MoonTxnWitness v0.3.0

MoonTxnWitness v0.3.0

• Deliver the native MLflow HTTP log-batch repair case: 30 fixed seed pairs, actual 40P01 evidence, 1000-to-623 key reduction and three fixed-side reruns of the identical final case. oneMinimal remains false.
• Run node examples/mlflow-http-roundtrip.mjs to diagnose and replay two unchanged real captures without a database. Source gaps and inconclusive model results remain visible.
• Add explicit bounded source-write buffering to configured user/model profiles while preserving legacy defaults, tail synchronization and refusal of incomplete seals.
• Replay 156 original public v0.2.8 cases across three genuine build contexts with fixed runtime identity checks and unchanged original expectations.

Formal capture-cost experiments completed, but the throughput/P95 targets were not met; failed CRUD cells remain ineligible. The measured scope and raw evidence identities are in the capture-cost report.

See the HTTP case and new-run instructions and support matrix. Install with moon add HhWw96/moontxnwitness@0.3.0 or use the portable ZIP for Node.js.

MoonTxnWitness v0.2.8

MoonTxnWitness v0.2.8

• Add diagnoseV3 / verifyDiagnosisV3 and CLI diagnosis/verification, combining the original serial analysis and model checks with independently verifiable direct evidence. Unknown analysis remains visible; INT does not become G2 or a PL3 refutation.
• Reuse parsed relation-reduction candidates while retaining full semantic revalidation, sources and budgets. Fixed relation-candidate workloads measured 1.124x speedup; this is not an overall v0.2.8 SDK or database-throughput claim.
• Replay all 46 genuine public v0.2.7 fixture families through the fixed original runtime, including configured model profiles, and retain the full toolchain licences and NOTICE.

Configured register/list/relation profiles, explicit model checks, missing-source gaps and existing budgets remain as documented. The MLflow case remains a component-level reproduction; the original HTTP log_batch repair comparison is unfinished.

See the support matrix, diagnosis guide and measured performance scope. Actual release, database, fresh-consumer and public-download receipts remain separate from human feedback and organizer decisions.

Install with moon add HhWw96/moontxnwitness@0.2.8 or use the portable ZIP for Node.js.

MoonTxnWitness v0.2.7

MoonTxnWitness v0.2.7

• 新增配置化 PostgreSQL 列表／关系来源采集，由 MoonBit 核验真实 SQL、结果交付与事务事实。
• 提供 SDK／CLI 分析、证据验证和完整来源包重放，保留无法判定与覆盖缺口。
• 采用 Unicode 流式 JSON 预检查，保留旧错误语义及公开 v0.2.6 的35家族兼容。

安装：moon add HhWw96/moontxnwitness@0.2.7，或下载便携包运行 node examples/model-profile-roundtrip.mjs。

用法与支持范围 · 固定工作负载实测

构建工具链的完整许可与署名另附：THIRD_PARTY_NOTICES.txt。

MoonTxnWitness v0.2.6

MoonTxnWitness v0.2.6

新增寄存器与有界列表的六模型检查：逐规则显示反证、完成或无法判定，保留原始来源、预算和复现依据。新增模型 SDK、命令行和独立复现包；原类别缩减入口保持兼容。

真实公开 v0.2.5 的21类旧证据加入兼容验证。旧实现仅在对应历史重放时按摘要校验后加载，日常当前版本分析不会启动历史工作进程。

三目标各748项核心测试、数据库回归、513项独立检查通过。性能完整输出一致，但1.15×目标与部分尾部耗时控制未通过，原始结果保留。本版不声明达到性能优化目标。

用法：模型检查 · 支持范围 · 实测结果

关系六模型、混合模型事务和模型保持缩减尚未纳入本版；严格模型正例中的合成证据与真实业务采集验收分开记录。

MoonTxnWitness v0.2.5

MoonTxnWitness v0.2.5

本版新增按异常类别定位、缩减和复现的完整入口：分别检查 G0、G1c 和 G2-item，能够发现被其他较短环遮住的目标异常；缩减时保留所选类别及必要读取来源。

• SDK：classifyV3、reduceClassificationV3、verifyClassificationV3。
• 示例：node examples/classification-roundtrip.mjs。
• 兼容真实公开 v0.2.4 的 15 类证据；旧 ORM 复现所需代码按需加载。
• 增加显式有界采集实验配置，公布完整 A/B/C 成本和未达标结果。本版不宣称分析提速或低开销采集。

本轮核心回归每目标 691 项，保留 10,320 份独立历史和 6,373 份环证据核对；新增分类、真实旧包兼容、数据库及全新安装验证均保留具体记录。

分类入口目前支持来源完整且必要版本顺序明确的点寄存器；未知来源、预算耗尽和不支持的模型保持可见。旧分析入口继续保留原有行为。

使用方法 · 性能实测 · 验证记录

安装：moon add HhWw96/moontxnwitness@0.2.5，或下载 portable ZIP 后使用 Node.js 20 及以上运行。

MoonTxnWitness v0.2.4

MoonTxnWitness v0.2.4

• Configure the fixed base-user profile at business startup to collect original SQL, delivered results and version sources. MoonBit validates the complete supported ID-CRUD window before analysis.
• New explicit ORM-profile SDK/CLI entries analyze, reduce within the source-closed predicate and replay exact source-bound capsules. Generic diagnostics retain their coverage gaps.
• Preserve eight genuine public v0.2.3 capsules and earlier released replay formats. Correct full capture cost startup synchronization and failed-cell handling.

