain; the public library interface
is unchanged. An additional Windows strict all-target check passed with
`moonc v0.10.17+093cb8f9c` / `moon 0.1.20261021`; that additional check alone does
not establish runtime compatibility on that version. CI now checks/formats all
automation scripts, runs native CLI tests on Linux/macOS and includes the output
helper. Windows native CLI remains excluded; local macOS execution, power-loss
durability and hostile directory modification are not claimed. Assess remote CI
on the new capability's own commit; earlier green runs do not validate it.

2026-10-12, review-version comparison adds four black-box test blocks (58 total). Tests cover additions/replacements/revocations, complete before/after source evidence, explicit empty versus absent decisions, Unicode/quoted text, reverse symmetry, decision-order independence, immutable earlier bundles, defensive result arrays, JSON/bundle round trips, normalized lineage coordinates, input/policy mismatch rejection, retained dropped structural values and revoked dependencies, and oversized JSON rejection with complete typed evidence/recovery. Windows actually executed all 58 on Wasm/Wasm-GC/JavaScript/native and all twelve Wasm CLI helpers. Format, interfaces and strict all-target checks passed. The new audit helper also actually ran on Ubuntu 22.04/WSL native, verifying a full 18-column OurAirports result, discarded/revoked evidence, stale rejection and deterministic retry. Windows native CLI remains excluded. Combination CI [38214015998](https://github.com/cn-hww/mntablemerge/actions/runs/38214015998) completed successfully at head `62526db611460c7bb816d80034f5119834fc5250`; assess comparison CI on its own commit.

2026-10-12, independent review combination adds four black-box test blocks (54 total). Tests cover disjoint/agreeing/repeated reviews, order independence, exact values and bundle round trips, 0/256/257 batch boundaries, partial review, disagreement, inherited-decision replacement/revocation rejection, atomic failure/retry, all original snapshots/key/order/lineage guards, normalized coordinates, indivisible concurrent insertion choices and new structural review phases. Windows actually ran all 54 on Wasm/Wasm-GC/JavaScript/native and all eleven Wasm CLI helpers. Format, generated interfaces and strict all-target checks passed. The two-owner OurAirports assignment/combine/replay helper also actually ran successfully on Ubuntu 22.04/WSL native. Windows native CLI remains excluded. Assignment commit CI [38212622912](https://github.com/cn-hww/mntablemerge/actions/runs/38212622912) completed successfully at head `27cbcf973f9413ec1edbe38656e664407bfadc97`; assess combination CI on its own commit.

2026-10-12, explicit review assignments add three black-box test blocks (50 total). Tests cover canonical assignment ordering, partial review with unassigned conflicts retained, empty assignments, exact Unicode/tab/newline values, TSV round trips, independent scope/evidence/coverage/input binding, duplicates/unknown/resolved targets, malformed JSON, Unicode, 512/513 key components and 100,000-target rejection. The existing full-capacity test now verifies a one-target assignment/application against a real 100,000-conflict plan. Windows actually ran all 50 tests on Wasm/Wasm-GC/JavaScript/native and all ten Wasm CLI helpers. Format, interfaces and strict all-target checks passed. The new assignment helper actually ran successfully on Ubuntu 22.04/WSL native. Full-sheet behavior and prior CLI workflows passed; Windows native CLI and spreadsheet I/O remain excluded. Assess this capability's remote CI on its own commit.

2026-10-12, editable review sheets add five black-box test blocks (47 total): exact TSV/bundle round trips, blank-versus-explicit-empty, Unicode/quoted custom values, canonical row reordering, partial/complete review, immutable snapshots, stale original data/order/policies, evidence/header/coverage/choice rejection, atomic failure/retry, new structural review phases, composed lineage coordinates, and rejection of 100,000 conflicts without truncation. Windows actually executed all 47 on Wasm/Wasm-GC/JavaScript/native and all nine Wasm CLI helpers. Format, generated interfaces and strict all-target checks passed. The new sheet helper also actually executed successfully on Ubuntu 22.04/WSL native, checking exact complete OurAirports output, invalid evidence/unused text, recovery and a stale nonconflicting cell. Windows native CLI remains excluded; spreadsheet import/export is not validated. Assess remote CI on this capability's own commit.

