@functional-systems/lambdadb-cli 0.1.4-dev.16 → 0.1.4-dev.17

package/VALIDATION.md

@@ -1,1017 +0,0 @@
-# Validation record
-
-## CLI 0.1.3 publication (2026-10-06)
-
-With explicit release authorization, annotated tag `v0.1.3` was created on main
-commit `90628ed3648aea0ae381c061a95aebbbf48e9259` and the stable
-[GitHub Release](https://github.com/lambdadb/lambdadb-cli/releases/tag/v0.1.3)
-triggered [OIDC publication](https://github.com/lambdadb/lambdadb-cli/actions/runs/37466450036).
-The workflow succeeded on attempt 1, repeating Node 22/24 validation and
-installed-tarball tests before publishing the exact tested artifact.
-
-- Registry reads confirmed `latest=0.1.3`, `dev=0.1.3-dev.15`.
-- `node /tmp/lambdadb-cli-release013/main/verify-publication.mjs` fetched registry
-  metadata, downloaded the tarball and verified its SHA-512 integrity. Its bytes
-  exactly match the final reviewed main candidate described below.
-- A clean, unqualified `npm install --prefer-online --cache <isolated-cache>
-  --no-audit --no-fund @functional-systems/lambdadb-cli` selected CLI 0.1.3 and
-  SDK 0.8.0. Version and query/update help passed. All 54 CLI contract tests
-  executed against this registry-installed binary, with no failures or skips.
-- `npm audit signatures --prefer-online --cache <isolated-cache>` verified four
-  registry signatures and three attestations. The isolated cache avoids stale
-  local npm metadata; the user's global installation was not changed.
-- Decoded [SLSA provenance](https://registry.npmjs.org/-/npm/v1/attestations/@functional-systems%2flambdadb-cli@0.1.3)
-  matches the downloaded artifact's digest, release commit, `refs/tags/v0.1.3`,
-  repository and `.github/workflows/publish.yaml` run 37466450036.
-- The registry's full metadata, npm installation metadata and signature lookup
-  became visible at different times. Initial reads returned 404, selected the
-  older latest package or returned ETARGET during audit. Only reads and isolated
-  installation were retried; publication was not repeated and no tag was moved.
-
-Published tarball SHA-256:
-`ef98f935c5274ee6b4c7991669e46648fee4070a1105b1a265df6a6faf791b07`.
-Its SHA-512 integrity is
-`sha512-vgS5U+dVBVpjpi1KVhHr6XL/VPj6gJ3nHxWg8osGwK2ZiDP8CzZ4o3ebMzEP13Pg+G98akZKTiitiM2prvCWsw==`.
-The matching immutable release `package-lock.json` has SHA-256
-`4e1a5028be3b610e3c55f2e91954270ae767e1d958810a3ac469f056f82d1c81`.
-These values prepare the 0.1.3 Homebrew formula; they do not publish the public tap.
-`bash scripts/test-homebrew.sh` passed locally on macOS arm64: formula style,
-installation, brew test, wrapper isolation from ambient Node and all 54 released
-CLI contracts. The test uninstalled its CLI and removed its disposable tap.
-This was a fresh local installation, not an existing-user upgrade.
-
-The post-release PR brings main history into develop, starts `0.1.4-dev.1`, and
-updates publication records and the local Homebrew formula. Source/query behavior
-is unchanged. Homebrew CI results are recorded in that PR; the public tap still
-serves 0.1.2 pending its separate reviewed update. No production validation,
-server deployment, billing verification or new search-default promotion occurred.
-The installed-candidate live smoke and verified Collection cleanup below remain
-applicable because the published runtime is byte-identical.
-
-Sanitized original logs, registry metadata, provenance, the publication tarball
-and the isolated consumer are retained under `/tmp/lambdadb-cli-release013/main/`.
-No scratch artifacts or credentials are committed.
-
-
-## CLI 0.1.3 release preparation (2026-10-06)
-
-Prepared from reviewed develop `caee1bcb6ce864c33647673b54a48f7cc06dcf84`
-([PR #18](https://github.com/lambdadb/lambdadb-cli/pull/18)); its tree matches
-feature head `9589c7a`, whose Node 22/24 CI and automatic review completed with
-no findings. The release branch changes only version/changelog and release
-evidence relative to develop. Runtime source, examples, tests and workflow
-remain identical. Stable publication is not authorized or claimed here.
-
-On Node 24.15.0, these commands passed:
-
-```sh
-npm ci
-npm run lint
-npm run check:version
-RELEASE_TAG=v0.1.3 RELEASE_PRERELEASE=false node scripts/check-release.mjs --release
-npm run typecheck
-npm test
-npm pack --json --pack-destination /tmp/lambdadb-cli-release013
-npm run test:package -- /tmp/lambdadb-cli-release013/functional-systems-lambdadb-cli-0.1.3.tgz
-```
-
-The metadata check validates release settings without creating a tag. All 79
-source tests and 54 separately installed CLI tests passed, with zero failures
-or skips. The candidate inventory contains 45 files, including the new examples
-and no credentials, scratch files or test harnesses.
-
-An isolated installation of the packed 0.1.3 candidate, with SDK 0.8.0, then ran
-the actual CLI against authorized development project `bench-recall` at
-`https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`:
-
-```sh
-npm install --prefix /tmp/lambdadb-cli-release013/consumer --no-audit --no-fund /tmp/lambdadb-cli-release013/functional-systems-lambdadb-cli-0.1.3.tgz
-LAMBDADB_TEST_CLI=/tmp/lambdadb-cli-release013/consumer/node_modules/@functional-systems/lambdadb-cli/dist/cli.js node --env-file=.env.local /tmp/lambdadb-cli-sdk080/run-live.mjs
-```
-
-The launcher used the existing private environment only in memory and ran
-`npm run test:live`; the harness selected the installed binary through
-LAMBDADB_TEST_CLI. The complete suite passed in 113.4 seconds: 1 test, zero
-failures/skips. It exercised analyzer metadata, ordinary/bulk imports, committed
-query/fetch contents, facets, native/legacy create/update and actual embeddings,
-Bayesian budgets and reranking, retrieval scores and all 30 server rejection
-checks described in the SDK 0.8.0 record below.
-
-Deletion and subsequent HTTP 404 verified cleanup of all three Collections:
-
-- `cli-smoke-29074d17-7e21-460a-9823-a6e49f30d067`
-- `cli-native-3c9bcae9-1b91-424f-8504-ac755078ac13`
-- `cli-native-3b18c438-83d7-44eb-8c8d-bfc7949f35c5`
-
-The live-tested tarball has integrity
-`sha512-GzQVy1ZJgtT8O2q5wXDV+mhegYk6MBzZ6hSNMOjKq1BGUu6dW1P0F69dHDFJC3Q8o6gvUYQGv41DZMvfMXxaIQ==`.
-It and its manifest remain under `/tmp/lambdadb-cli-release013/live-tested/`;
-logs and the installed consumer remain in the parent directory. This evidence
-section changes packaged documentation after the live run. The final package is
-repacked and checked by the installed CLI suite; its runtime and examples are
-compared byte-for-byte to the live-tested installation. Its final integrity and
-remote CI/review results are recorded in the release PR.
-
-The earlier same-day deployment provenance record below identifies backend
-`9072a1bc8925954369a887f558f1eaf387b7ea0e`. This bounded test does not establish
-production availability, ranking quality, load or billing behavior. No new
-Homebrew installation or older-server tests were run. The checked-in/public tap
-still points to published 0.1.2; its 0.1.3 update follows a separately authorized
-stable publication. No server deployment, npm latest change, tag or GitHub
-Release was performed by this preparation.
-
-
-## SDK 0.8.0 CLI compatibility (2026-10-06)
-
-Candidate: `feat/sdk-0.8.0`, based on develop `8e76422`; CLI development version
-`0.1.3-dev.1`, exact SDK `0.8.0`, Node `24.15.0`. This change does not publish
-or promote a package. Repository instructions contain no local AGENTS.md;
-CONTRIBUTING.md, RELEASING.md, CI, runtime/input/output and adjacent commands
-were inspected before implementation.