Validation: 672 core tests per JS/WasmGC/Wasm target; 18 driver contracts; 50 host checks; two current real PostgreSQL business runs; existing PostgreSQL regressions; independent clean-archive review; fresh source/portable consumers. Formal performance gains are not claimed in this release.

Scope and setup: https://github.com/HhWw96/moontxnwitness/blob/v0.2.4/docs/orm-profile-sdk.md
Evidence: verification/releases/0.2.4/

Install: moon add HhWw96/moontxnwitness@0.2.4 or use the Node.js portable ZIP.

MoonTxnWitness v0.2.3

MoonTxnWitness v0.2.3

• Bounded batching for segmented event journals and optional provenance source logs, with explicit enqueued and durable watermarks and tail synchronization.
• Export full sealed journals while retaining original logs, bound active diagnostic state, and reject incomplete or failed persistence.
• Preserve existing synchronous source semantics and verify eight genuine public v0.2.2 capsules alongside the earlier public20/public21 fixtures.

The lifecycle observation file retains its original close-time synchronization contract. The A/B/C cost protocol is included; formal throughput and latency results are not yet claimed.

Validation:629 core tests per JS/WasmGC/Wasm target, independent histories and cycle certificates, grouped persistence and failure tests, six actual PostgreSQL18.6 instrumentation cases, and fresh source/portable consumers. Scope and evidence: docs/grouped-source-persistence.md and verification/releases/0.2.3/.

Install: moon add HhWw96/moontxnwitness@0.2.3 or use the Node.js portable ZIP.

MoonTxnWitness v0.2.2

MoonTxnWitness v0.2.2

• Attach collection to one SQLAlchemy engine or a local psycopg factory at startup, with sync/async support and no global driver patch or database migration.
• Preserve actual result delivery, exceptions, executemany consumption, pool/savepoint lifecycle and native COMMIT/ROLLBACK observations. Lost COMMIT responses remain unknown.
• Replay eight genuine public v0.2.1 capsules alongside the original14 public v0.2.0 capsules without rewriting observations or version-shaped business values.

This entry produces execution diagnostics with visible coverage gaps. Automatic version provenance, complete ORM histories and collection-performance measurements follow in later versions.

Validation:38 original instrumentation protocol tests, six real PostgreSQL18.6 sync/async and lost-response cases,629 core tests per JS/WasmGC/Wasm target, independent histories/cycle certificates, compatibility and fresh source/portable consumers. Exact receipts and scope: docs/diagnostic-instrumentation.md and verification/releases/0.2.2/.

Install: moon add HhWw96/moontxnwitness@0.2.2 or use the Node.js portable ZIP.

MoonTxnWitness v0.2.1

MoonTxnWitness v0.2.1

• Pin the actual public v0.2.0 runtime and input hashes for complete SDK benchmarks: register, list, relation, normalization and reduction plus20 unchanged controls.
• Replay14 genuine public20 model, execution and invalid-diagnostic capsules with155 compatibility checks.
• Preserve both measured optimization candidates and their failed P95 controls. Structured normalization is withdrawn from the stable default; no stable21 speedup is claimed.

Run node scripts/performance-v020.mjs for the full benchmark, or --quick for preparation only. See docs/performance-v0.2.1-withdrawal.md and CHANGELOG.md for scope and evidence.

Validation includes629 core tests on JS/WasmGC/Wasm, Native check, independent full-output and replay checks, PostgreSQL18 scenarios and fresh source/portable consumers. Existing SDK and replay entry points remain compatible.

Install: moon add HhWw96/moontxnwitness@0.2.1 or use the Node.js portable ZIP.

MoonTxnWitness v0.2.0

MoonTxnWitness v0.2.0

• One bounded support matrix for registers, fixed-key presence, SQL templates/cache, append lists, relation facts, selected-cycle explanation and execution capsules.
• Preserve explicit budgets, source gaps and replay rules, including v0.1.8 model and v0.1.7/v0.1.8 execution capsules.
• Keep full automatic SQL/ORM, six-model classification and complete production integrations on the future roadmap.

See the stable scope and examples. Actual regression, database, independent review, fresh installation and public download receipts are recorded separately; no unmeasured performance figure is claimed here.

Install with moon add HhWw96/moontxnwitness@0.2.0 or use the portable ZIP for Node.js.