Quoted TSV transport adds three test blocks (42 total): 100 exact-string round trips across CSV/TSV, unchanged CSV bytes, tabs/quotes/newlines/Unicode/BOM, malformed input and error recovery, 512/513 columns, structural review and explicit empty-versus-missing values. Windows ran all 42 tests on Wasm/Wasm-GC/JavaScript/native and all eight Wasm CLI helpers; the new TSV helper also actually ran on Ubuntu 22.04/WSL native. Format, generated interfaces and strict all-target checks passed. The helper converts seven complete OurAirports fixtures, verifies all retained cells, driver rejection/retry and guarded review/replay. Windows native CLI remains excluded. Assess remote CI on this capability's own commit.

2026-10-12, review revision/revocation adds six test blocks (39 total). Tests cover replacing/revoking/redoing choices, unchanged old plans/evidence, deterministic retained decision order, JSON/bundle round trips, atomic failure and recovery, explicit removal of structural dependencies, column deletion and concurrent-row source changes, reopened conflicts, lineage/order preservation, stale original inputs, duplicate/unknown/premature/invalid choices, malformed JSON/Unicode, 512/513 key components, edit count and raw/escaped wire limits. Windows executed all 39 on Wasm/Wasm-GC/JavaScript/native and all seven Wasm CLI helpers. The revision helper actually executed successfully on Ubuntu 22.04/WSL native. Strict all-target checks, format and regenerated interfaces passed. CLI merge errors now use `MergeError::diagnostic`, preserving actionable reasons previously hidden by generic Error display. Windows native CLI remains excluded.

Row-identity commit CI [38203809513](https://github.com/cn-hww/mntablemerge/actions/runs/38203809513), head `372826174605f294d0a7304d51589d5f49c68018`, completed successfully. Revision capability CI must be assessed on its own commit; this prior run does not validate it.

2026-10-12, explicit retained-row lineage adds five test blocks (33 total): independent edits during ID migration, simultaneous key cycles, mapping-order independence, compound/Unicode keys, column-lineage composition, output-key conflicts, original-input review/replay, stale IDs/data, collision/absence/duplicate/malformed declarations, 512/513 key components, declaration count limits, ownership, CSV/JSON/bundle round trips and structural-rule preservation. Windows executed all 33 on Wasm/Wasm-GC/JavaScript/native; all six Wasm CLI helpers passed. The new identity CLI helper also actually ran successfully on Ubuntu 22.04/WSL native. Format, interface generation and strict all-target checks passed. Native Windows CLI remains excluded. Remote CI for this capability must be assessed on its own commit.

Column-lineage commit CI [38130530391](https://github.com/cn-hww/mntablemerge/actions/runs/38130530391) passed all three OS jobs. This is prior-capability evidence, not a substitute for the row-lineage run.

Explicit column lineage adds five test blocks (28 total) covering rename versus value edits, simultaneous swaps, renamed compound keys, mapping-order independence, defensive copies, CSV/lineage/review round trips, normalized conflict coordinates, cumulative resolutions, original-input stale rejection, missing/duplicate/colliding/malformed mappings, Unicode, the 512/513 declaration boundary, and unchanged structural deletion rules. `scripts/rename-demo.mbtx` exercises merge, conflict review/resolve/replay and stale rejection on real source records, plus Git-driver collision preservation/retry. Ordering commit CI [38129259943](https://github.com/cn-hww/mntablemerge/actions/runs/38129259943) passed all three OS jobs; lineage must be assessed on its own run.

Before the lineage commit, Windows and Ubuntu 22.04/WSL each executed all 28 tests on Wasm/Wasm-GC/JavaScript/native. All five CLI helpers passed on Windows Wasm and on Linux Wasm/native. Strict checks, format and interface generation passed. One overlapping Windows/WSL run failed a Wasm review subprocess while both used the same build directories; the subsequent serial Linux full run passed. These results do not add Windows native CLI support.