-
-### Local and installed validation
-
-The following commands passed:
-
-```sh
-npm ci
-npm run lint
-npm run check:version
-npm run typecheck
-npm test
-npm run test:package
-git diff --check
-```
-
-`npm test` executed 79 tests, with no failures/skips. `npm run test:package`
-packed and separately installed the CLI, checked inventory/version/help, and
-executed all 54 CLI contract tests with no failures/skips. This is executed
-subprocess coverage, not merely compilation. New coverage includes Bayesian
-JSON mapping and budgets, omitted/null defaults, ref/input precedence, unchanged
-ordinary fusion requests, local versus server errors, native/legacy create and
-update, contradictory embedding inputs, uncertain mutations without retries,
-and full JSON/human results with rerank metadata through inline and docsUrl
-responses. No SDK query DSL validator was added.
-
-### Current deployment provenance
-
-Read-only AWS inspection used profile `dev`, region `ap-northeast-2`:
-`aws ecs list-services`, `describe-services`, `list-tasks`, `describe-tasks`,
-and `aws ecr describe-images --image-ids imageTag=dev-v3-9072a1b` for the
-Gateway and Query Executor repositories. Both services had desired/running
-counts 1/1, task definition revision 16 and rollout `COMPLETED`.
-Running digest and the ECR `dev-v3-9072a1b` tag agreed:
-
-- Gateway: `sha256:300f65269579fcff327366490505327a549e38249c371e93df07f2ae0669bfef`.
-- Query Executor: `sha256:bf87b32fdcbaf1b17f5cacb3ec8e9f1af1eb7ef51566d06cfb74273e3f857a54`.
-
-`gh run view 37422611173 --repo lambdadb/lambdadb --json headSha,conclusion,url`
-confirmed the [successful deployment](https://github.com/lambdadb/lambdadb/actions/runs/37422611173)
-for pinned backend `9072a1bc8925954369a887f558f1eaf387b7ea0e`. The deployment
-run, image tag and running digests were compared together; no deployment or
-infrastructure configuration was changed.
-
-### Actual CLI on the authorized development target
-
-Target: `bench-recall` at
-`https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`, following the
-existing repository test-environment workflow. The repository's private
-`.env.local` supplied its existing opt-in and credential; nothing was borrowed
-from another repository. The command was:
-
-```sh
-node --env-file=.env.local /tmp/lambdadb-cli-sdk080/run-live.mjs
-```
-
-This uncommitted launcher verified the exact endpoint/project and opt-in,
-mapped `LAMBDADB_BASE_URL`, `LAMBDADB_PROJECT_NAME`, and
-`LAMBDADB_PROJECT_API_KEY` in memory to the CLI variables, set
-`LAMBDADB_LIVE_CONFIRM_PROJECT`, and ran `npm run test:live` (build followed by
-`node --test test/live/*.test.mjs`). Credentials never entered command arguments,
-logs or committed files. The suite passed in 94.8 seconds: one test, zero
-failures/skips, using the actual CLI for all creation, updates, imports and reads.
-
-- Doctor, all 49 analyzer presets, ordinary/bulk imports, exact committed
-  query/fetch document contents, facets, and default/null/custom reranking passed.
-- Both native embedding-only and explicit legacy true configurations passed
-  create/update and normalized metadata checks with explicit 256 dimensions and
-  cosine similarity inside embedding. Actual OpenAI document/query embeddings
-  worked through ordinary upsert and KNN queryText.
-- Bayesian retrieval returned all expected documents. Holding candidateSize=30
-  while changing size from 3 to 1 preserved the result prefix. Ordinary
-  text/KNN/RRF/Min-Max/L2 worked without a top-level candidate budget.
-- Default and explicit rerank budgets both produced `applied`, three candidates,
-  three scored candidates and two final results. Finite final scores were in
-  [0,1], and retrievalScore exactly matched the baseline Bayesian scores. Null
-  rerank preserved the baseline documents/scores without rerank metadata.
-- For each native/legacy Collection, 15 invalid requests retained API_ERROR,
-  HTTP 400 and exit 3: missing/invalid/conflicting candidate budgets, wrong signal
-  counts, boosts including Boolean descendants, nested fusion and a Bayesian
-  budget on an ordinary query.
-- Cleanup deleted and verified HTTP 404 for all temporary Collections:
-  `cli-smoke-61877546-2601-4df9-aeeb-91d87f44c2a5`,
-  `cli-native-4e72c926-fe9e-4bbd-822e-1667be3e619f`, and
-  `cli-native-2892e6ab-af3c-4a23-aff7-bb5793f7f31e`.
-
-Sanitized logs and deployment evidence are retained outside the repository in
-`/tmp/lambdadb-cli-sdk080/`. No temporary test data remains in those Collections.
-The live run used the built source CLI; installed-artifact validation used
-loopback servers. Full downloaded documents are covered by local transport tests;
-large live results alone do not prove that the server selected docsUrl.
-Older-server deployment, production behavior, search quality, load, billing,
-and Homebrew installation were not tested. Older servers may require explicit
-`managedEmbedding: true` and may reject Bayesian search. Existing flag precedence,
-ordinary query behavior, output schema and SDK transfer handling are preserved.
-
-
-Updated: 2026-10-04. Local contract tests use loopback servers and synthetic
-credentials. Authenticated development-project evidence is recorded separately.
-
-## CLI 0.1.2 publication (2026-10-04)
-
-[PR #15](https://github.com/lambdadb/lambdadb-cli/pull/15) merged as
-`6628fe0ed1f405c88fbd91d591e5223d3250dd73`. The annotated v0.1.2 tag points to
-that commit. [GitHub Release v0.1.2](https://github.com/lambdadb/lambdadb-cli/releases/tag/v0.1.2)
-triggered [OIDC publication](https://github.com/lambdadb/lambdadb-cli/actions/runs/37191188846),
-which passed Node 22/24 validation and exact-artifact tests on attempt 1.
-
-- Registry latest=0.1.2; dev=0.1.2-dev.12 remained unchanged.
-- Downloaded npm tarball is byte-identical to the reviewed final candidate and
-  merged-main pack. SHA-256:
-  `e63c2c0fd624e1d81c45c5ba24d641a11ee0ecdc60f063aba4268e06547761ea`.
-  Registry SHA-512 integrity:
-  `sha512-9l1WPN2vrX+JXRjuOh7Uz5DCVcaiV/YLpYKgZb7ORU1jphaAIx38VSWeiQitz0V2Nw/b1vTxfcXzvra6EAkaFg==`.
-- An isolated registry consumer passed all 46 installed CLI contracts and
-  version/help. Its lockfile integrity matched npm and SDK resolves to 0.7.0.
-  npm audit signatures verified four registry signatures and three attestations.
-- Decoded [provenance](https://registry.npmjs.org/-/npm/v1/attestations/@functional-systems%2flambdadb-cli@0.1.2)
-  matched the artifact digest, repository, release commit, refs/tags/v0.1.2,
-  publish.yaml and workflow run 37191188846. Verification retried registry reads
-  during propagation without another publication. Temporary consumer, archive
-  and cache were removed.
-- The candidate live evidence below applies to the identical published artifact;
-  no additional live run was performed. The earlier failed sample remains recorded.
-- Formula handoff pins this npm tarball and release lockfile SHA-256
-  `e2d12af262d13c3a23073d1313b183766bb40b5360abeb0d773b5100819dac65`.
-  The completed tap handoff and bounded upgrade evidence are recorded below.
-
-These checks establish artifact publication and the recorded development sample;
-production deployment, search quality, load/failure coverage and billing readiness
-remain separate dependencies.
