Files
alkblobs/tasks/changelog-and-readme.md
T
glm-5.3-flash 825b0e3187 docs(release): CHANGELOG.md [0.1.0] and the root README (wave 12's changelog-and-readme)
- CHANGELOG.md: the family Keep-a-Changelog form (alkcall/alkstore
  precedent), the prose-led initial-release entry summarizing the
  shipped surface - Store facade, pins/batches, GC protocol, five
  engines, fleet GC, the feature-gated ops module, fuzzing - with
  link refs into docs/architecture/decisions/ and a Verified block
- README.md: tagline, the is/is-not posture, three quick-start
  examples, the ADR-013 \u00a74 feature table, the engine table, pointers;
  written honestly - the sweep's refuse-to-run posture shown as-is,
  read_range's slice digest stated as out-of-band (no invented read
  gate), no claims beyond shipped constructors
- tests/readme_examples.rs (kept): the quick start's durable
  compile-check - each fence's call shape runs against the real API;
  the pg row feature-gated, runs against the harness env
- tasks/changelog-and-readme.md: status completed, Notes/Summary
  filled

Verified: cargo test 209 (default) and 460 (--all-features), both
against the live pg harness clippy both configs -D warnings clean,
fmt clean, cargo doc warning-free, fuzz corpus replay 7/7
2026-10-16 11:09:08 +00:00

6.2 KiB

id, name, status, depends_on, scope, risk, impact, level
id name status depends_on scope risk impact level
changelog-and-readme CHANGELOG.md (v0.1.0) and the root README completed
crate-docs
docs-consistency-review
feature-graph-verification
moderate low project implementation

Description

The release-facing prose, written after the docs and matrix work so it describes verified reality.

CHANGELOG.md — the first entry, [0.1.0], in the family's Keep-a-Changelog form (precedent: alkcall/CHANGELOG.md, alkstore/CHANGELOG.md — the two-line header citing Keep a Changelog and Semantic Versioning, a dated prose-led entry, link references at the foot). Content: the initial release's full surface, written from the shipped crate — the Store facade (put/get/stat/range, pins and batches, sweep), the five engines (kv: sqlite default / postgres / mem; large: local / pg-lo), fleet GC (DB-backed pins, sweep lease, delete arbitration), the feature-gated ops module (alkcall-backed op family, ACL mapping, pin tokens), the git-family canonical hash and key encoding, and the fuzzing posture (the sidecar targets + the corpus-replay gate). The alkstore 0.1.0 entry is the closest shape precedent (an initial release's prose summary); the date is the release-prep date. Link refs point into docs/architecture/ ADRs where the entry names a decision.

README.md — written last, honestly (the plan's words). Precedent: alkcall/README.md, alktty/README.md, alksocks/README.md. Shape:

  • Tagline (the manifest description's territory: content-addressed blob storage, multi-hash, pluggable backends, transport-agnostic).
  • What it is / is not: the store layer only — no transport, no protocol in-core; ops rides alkcall behind a default-off feature (ADR-001's posture, stated plainly).
  • Quick start: construct a store (the default dual-tier shape and one feature-gated variant), put/get round trip, a pin + sweep taste. Examples must compile against the real API — verify them (a doc-test or a manual compile against the built crate), not from memory.
  • The feature table (ADR-013 §4's graph: default features, what each engine feature pulls).
  • The engine table (tier / engine / feature / durability posture — one honest line each; pg-lo's named deltas included).
  • Pointers: docs/architecture/ as the spec/ADR record, CHANGELOG, license (MIT OR Apache-2.0).

No overselling: no claims the crate doesn't make (no benchmarks beyond the recorded POC numbers' posture, no "production-ready" beyond what the gates prove).

Acceptance Criteria

  • CHANGELOG.md exists with the [0.1.0] entry in the family format; the surface summary matches the shipped crate
  • README.md exists with tagline / posture / quick start / feature + engine tables / pointers
  • README examples verified against the real API (compile-checked)
  • Feature and engine tables match ADR-013 §4 and the shipped constructor surface
  • No claims beyond shipped reality (the honesty bar the wave-12 review audits)

References

  • /workspace/@alkdev/alkcall/CHANGELOG.md and README.md (format and shape precedent)
  • /workspace/@alkdev/alkstore/CHANGELOG.md (the 0.1.0 initial-release entry shape)
  • docs/architecture/decisions/013-…-feature-graph.md §4 (the feature table's source)
  • docs/architecture/backends-and-dispatch.md (the engine table's source)

Notes

  • CHANGELOG.md — the [0.1.0] - 2026-10-16 entry in the alkcall / alkstore Keep-a-Changelog form (two-line header, prose-led entry, link references at the foot pointing into docs/architecture/decisions/). The prose covers the full shipped surface: the Store facade, pins and batches, the GC protocol, the five engines (durability postures stated per engine), fleet GC, the feature-gated ops module, and the fuzzing posture.
  • README.md — tagline (the manifest description's territory), the is/is-not posture (store layer only; ops default-off, ADR-001), three quick-start examples (mem round trip + pin/sweep taste, the default dual-tier sqlite shape, the feature-gated kv_only_postgres variant), the ADR-013 §4 feature table, the engine table (tier/engine/feature/durability), and pointers (architecture docs, CHANGELOG, fuzz README, license). Every claim cross-checked against the shipped constructors, the facade docs, ADR-013 §4, and backends-and-dispatch.md — two honesty corrections found during that pass are worth recording: the sweep's refuse-to-run posture (zero registered liveness sources aborts with NoLivenessSources, so the quick start shows that rather than an invented reap), and read_range's slice digest is carried out-of-band by the caller, not an in-core read-gate (verification-on-get fires on the kv tier's materialized reads; ADR-006).
  • tests/readme_examples.rs (kept) — the durable compile-check: each fence's call shape runs against the real API; the pg row is feature-gated and runs against the harness env, server-less it is absent like the conformance pg columns. Implementation lesson recorded there: a standalone tokio_postgres::connect client must have its Connection future spawned or every query hangs (the first draft hung exactly that; the engine's own harness does it right).
  • Cargo manifest untouched — readme = "README.md" and the exclude list's fuzz/ belong to publish-prep, per the wave-12 decomposition.
  • The interrupted first pg-column run left three schema debris mints (alkblobs_readme_pg_kv_*); dropped manually against the harness.

Summary

Both release-facing prose files landed as the wave-12 gate's final authoring step. CHANGELOG.md carries the family-format [0.1.0] entry summarizing the shipped crate with ADR link refs (ADR-001/002/ 003/005/006/007/008/009/012/013/014/015/016/017) and a Verified block recording the real numbers. README.md describes verified reality at the honesty bar: no capabilities beyond shipped constructors, no benchmark claims beyond the recorded POC posture, quick-start examples executable against the crate. Verified: the full block — cargo test 209 (default features) and 460 under --all-features, both against the live pg harness; both clippy configurations -D warnings clean; cargo fmt --check clean; cargo doc --no-deps warning-free; the fuzz corpus-replay gate 7/7.