The ordering capability adds three test blocks (23 total): surviving/deleted/new row order, compound keys, empty preferred input, unchanged default, deterministic output, CSV/review round trips, cumulative resolution, strict metadata and recovery. `scripts/order-demo.mbtx` runs a real-data CLI merge and input-bound review/replay, including override rejection. Its native run requires the same supported Linux/macOS host as the other CLI demos.

Before the ordering commit, all 23 test blocks ran successfully on Windows Wasm/Wasm-GC/JavaScript/native, every Wasm CLI helper passed, and the ordering CLI workflow actually executed successfully on Ubuntu 22.04/WSL native. The third commit's CI [38127787040](https://github.com/cn-hww/mntablemerge/actions/runs/38127787040) also passed all three OS jobs. New capability CI evidence must be assessed on its own commit.

2026-10-11, MoonBit stable `moonc v0.10.14+7d59c7ec9` / `moon 0.1.20260920`. Initial capability evidence:

- Windows: strict check of all targets; 13 test blocks executed successfully on Wasm, Wasm-GC and JavaScript. Native pure-library tests also executed successfully with LLVM/MinGW (not merely compiled).
- Ubuntu 22.04 under WSL: the same toolchain actually executed all 13 native test blocks and the native CLI on the complete OurAirports merge. The end-to-end helper checks CSV output, conflict/error statuses and resolution. Native async dependency C sources produce compiler warnings on this toolchain; MoonBit project sources pass `--deny-warn`.
- Wasm CLI: offline real-data end-to-end helper passed, including a missing intersection and explicit recovery.
- Windows native CLI: **not supported/validated in this environment**. `moonbitlang/async@0.21.3` native Windows I/O requires MSVC; LLVM/MinGW compilation fails in upstream C sources. CI tests the Windows native pure library only, with Wasm CLI I/O. Linux/macOS jobs exercise native CLI I/O.
- CI targets Linux, Windows and macOS. [Run 38126708323](https://github.com/cn-hww/mntablemerge/actions/runs/38126708323) for the second commit passed all three jobs, including all 16 core tests on all four backends, Wasm CLI workflows on every OS, and actual native CLI workflows on Linux/macOS. Windows native CLI remains explicitly excluded.

Reproduce with `moon run scripts/verify.mbtx`. On Windows, an optional compiler-bin directory is prepended to PATH. Tests use no private infrastructure. See [demo](../examples/airports/README.md). No performance, stress-scale or cryptographic review-authenticity results are claimed.

The second capability adds 3 review test blocks (16 total), checking bundle round trips, cumulative structural/cell decisions, stale input/order/key rejection, malformed envelopes and recovery. The review helper runs export → resolve → guarded replay → stale rejection using actual CLI processes.

Remote run [38126449249](https://github.com/cn-hww/mntablemerge/actions/runs/38126449249) for the first commit failed on all three hosts before compilation: the freshly installed toolchain had no registry index, so the `.mbtx` dependency was unavailable. CI now explicitly runs `moon update` before starting the script. Review tests passed on all four backends locally; the review CLI workflow passed on Windows Wasm and actual Linux native execution.

The third capability adds actual Git branch-merge integration plus malformed-CSV/UTF-8, alias and sidecar-collision checks and recovery, including byte preservation of ancestor/other inputs. Four additional preservation test blocks (20 total) reject lone UTF-16 surrogates in programmatic tables/decisions, malformed key dimensions/deep JSON, and cover a stored header BOM plus CSV quote expansion beyond 16 million syntax units. CSV transport limit is 40 million units, separate from the 16 million semantic table limit, so every valid table can round-trip after escaping. The Git helper exercises Wasm and Linux/macOS native; Windows tests the Wasm driver. Assess its CI results on the third commit's own run; the second run does not prove this new integration.

Before the third commit, all 20 tests executed successfully on Windows Wasm/Wasm-GC/JavaScript/native, and the full Wasm CLI workflows passed. Actual native CLI merge, review and Git-driver workflows also passed under Ubuntu 22.04/WSL. Format checking and regenerated interfaces produced no drift.