-
-## CLI 0.1.2 post-release verification (2026-10-04)
-
-[PR #16](https://github.com/lambdadb/lambdadb-cli/pull/16) merged as
-`aefbaeed9c799e76dd7a9501ecda6ab80dc6e958`, preserving release commit `6628fe0`
-as an ancestor of develop and starting the 0.1.3-dev.1 development base.
-
-- [Automatic publication](https://github.com/lambdadb/lambdadb-cli/actions/runs/37192027479)
-  passed Node 22/24 validation and published 0.1.3-dev.13. An independent isolated
-  registry consumer passed all 46 CLI contracts, version/help, SDK 0.7.0 and
-  integrity checks. npm audit signatures verified four signatures and three
-  attestations. Decoded provenance matched the artifact digest, merge commit,
-  repository, develop ref and workflow run. At verification, dev=0.1.3-dev.13
-  and latest=0.1.2. Temporary consumer, source archive and cache were removed.
-- [Homebrew PR #6](https://github.com/lambdadb/homebrew-tap/pull/6) merged as
-  `ae9a7456acd10015d51b9ea8d58aee76afe3da28`. The public formula matches the
-  CLI-side formula and pins the verified npm 0.1.2 artifact and release lockfile.
-- [Public-tap CI](https://github.com/lambdadb/homebrew-tap/actions/runs/37192619208)
-  passed on macOS arm64 and Linux x86_64. It installed from the remote public tap
-  and checked formula equality, style, version, JSON configuration, file
-  permissions and Node runtime selection. Existing migration checks also passed;
-  no migration formula was changed. Hosted test installations and taps were removed.
-- A separate local macOS arm64 test installed 0.1.1 into a disposable tap, replaced
-  its formula with the exact merged public 0.1.2 formula, ran brew upgrade and
-  verified version 0.1.2 and brew test. Both test-owned versions and the temporary
-  tap were removed; the original user taps were preserved. This establishes the
-  formula version transition, not an existing user's public-tap upgrade or Linux
-  upgrade coverage. No LambdaDB service resources or credentials were used.
-- Merged feature branches were removed and both primary checkouts were clean.
-  The candidate and development-service evidence below remain historical samples;
-  this post-release verification did not run another live service smoke.
-
-## CLI 0.1.2 stable candidate (2026-10-04)
-
-This section records the pre-publication candidate milestone. Publication and
-post-release results are recorded above.
-
-Prepared release/0.1.2 from reviewed develop merge `aef74a3`, retaining SDK 0.7.0
-and all reviewed runtime/test changes. The release branch changes only version
-metadata, the dated changelog and release/validation documentation. Stable npm
-0.1.2 was absent when checked; latest remains 0.1.1. No stable tag, GitHub Release,
-package publication or main merge was performed during preparation.
-
-- npm ci, lint, typecheck, version checks and the explicit
-  RELEASE_TAG=v0.1.2 / RELEASE_PRERELEASE=false preflight passed on Node 24.15.0
-  with npm 11.12.1. All 71 source tests and 46 separately installed CLI tests passed.
-  The executable and lockfile agree on 0.1.2, and the installed SDK is 0.7.0.
-- A candidate tarball was installed into an isolated consumer with its own npm
-  cache. Its executable was selected via LAMBDADB_TEST_CLI for the authorized
-  development smoke using the existing .env.local target and in-memory mapping.
-  The successful run took 54.4 seconds: accepted writes at about 3 seconds,
-  committed query/fetch contents at about 52 seconds, exact facets and managed
-  reranking default/null/custom by about 54 seconds. All 49 preset metadata
-  checks passed. Cleanup returned HTTP 404 for
-  `cli-smoke-14b21fee-59e1-4898-9ef2-8a57db823de0`.
-- The first candidate live attempt failed after 53.2 seconds during query
-  observation with exit 3 and no HTTP status. The log does not distinguish a
-  transport timeout from another request failure; this is not a passing sample
-  or proof of a candidate defect. No runtime code was changed for the retry.
-  Cleanup was not confirmed by that run. An independent SDK get found its
-  test-owned collection still present; delete was accepted and a subsequent
-  get returned HTTP 404 for `cli-smoke-c42a987e-592a-4611-a8d7-04cf5b90ad15`.
-- Every live launcher checked .env.local's SHA-256 before/after execution. It
-  remained byte-for-byte unchanged. Temporary consumers, caches and initial
-  candidate tarballs were removed.
-- Only documentation was updated after the successful live run. The final
-  review tarball is rebuilt and checked again through the installed-package
-  suite; generated dist remains ignored. Candidate publication will be verified
-  independently after approval, rather than assuming dev provenance applies.
-- The public Homebrew tap was read and matches the verified 0.1.1 formula. Its
-  0.1.2 update waits for the stable npm artifact and immutable release lockfile.
-
-The live checks are bounded development evidence. Provider failure/load behavior,
-dense vector reranking, new-preset language quality, production deployment and
-billing readiness remain outside this verification. The PR must pass Node 22/24
-CI and receive review before main promotion. Tagging and publication are separate
-approved actions; synchronize main back into develop with 0.1.3-dev.1 afterward.
-
-## Published SDK 0.7.0 development consumer (2026-10-04)
-
-The [develop merge workflow](https://github.com/lambdadb/lambdadb-cli/actions/runs/37188862636)
-completed successfully and published
-[`0.1.2-dev.12`](https://www.npmjs.com/package/@functional-systems/lambdadb-cli/v/0.1.2-dev.12)
-from `aef74a3deeacb1e5d3f50856301562b58739fcdd` on attempt 1. Required Node
-22/24 validation and the exact publication artifact tests passed. Registry
-visibility was checked with reads while the automatic job completed; no second
-publication was attempted. At verification, dev=0.1.2-dev.12 and latest=0.1.1.
-
-- Downloaded tarball SHA-512 matched registry integrity and the clean consumer
-  lockfile. SDK dependency and the installed SDK both resolve exactly to 0.7.0.
-  Integrity: `sha512-AyGfP2ibeu/unYj5RvkkBO2WICkCMk6dp+4TauH0Da2xBsN6g3Nub7MmM/ZeiRE93wwV9r05IhXV2DyS28WY1A==`.
-- Registry/installed gitHead and decoded
-  [provenance](https://registry.npmjs.org/-/npm/v1/attestations/@functional-systems%2flambdadb-cli@0.1.2-dev.12)
-  agree on the merge commit, repository, develop ref, publish.yaml workflow and
-  run 37188862636; the provenance subject digest matches the downloaded artifact.
-- A fresh tarball consumer passed version/help and all 46 CLI contracts. Tests
-  came from an archived merge-source snapshot, with its expected package version
-  set to the generated published version. A separate registry-version consumer
-  with an isolated cache verified matching lockfile integrity and SDK version;
-  npm audit signatures verified four registry signatures and three attestations,
-  including the published CLI. The preliminary local-tarball audit verified only
-  the three registry dependencies, so the registry consumer verified the CLI too.
-- The published CLI's executable was selected through LAMBDADB_TEST_CLI for the
-  expanded authorized development smoke. All 49 preset metadata, ordinary/bulk
-  writes, full committed query/fetch contents, exact facets and managed reranking
-  default/null/custom checks passed in 81.9 seconds. Query contents appeared at
-  about 78 seconds; acceptance at about 2 seconds did not imply search readiness.
-- Cleanup was accepted and HTTP 404 confirmed for
-  `cli-smoke-5028c4c7-535c-4b29-8978-b3b805054d7d`. Temporary consumers, caches,
-  source snapshots and tarballs were removed. .env.local remained byte-for-byte
-  unchanged. The designated target and in-memory environment mapping are recorded
-  in the authenticated follow-up below.
-
-This establishes the published development artifact and a bounded sample on one
-development target. It does not establish stable publication, production feature
-deployment, language-specific search quality, load/failure coverage or billing
-readiness.
-
-## SDK 0.7.0 local compatibility (2026-10-04)
-
-The checkout began clean on develop at `3903531`. npm now pins the released
-SDK 0.7.0 in both package.json and package-lock.json. Existing 0.1.1 publication
-and live evidence below remain historical; no new CLI package was published.
-
-- `npm ci`, lint, typecheck and version/lockfile checks passed with Node 24.15.0
-  and npm 11.12.1. npm reported zero dependency vulnerabilities.
-- Source suite: 69 tests passed on Node 24.15.0 and Node 22.23.3, with no failures
-  or skips. Installed tarballs: 44 CLI tests passed on each runtime in temporary
-  consumers; package inventory and executable version/help checks passed.
-- Expanded analyzer forwarding covers all 49 names, existing omitted/default
-  serialization and lowercase rejection. Rerank cases check custom criteria,
-  null/omission, separate size/candidate cap/vector k, local validation, final and
-  retrieval scores (including zero/precision), tie order, metadata and facets.
-  Applied/empty/fallback/unused results are checked inline and through docsUrl;
-  API failures remain errors with returnOriginal. These are synthetic transport
-  contracts, not induced provider failures or language-quality measurements.
-- Repeated TypeScript builds produced identical hashes for all 21 generated
-  files. No separate schema/model generator exists; generated dist stays ignored.
-- No local Qdrant mapping exists. The upstream stricter schema-option behavior
-  and unchanged type-only mapping are documented in DESIGN.md.
-
-The initial update did not inspect the existing ignored `.env.local` and therefore
-ran only local tests. Steven subsequently explicitly authorized using that file;
-the development-service follow-up below supersedes the initial live-test omission.
-Supporting production deployment, search quality, load/failure coverage and billing
-readiness remain unverified. No merge, deployment or publication was performed.
-
-## SDK 0.7.0 review follow-up: rerank token redaction (2026-10-04)
-
-The PR review identified a valid protocol regression: a credential equal to
-`applied` rewrote `data.rerank.status` to `[REDACTED]`. The new CLI regression
-failed before the fix, while the negative test for arbitrary metadata passed.
-The output sanitizer now preserves exact known status/provider/model/criteria
-version/reason tokens only at their contract paths. Unknown server values,
-resolvedModel and document content retain credential redaction; whole metadata
-paths are not exempted because the SDK accepts open strings for several fields.
-
-On Node 24.15.0, lint, typecheck, version checks, all 71 source tests and all 46
-separately installed CLI tests passed with no failures or skips. The two new
-cases cover every fixed token and a one-character credential, JSON and human
-output, docsUrl results, nested document maps and unknown provider/model/reason
-values. Repeated builds remain identical for all 21 generated files. This is a
-local redaction fix; the prior development-service evidence remains applicable
-to SDK API behavior and was not rerun with synthetic colliding credentials.
-
-## SDK 0.7.0 authenticated development follow-up (2026-10-04)
-
-Tested the built CLI runtime from `b6b9cbc` (CLI `0.1.2-dev.1`, SDK `0.7.0`) on
-Node 24.15.0 with the expanded live harness in this PR. Target:
-`bench-recall` at `https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`.
-The existing `.env.local` supplied the API key and `LAMBDADB_RUN_LIVE_TESTS=1`.
-A `node --env-file=.env.local` launcher mapped `LAMBDADB_BASE_URL`,
-`LAMBDADB_PROJECT_NAME` and `LAMBDADB_PROJECT_API_KEY` in memory to the CLI's
-`LAMBDADB_ENDPOINT`, `LAMBDADB_PROJECT` and `LAMBDADB_API_KEY`, and set
-`LAMBDADB_LIVE_CONFIRM_PROJECT` to the same authorized project. It ran
-`node --test --test-reporter=tap test/live/cli.test.mjs` without printing credentials.
-
-- Doctor and temporary collection creation passed. All 49 fixed presets were
-  accepted as individual text field configurations, and describe metadata
-  preserved each requested configuration. The keyword analyzer and keyword field
-  type remained distinct. Unpopulated preset fields test acceptance/metadata,
-  not language-specific indexing or search quality.
-- Two ordinary documents with 3 MiB payloads each and one bulk document were
-  accepted at about 4 seconds. Committed query and fetch contents for all three
-  passed at about 100 seconds, without a pending-write overlay.
-- Match-all facet-only and document-plus-facet queries returned exact expected
-  document counts and category buckets.
-- Default reranking and custom criteria returned `applied`, candidate/scored
-  counts of three, expected provider/model and criteria-version markers, finite
-  final evaluation and retrieval scores, descending final-score order and matching
-  maxScore. Null rerank retained ordinary results with no rerank metadata or
-  retrievalScore. All three expected IDs were preserved. These checks exercised
-  the CLI, its published SDK and the designated development service, with no Jev key.
-- The final run passed in 102.4 seconds. Cleanup was accepted and get returned
-  HTTP 404 for `cli-smoke-571da4e7-2402-4652-aa81-c44824964fdc`.
-- Two earlier attempts failed only in newly added test metadata assertions:
-  first the response's collection wrapper was missed, then the server-added
-  default id index was compared as an unexpected field. The harness now compares
-  every requested index configuration while allowing server-added fields.
-  Both temporary collections were deleted and HTTP 404 confirmed:
-  `cli-smoke-ec8b9c3f-c94f-4c4a-bfff-3dad2a24516b` and
-  `cli-smoke-7fed413e-cbe5-4e15-a7dc-e334a3fac537`.
-- Every attempt verified `.env.local` remained byte-for-byte unchanged using
-  SHA-256. Temporary local files were removed by the harness. Lint, typecheck
-  and diff checks passed after the harness changes.
-
-This is bounded development evidence. The live harness does not inspect wire
-responses to prove docsUrl selection, induce provider fallback or load failures,
-measure ranking quality, test dense vector reranking, or establish production
-feature deployment or billing readiness. Local transport regressions cover
-zero/precision/ties, docsUrl metadata, independent vector k and failure handling.
-
-## CLI 0.1.1 publication
-
-On 2026-09-29, [release v0.1.1](https://github.com/lambdadb/lambdadb-cli/releases/tag/v0.1.1)
-was published from tag/main commit `80a4c10628ad2ad4ebf8d90a01924d27dcb4a790`,
-whose tree equals candidate `0291970`. The
-[release workflow](https://github.com/lambdadb/lambdadb-cli/actions/runs/36557578359)
-passed on attempt 1, including Node 22/24 and exact publication-artifact tests.
-
-- Registry reads confirmed `latest=0.1.1`, `dev=0.1.1-dev.10`. The stable tarball
-  matches the final tested candidate byte-for-byte. SHA-512 integrity:
-  `sha512-5eYw4NSCDPQrJZVDwfAg9cPtdSuipPy6/FzTj08EDn94/rRJEFLb0Meih2hzbzOSqvNLOn+MJD8Ij0a+nXe99A==`.
-- Decoded provenance identifies the same release commit, `refs/tags/v0.1.1`,
-  `.github/workflows/publish.yaml`, run `36557578359`, attempt 1; its subject digest
-  matches the downloaded tarball. Stable metadata does not expose `gitHead`, so
-  source identity is established by provenance and the artifact comparison.
-- A clean unqualified consumer install selected CLI 0.1.1 and SDK 0.6.0 and passed
-  all 40 CLI contracts. `npm audit signatures` verified all four packages' registry
-  signatures and three packages' attestations. No publication was retried.
-- Credentialed live evidence remains the installed-candidate smoke below; no new
-  live call was needed for byte-identical published runtime/package contents.
-- The prepared Homebrew formula pins the verified 0.1.1 tarball and lockfile from
-  `80a4c10628ad2ad4ebf8d90a01924d27dcb4a790`. Local macOS checks passed formula
-  style, installation, `brew test`, managed Node selection and all 40 installed
-  CLI contracts. The harness removed its temporary installation and tap. Remote
-  tap publication and upgrade behavior await the separate tap PR.
-
-Post-release development changes preserve this release and start `0.1.2-dev.1`.
-The registry-verification change passes 13 focused tests, including the real script
-entrypoint with a synthetic npm command: successful publication plus timeout exits
-0 with an Actions warning/summary, while failed publication exits 1. Transient
-read failures remain bounded; authentication, schema and artifact mismatches fail.
-No live publication was used to test that change. The full local suite passed 65
-tests and the installed-package suite passed 40.
-
-## CLI 0.1.1 release preparation
-
-On 2026-09-29, candidate commit `6f7bb23` prepared stable metadata from develop
-`a65a2b198b484d2e4da4ca2ff438779806a87cce`. Runtime source, tests, examples,
-SDK dependency and workflows are unchanged from that reviewed integration.
-Version 0.1.1 is a release candidate here, not an npm publication claim.
-
-- Clean candidate checkout on Node.js 24.15.0: `npm ci`, lint, version metadata,
-  typecheck and all 64 tests passed. Explicit release preflight with
-  `RELEASE_TAG=v0.1.1 RELEASE_PRERELEASE=false` passed without creating a tag.
-- Packed the candidate once and ran `npm run test:package -- <tarball>`:
-  all 40 installed-CLI tests passed. A separate consumer installation verified
-  CLI 0.1.1, SDK 0.6.0 and a LICENSE identical to the source.
-- Installed-tarball live smoke used the designated `bench-recall` development
-  project at `https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai` with
-  the existing secure in-memory environment mapping. The candidate tarball had
-  SHA-512 integrity
-  `sha512-1MRwAbI7kbAPKNUz2L+rd/F2tIF2I3r3wdI4kpsBNE84Y13TUYwTQF+gIDB95tvByyR1RdUi30ncH699rBhu9w==`.
-- Doctor, creation with `chinese` and `english`, ordinary/bulk acceptance, exact
-  committed query/fetch contents, match-all facet-only and document+facet buckets
-  passed in 61.0 seconds. Temporary Collection
-  `cli-smoke-a2894dda-50a2-4fa9-b320-fcc142bdf1fb` was deleted and verified by a
-  subsequent HTTP 404. No existing Collection or index was changed.
-- This evidence-only addition changes packaged documentation after that smoke.
-  The final promotion tarball is checked again by the installed-package suite;
-  runtime source, compiled JavaScript, examples and package metadata must remain
-  identical to the live-tested candidate. The promotion PR records the final
-  artifact digest and CI results separately.
-
-Publication remains a separate step after main promotion and authorization.
-The release workflow will rebuild and verify its own exact publication tarball.
-The current Homebrew formula still selects published 0.1.0; updating it requires
-verified 0.1.1 registry metadata and immutable release-commit checksums.
-
-## SDK 0.6.0 integration
-
-Validated on 2026-09-29 from an isolated worktree based on develop
-`513af6e4d262edd380013c86d51a20aad16274d7`, with Node.js 24.15.0 and the
-published `@functional-systems/lambdadb@0.6.0` dependency. npm `latest` was 0.6.0.
-The release tag resolves to `491d01e0eb54bd135823fab79cbe32396ec03691`;
-installed query schema, analyzer schema and client facade source files matched
-that tag byte-for-byte. The lockfile records the published package integrity.
-
-- `npm ci`, `npm run lint`, `npm run check:version`, `npm run typecheck`: passed.
-- `npm test`: 64 tests passed, including subprocess requests through the actual
-  SDK for facet-only success, zero without facets rejection, omitted query,
-  malformed facets, bounds, null/default preservation and ref consistency.
-- `npm run test:package`: 40 installed-tarball CLI tests passed. Facet metadata,
-  inline documents and `docsUrl` downloads survive JSON output; signed URLs and
-  credentials remain excluded. Arbitrary facet-name redaction preserves collisions;
-  short credentials do not change fixed bucket/result schema keys.
-- All 16 analyzer names passed create serialization and reached the loopback API;
-  unknown/case-mismatched names failed before HTTP. Omitted, empty and duplicate
-  analyzer lists were forwarded unchanged. This does not mean the live server
-  accepts duplicate names.
-- `git diff --check`: passed. GitHub Node 22/24 CI results are reported on the PR;
-  the local run above used Node 24 only.
-
-The credentialed live harness (`node --test test/live/cli.test.mjs`, after build)
-used the previously designated development endpoint
-`https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`, project `bench-recall`.
-An in-memory launcher used the existing ignored `.env.local` mapping documented
-below; no credential file was changed or copied into the worktree.
-
-- Doctor and temporary Collection creation with `english` and new `chinese`
-  analyzers succeeded. Two ordinary 3 MiB writes and one bulk write were accepted
-  at about 2 seconds in the final run.
-- Committed query/fetch contents for all three documents were verified at about
-  87 seconds in the final run, without a pending-write overlay.
-- Both checked-in facet examples passed: omitted-query `size: 0` returned no
-  documents; document+facet returned three. Both returned exact category buckets
-  `database: 2` and `developer-tools: 1` through CLI JSON output.
-- The final test passed in 87.8 seconds after the short-credential output fix.
-  Temporary Collection `cli-smoke-dcd3b60a-b65d-4126-b6e4-ad78904f4eb3` was deleted
-  and a subsequent SDK get returned HTTP 404. The earlier integration smoke also
-  passed in 53.6 seconds; its temporary Collection
-  `cli-smoke-e9f9a7c8-babf-4a62-a8d8-48bfa4683f66` was likewise deleted with 404
-  verification.
-
-This proves the bounded sample against one supporting development deployment,
-not production availability, all analyzer language behavior or old-index support.
-No existing Collection/index was migrated. No CLI merge, tag, GitHub Release or
-npm publication was performed as part of this integration.
-
-## Development publication after Homebrew documentation
-
-On 2026-09-19, the [develop push workflow](https://github.com/lambdadb/lambdadb-cli/actions/runs/35434313586)
-completed successfully for `225dec37f48dd5603af7fe777c1eb24d9041188c`, the merge of
-PR #8. Node 22/24 validation and tests of the exact publication tarball passed.
-The publish job accepted `0.1.1-dev.8`, retried registry reads during propagation,
-and finished with `result=published` without repeating the write.
-
-Subsequent read-only registry checks confirmed `dev=0.1.1-dev.8`,
-`latest=0.1.0`, and the package's `gitHead` matching the merge commit. Decoded
-provenance metadata identifies that same commit, `refs/heads/develop`,
-`.github/workflows/publish.yaml` and workflow attempt 1; its subject digest matches
-registry integrity. This metadata inspection did not repeat signature verification,
-a consumer installation or the LambdaDB live smoke.
-
-## Homebrew publication
-
-On 2026-09-19, the public
-[lambdadb/homebrew-tap](https://github.com/lambdadb/homebrew-tap) was initialized at
-`ca84485149d7915131df686b4c68868d47c00cd5`. Its CLI formula is byte-for-byte equal
-to the formula merged in CLI PR #7 (`3eefaaa`), selecting stable `0.1.0` and the
-release lockfile at `c8d389f`.
-
-- Local macOS 26.4 arm64 / Homebrew 6.0.17 verification used the tap's new harness
-  in both local and remote modes. Remote mode invoked
-  `brew install --formula lambdadb/tap/lambdadb-cli` with no existing tap or CLI,
-  exercising automatic public tap discovery and formula-specific trust.
-- Installation, downloaded formula equality, Homebrew style, `brew test`, version
-  `0.1.0`, JSON configuration with mode 0600, and the wrapper with an unusable
-  ambient Node all passed. No trust checks were disabled.
-- Test installations and taps were removed; existing taps, fnm Node and CLI
-  `.env.local` were preserved. Dependencies and caches remain installed.
-- [Tap CI](https://github.com/lambdadb/homebrew-tap/actions/runs/35433096180)
-  passed remote installation, style, smoke, runtime selection and cleanup on both
-  macOS 15 and Ubuntu 24.04 at the published tap commit.
-- Simulated existing CLI, broken prefix executable symlink and existing remote
-  tap cases stopped the new harness before mutations; invalid mode returned 2
-  without invoking Homebrew.
-- Tap main protection requires both installation checks, one approving review,
-  resolved conversations and an up-to-date base; force pushes and deletion are
-  disabled. Organization-level default-branch rules also apply.
-
-This publishes an installation path for the existing stable artifact, without a
-new npm release or another LambdaDB live-service smoke. Upgrade behavior between
-two distinct stable versions remains unverified until the next formula update.
-The earlier 35-contract Homebrew verification below used the same stable artifact;
-this tap publication ran the installation smoke checks rather than repeating it.
-
-## Homebrew preparation
-
-On 2026-09-19, macOS 26.4 arm64 / Homebrew 6.0.17 installed stable `0.1.0`
-from `packaging/homebrew/Formula/lambdadb-cli.rb` through a disposable local tap.
-The npm tarball SHA-256 was checked against the downloaded bytes, whose SHA-512
-matched registry integrity. The release lockfile resource is fixed to `c8d389f`.
-
-- Homebrew formula style, installation and `brew test` passed. The formula checks
-  CLI version, JSON configuration and mode 0600 without contacting a service.
-- A deliberately unusable `node` placed first on ambient PATH did not prevent
-  the installed wrapper from running; it selected Homebrew's Node 24.21.0.
-- All 35 CLI contracts from the stable release commit passed against the installed
-  Homebrew package. The harness uses the release commit's tests/examples so future
-  development-only features do not create false failures for an older stable CLI.
-- Simulated existing-formula and broken-executable-symlink cases both stopped the
-  harness before any installation or tap mutation.
-- The temporary CLI installation and tap were removed. The final harness disables
-  automatic dependency removal and cleanup, and does not persist developer mode.
-  Homebrew dependencies/caches remain: installing Node also upgraded the required
-  ca-certificates, openssl@3, readline, sqlite and xz packages during initial setup.
-  The existing fnm-managed Node and global npm CLI were not replaced.
-- Normal clean-install, lint, version/type checks, 59 source tests and 35 npm
-  installed-package contracts also passed. CLI runtime and SDK dependencies did
-  not change.
-
-macOS/Linux Homebrew CI is configured separately from runtime CI; current PR job
-results are the authority for hosted-runner coverage. This initial preparation
-did not create the public tap; publication evidence is recorded above. No new npm
-version was published by the preparation. LambdaDB service smoke was not repeated for
-this packaging-only change; the stable candidate evidence below remains separate.
-
-## Stable 0.1.0 publication
-
-On 2026-09-19, [GitHub Release v0.1.0](https://github.com/lambdadb/lambdadb-cli/releases/tag/v0.1.0)
-and its annotated tag identified main commit
-`c8d389f3264d641ae6eafc7737fedd5ea046ddf1`. Its tree exactly matches the reviewed
-candidate `658ea7ccdce535bbd4f2e29ae06813e2f2e716e2` from PR #5, including the
-runtime verified by the installed-candidate development smoke below. That smoke
-was not repeated because release contents were unchanged.
-
-The [release-event workflow](https://github.com/lambdadb/lambdadb-cli/actions/runs/35421226854)
-passed on attempt 1. Node 22/24 validation and exact-tarball contracts passed;
-Trusted Publishing published `0.1.0` on `latest` at 04:27:06 UTC. Version/tag
-metadata and attestation metadata propagated separately. Verification retried
-reads only; no second publication or workflow rerun was needed.
-
-A clean, unqualified npm install with an isolated cache resolved to `0.1.0` and
-passed version/help plus all 35 CLI contract tests. The downloaded tarball digest,
-installed lockfile integrity and provenance subject matched. The decoded SLSA
-provenance identified this repository, `.github/workflows/publish.yaml`,
-`refs/tags/v0.1.0`, release commit `c8d389f` and workflow attempt 1.
-`npm audit signatures` reported no invalid or missing signatures.
-At verification, `latest=0.1.0` and `dev=0.1.0-dev.5`; the stable release replaced
-the initial bootstrap latest tag without changing the dev channel.
-
-The subsequent main-to-develop synchronization sets `0.1.1-dev.1` as the next
-base. Its local clean install, lint, types, version checks, 59 source tests and
-35 installed-package contracts passed. This maintenance change updates versions
-and documentation only; CLI runtime, dependencies, tests and workflows match
-`v0.1.0`. Its PR validation does not itself publish a new dev version.
-
-## Stable 0.1.0 candidate
-
-Prepared `release/0.1.0` from reviewed develop commit
-`c8947e96308283ac593e712716f0732121bb1958`. Relative to that commit, only package
-versions and release documents change. Runtime source, dependencies, tests,
-examples and workflows are unchanged. This records the pre-publication candidate.
-
-On Node.js 24.15.0 / npm 11.12.1, a new worktree passed `npm ci`, lint, type
-checking, all 59 source tests, and 35 installed-package contract tests.
-`RELEASE_TAG=v0.1.0 RELEASE_PRERELEASE=false npm run check:version -- --release`
-passed and selected `latest`. CI results for the final commit belong to the
-promotion PR; local success is not a substitute for those required checks.
-
-A newly packed `0.1.0` tarball was installed into an isolated temporary consumer.
-Its executable reported `0.1.0` and was selected with `LAMBDADB_TEST_CLI` while
-running `node --test --test-reporter=tap test/live/cli.test.mjs`. The launcher used
-the maintainer's existing opt-in development settings and the in-memory variable
-mapping described below; it did not change or copy `.env.local` into the worktree.
-Target: `bench-recall` at
-`https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`.
-
-- Doctor and temporary collection creation passed within the first second.
-- Two ordinary documents with 3 MiB payloads each and one bulk document were
-  accepted at approximately 3 seconds. Acceptance did not imply search visibility.
-- Query and ID fetch on `branch:main` returned all three documents with exact
-  contents at approximately 69 seconds, without a pending-write overlay.
-- The live test passed in 69.6 seconds. Cleanup was accepted, then an independent
-  SDK get returned `ResourceNotFoundError` for
-  `cli-smoke-b854ef72-1448-42ae-8c9a-ce5c0441f104`, confirming its absence.
-- The source `.env.local` was byte-for-byte unchanged, and the temporary consumer
-  and tarball were removed. No key or signed URL is recorded here.
-
-Only documentation was updated after this live run; the final package is checked
-again with the installed-package contracts. This is development-service evidence
-for a bounded sample, not a latency/readiness guarantee or live tag/alias coverage.
-The stable npm artifact and release-event workflow were not verified at candidate
-preparation time; the subsequent publication evidence is recorded above.
-
-## Public bootstrap and OIDC publication
-
-On 2026-09-18, the maintainer authorized the public bootstrap
-[`0.1.0-dev.1`](https://www.npmjs.com/package/@functional-systems/lambdadb-cli/v/0.1.0-dev.1)
-from reviewed develop commit `f2397ba`. Clean installation, lint, types and
-53 source tests passed; the exact tarball passed 35 installed-CLI tests and its
-published integrity matched. This local publication has no GitHub provenance.
-
-After npm Trusted Publisher setup and repository variable activation, the
-[develop push workflow, attempt 2](https://github.com/lambdadb/lambdadb-cli/actions/runs/35340278724/attempts/2)
-passed Node 22/24 validation, prepared `0.1.0-dev.4`, tested its exact tarball,
-and published with OIDC and provenance. The old six-attempt verification loop
-exhausted its five 5-second sleeps before npm metadata propagation completed, so
-the job failed after a successful publication. Read-only checks subsequently
-confirmed the artifact. The
-[protected rerun, attempt 3](https://github.com/lambdadb/lambdadb-cli/actions/runs/35340278724/attempts/3)
-completed with `result=already-published`, proving the same source/artifact was
-recognized without a second registry write.
-
-The public [`0.1.0-dev.4` provenance](https://registry.npmjs.org/-/npm/v1/attestations/@functional-systems%2flambdadb-cli@0.1.0-dev.4)
-identifies `lambdadb/lambdadb-cli`, `.github/workflows/publish.yaml`,
-`refs/heads/develop`, source commit `f2397ba102c67f1af9ed5ae0b8b8d47c06da7d4b`,
-and workflow attempt 2. Registry `gitHead`, downloaded tarball digest, installed
-lockfile integrity and provenance subject agreed. A clean `@dev` installation
-using an isolated npm cache passed version/help and all 35 CLI contract tests
-with the generated package metadata. `npm audit signatures` verified four package
-signatures and three attestations, including the CLI. No credentials were added
-to GitHub validation jobs or repository files.
-
-At this verification, `dev=0.1.0-dev.4` and `latest=0.1.0-dev.1`. Bootstrap had
-created both tags despite `--tag dev`; a removal attempt for `latest` returned
-HTTP 400. The automated dev publication left `latest` unchanged. This observation
-does not establish a general version-count rule for npm. Bootstrap and OIDC setup
-were complete at that stage; main promotion and stable publication followed as
-recorded above.
-
-## Publication verification hardening
-
-The follow-up script uses a five-minute monotonic verification budget, 15-second
-maximum read subprocesses and 10-second maximum polling sleeps. It retries only
-missing/lagging metadata and transient post-publication reads. It reports an
-accepted write separately from `published-verification-pending`, which still exits
-nonzero. Authentication, malformed responses and source/artifact mismatches fail
-closed. No late response can satisfy the deadline, and no verification path
-repeats the registry write.
-
-Deterministic tests exercise 150/180-second metadata/tag propagation, five-minute
-expiration, read time consuming the budget, capped final I/O/sleep, late success,
-temporary HTTP/transport failures and permanent errors. The changed publication
-logic passed all 59 source tests (12 release-automation tests), lint, type checks
-and version checks on Node 24.15.0; the installed tarball passed 35 CLI contracts.
-A read-only probe used the actual npm CLI and new bounded-read flags to verify
-the existing `0.1.0-dev.4` manifest/tag. Only its preflight and accepted write were
-simulated; the probe performed no registry mutation.
-
-The [ordinary develop merge workflow](https://github.com/lambdadb/lambdadb-cli/actions/runs/35344153978)
-published `0.1.0-dev.5` from `c8947e96308283ac593e712716f0732121bb1958`
-on attempt 1. Node 22/24 validation and exact-artifact tests passed. Publication
-succeeded at 12:22:09 UTC; registry verification completed at 12:25:10 UTC with
-`result=published`. The approximately 181-second propagation delay required only
-read retries, with no manual rerun or repeated publication.
-
-A fresh npm `@dev` install with an isolated cache passed all 35 CLI contract tests,
-version/help, tarball/lockfile integrity checks and npm signature verification.
-Provenance identified the expected repository, develop ref, workflow, source
-commit and attempt 1. `dev=0.1.0-dev.5`; `latest` remained `0.1.0-dev.1`.
-These registry checks are separate from the development-service smoke below.
-
-## Initial local automation validation (before publication)
-
-On Node.js 24.15.0, lint, type checking and all 46 source tests passed, including
-six release-automation tests. The tests cover deterministic first-parent version
-numbers, numeric channel ordering, stale commits, registry lookup failures,
-artifact/commit conflicts, identical reruns and uncertain publication outcomes.
-Registry writes use an injected fake npm runner; no package was published.
-A real read-only npm lookup also confirmed the structured E404 handling.
-
-A temporary local Git repository and bare remote exercised the actual prepare
-command, including rejected PR/wrong-commit contexts and stale remote heads.
-A separate clean source copy generated `0.1.0-dev.2`, installed locked
-dependencies, packed and installed the generated artifact, and passed all 35
-installed-CLI contracts. Its source `gitHead` and license metadata were preserved
-and its Git HEAD did not change. The normal `0.1.0-dev.1` artifact also passed
-all 35 installed-CLI contracts.
-
-`actionlint` 1.7.12 accepted the combined CI/publishing workflow. The existing
-`Node.js 22` and `Node.js 24` check names are retained. Publishing needs both jobs
-and is restricted to explicit releases or opted-in develop pushes. The repository
-variable enabling automatic dev publication was not set during this validation.
-At that stage, OIDC authentication, registry writes, provenance and GitHub job
-ordering were unverified. The later publication evidence above supersedes that
-limitation; it does not turn these original local tests into live registry writes.
-
-## Authenticated development-project smoke
-
-Tested the CLI from develop commit `a040f53` (version `0.1.0-dev.1`, SDK `0.5.1`)
-with Node.js 24.15.0 against the development target supplied by the maintainer in
-the primary checkout's ignored `.env.local`. The designated development target is:
-
-- Endpoint: `https://internal-dev-aws-apne2-v3-c05a2b5d492a.lambdadb.ai`
-- Project: `bench-recall`
-
-The API key is excluded from this record. The file used `LAMBDADB_BASE_URL`,
-`LAMBDADB_PROJECT_NAME`, `LAMBDADB_PROJECT_API_KEY` and an explicit
-`LAMBDADB_RUN_LIVE_TESTS=1`. An in-memory launcher mapped the first three to
-`LAMBDADB_ENDPOINT`, `LAMBDADB_PROJECT`, `LAMBDADB_API_KEY`, and supplied the
-confirmed development project as `LAMBDADB_LIVE_CONFIRM_PROJECT` for this run.
-Neither the file nor the CLI's environment-variable contract was changed.
-
-After `npm run build`, the launcher ran `node --test test/live/cli.test.mjs`:
-
-- The original 75-second query observation window failed after writes were
-  accepted. This was a failed smoke, not proof of a service or CLI defect.
-- The follow-up harness allows up to 300 seconds per read stage, matching the
-  SDK live smoke's committed-document observation window. It reports elapsed
-  time and matched-document counts without printing responses or credentials.
-- The follow-up run passed: doctor and creation completed within the first
-  second; two ordinary writes (3 MiB payload each) and one bulk write were
-  accepted at about 2 seconds. Query returned all three expected documents with
-  exact contents at about 65 seconds, and ID fetch verified them at about 66
-  seconds. Both reads selected `branch:main` without a pending-write overlay.
-- Cleanup was accepted for both temporary collections. Subsequent SDK get calls
-  returned `ResourceNotFoundError` for
-  `cli-smoke-9797d164-abb3-4a7e-ae26-f5ae251ad72c` and
-  `cli-smoke-b926fd65-c371-462f-9727-9d5d9edecc1c`, confirming their absence.
-
-The two runs demonstrate variable observation time; 300 seconds is a test budget,
-not a readiness or latency guarantee. Full content was verified for a combined
-response above 6 MiB, but the CLI harness did not inspect the service's wire
-response to prove selection of `docsUrl`. Live tag/alias reads remain unverified.
-
-### Review follow-up: enforce the observation budget
-
-The harness now uses a monotonic deadline, caps CLI and child-process timeouts
-to the remaining budget, and limits the final polling sleep. It checks the
-deadline after each observation, including content comparison, before accepting
-success. Process termination and event-loop scheduling can delay reporting, but
-a response at or after the deadline cannot pass the stage.
-
-On Node.js 24.15.0, lint, type checking and all 53 source tests passed. Seven new
-deterministic tests cover timely success, late success/failure, success exactly
-at the deadline, a capped final sleep, an in-flight timeout and an earlier fatal
-error. The regression reproduces a response arriving at 318 seconds and rejects
-it; the final request receives only the remaining 2-second budget.
-The packed and separately installed CLI also passed all 35 contract tests.
-
-Re-ran `node --test test/live/cli.test.mjs` with this review fix in the working
-tree based on `6642eeb`, using the same designated endpoint/project and launcher
-above. The CLI runtime source remained identical to `a040f53`; SDK version was
-`0.5.1`. Doctor and collection creation passed within the first second, ordinary
-and bulk writes were accepted at about 3 seconds, and exact query/fetch contents
-passed at about 91 seconds (test duration: 91.231 seconds). Both reads used
-`branch:main`. Cleanup was accepted for
-`cli-smoke-3cb15833-414d-4265-9b57-70df164b28ef`; a subsequent SDK get returned
-`ResourceNotFoundError`, confirming absence. Deadline boundary cases were
-verified deterministically above, not by inducing a live service timeout.
-
-## 0.1.0-dev.1 release preparation
-
-Revalidated on 2026-09-18 in a fresh worktree based on develop commit `52e7aa2`
-with Node.js 24.15.0 and npm 11.12.1. `npm ci`, lint, type checking and version
-checks passed. Source tests passed (40), and the separately installed tarball's
-CLI tests passed (35), with no failures or skips. The version assertion now
-compares the executable against package metadata instead of a fixed version.
-
-`RELEASE_TAG=v0.1.0-dev.1 RELEASE_PRERELEASE=true npm run check:version -- --release`
-passed and selected the `dev` channel. This checks release metadata only; it does
-not publish or establish main ancestry, npm permissions or live-service behavior.
-At the initial release-preparation stage no development connection settings were
-supplied, so live validation was not attempted then. See the later smoke above.
-
-## Completed local validation
-
-| Check | Result |
-| --- | --- |
-| `npm run typecheck` | Passed with TypeScript 5.8.3. |
-| `npm run lint` and `npm run check:version` | Passed; package and lockfile metadata agree. |
-| `npm test` on Node.js 24.15.0 | 40 tests passed, zero failed or skipped. |
-| Prior CLI MVP suite on Node.js 22.23.2 via `npm exec --yes --package=node@22 -- node --test test/*.test.mjs` | 40 tests passed, zero failed or skipped; new automation tests are recorded separately above. |
-| Fresh temporary source directory: `npm ci`, `npm run build`, CLI version | Passed without the development checkout's node_modules or dist. |
-| `npm run test:package`: pack, inspect inventory, install in a temporary consumer, invoke executable version/help and repeat CLI contracts | 35 installed-CLI tests passed on Node 24. Package was never published. |
-| `actionlint` on CI and publish workflows | Passed. This is static validation, not an OIDC publication. |
-| Live command without opt-in | Expected nonzero exit before network calls; not counted as a live test pass. |
-| npm install dependency audit | Zero vulnerabilities reported at installation time. |
-| Apache-2.0 license packaging | Official license text from apache.org is included byte-for-byte as `package/LICENSE` in the built tarball; package.json and lockfile license metadata agree. |
-| Reference SDK status and specified sbrain file status | Unchanged. |
-
-Tests use the actual published SDK and CLI subprocesses; they do not replace the
-SDK with a fake client for transport tests. The workflow unit tests separately
-cover byte accounting and cancellation between batches.
-
-## Behavior exercised
-
-- Configure, doctor, create (HTTP 201), JSONL upsert (HTTP 202), query, fetch,
-  describe and paginated collection listing.
-- Exact branch/tag/alias request bodies and result targets on query and fetch;
-  write branch forwarded on ordinary and bulk requests.
-- SDK pagination with an opaque continuation token.
-- SDK bulk metadata, signed headers, upload body and finalization; no project
-  credential sent to object storage.
-- SDK offloaded query/fetch downloads for each ref kind, including 1 MiB returned
-  document bodies; consumed signed URLs omitted from CLI output.
-- Complete-file preflight rejects malformed later JSONL lines, invalid IDs,
-  empty files and oversized individual batch entries before any request.
-- Invalid query fields, conflicting file/flag refs, invalid query size and
-  branch-only consistency constraints rejected before requests.
-- Query `size: null` normalized to SDK omission and branch consistency forwarded.
-- Accepted/rejected/not-attempted batch counts and physical line ranges.
-- Disconnect, HTTP 503, malformed 202 and timeout classified as unknown writes
-  without mutation retries, preserving previously accepted batches.
-- Failed bulk upload not retried or finalized; reported conservatively as unknown
-  because the SDK helper has no structured error stage.
-- Shared deadline applies to API calls, signed downloads and signed uploads;
-  cancellation between batches leaves unsent rows not attempted.
-- Actual SIGINT/SIGTERM during an active second write preserve the accepted first
-  batch, mark the active batch unknown and leave the third batch unattempted.
-- Release channels and rejection of version drift, invalid tags/flags, private
-  publication and missing license metadata.
-- Empty results succeed; 401/403 authentication failures, other API failures and
-  invalid response schemas fail; creation conflicts differ from uncertain writes.
-- All documented exit codes, JSON parseability, stdout/stderr separation, human
-  summaries, help and version.
-- Flag/environment/saved-config precedence, explicit missing config, missing
-  selected key, custom key-variable names and POSIX config file mode 0600.
-- SDK debug suppression and credential redaction in errors and returned documents,
-  including secret strings containing JSON structural characters.
-- Short keys preserve JSON envelope fields, CLI command/state/check tokens,
-  ref kinds and error codes. Arbitrary document/index/tag keys and variable
-  values remain redacted; human output retains its structured data section.
-- Redacted map-key collisions preserve all entries and unchanged field names.
-  Query/fetch regressions cover both input orders, multiple colliding renamed
-  keys, preexisting suffixes and nested maps; collection metadata covers tag and
-  index-config collisions. Generated suffixes cannot overwrite reserved names.
-- Exactly 100,000 documents are accepted by input preflight; excess rows are
-  rejected with their physical line number, excluding blank lines from the count.
-  CRLF and a final line without a newline are covered.
-- A 3 MB file of one million small documents exits 2 before any API request under
-  a 128 MiB V8 heap limit. The pre-fix reader reproduced heap exhaustion with
-  that input. Ten million blank lines followed by invalid JSON also exit 2 within
-  that heap limit, preserving the final line number without a split array.
-
-## Source and public-contract inspection
-
-Verified npm latest and installed SDK 0.5.1; checked current local SDK source and
-published OpenAPI/quickstart. Evidence revisions and compatibility findings are
-in [DESIGN.md](DESIGN.md). These checks establish the chosen contract, not the
-deployment state of an individual LambdaDB endpoint.
-
-## Not performed / not established
-
-- The live smoke verifies one designated development target and a bounded sample;
-  it does not establish general query/index readiness, search ranking, managed
-  embedding behavior, exhaustive signed-storage policies or production-scale
-  performance. Credentials were not searched for in other repositories.
-- No proof of exactly-once delivery, durable resume, or a committed-through write
-  receipt. Those capabilities are not implemented.
-- No Windows-specific permission or signal validation. POSIX modes and injected
-  SIGINT/SIGTERM were checked on macOS; CI targets Linux.
-- Mock large-response tests cover SDK routing/credential separation at 1 MiB;
-  the live sample verifies combined content above 6 MiB. Neither is an exhaustive
-  memory/size stress test. Imports buffer a maximum 64 MiB source file
-  and retain at most 100,000 documents. Complex documents can still use much more
-  memory than their serialized size; the count limit is not a process memory cap.
-
-## Requirements for live verification
-
-Follow the explicit settings and `npm run test:live` procedure in
-[RELEASING.md](RELEASING.md#explicit-live-smoke). The authenticated main-branch
-flow and cleanup have been exercised as recorded above. For non-main refs,
-supply existing development branch/tag/alias
-targets whose expected contents are known. Record evidence without keys or signed
-URLs. Temporary collection deletion is performed by the harness through the SDK,
-not exposed as a CLI command.
-
-During expanded validation, a second-write timeout hung repeatedly in the source
-and installed package. An explicit timer alone was insufficient; binding the
-command signal through SDK HTTPClient fetch dispatch resolved the observed local
-regressions. See [DESIGN.md](DESIGN.md) for the compatibility boundary; the exact
-upstream cause is not claimed. Commits, pushes and CI runs are recorded in Git
-and the pull request; local checks do not imply service or publication success.