Release retrospective: v19.0.0

[flyingrobots]

Release retrospective: v19.0.0

Released

v19.0.0 is the intentional graph-first-to-runtime application break and the completion of the emergency storage-ownership/performance pivot.

Released scope:

• Runtime as the sole root runtime value, with type-only Lane, Intent, Observer, Observation, Reading, Evidence, Tick, settlement, and Receipt contracts.
• Wesley-generated validated Intent and Observer SDKs; lazy single-execution Observations; runtime-backed strands and revalidated settlement plans.
• Explicit /advanced, /diagnostics, /charts, and /testing surfaces, with storage, legacy, and browser absent from the package export map.
• git-cas 6.5.5 ownership of CAS objects, pages, bundles, cache lifecycle, retention, and physical diagnostics; WARP-owned duplicate caches and raw object writers removed or prohibited by CI.
• A one-shot git-warp-v18-to-v19 retained-substrate migrator under scripts/v18-to-v19/, with non-mutating disposable rehearsal, compare-and-swap promotion, additive recovery refs, and authentic published-v18 fixtures. No v18 compatibility runtime remains in product code.
• Versioned cold/warm/incremental materialization and oversized Observer gates, plus an authentic published-v18 versus migrated-v19 retained-read comparison.
• CLI and MCP vocabulary aligned with Runtime/Lane/Intent/Observation/Reading/Receipt.
• npm, JSR, GitHub Release, annotated tag, release workflow, and exact-main performance evidence aligned on one commit.

Not released

• The v18 graph-first application API is not retained or hidden under another subpath. Consumers must migrate application code deliberately.
• Think itself was not switched to v19 as part of this package release. Its graph-first application layer and durable repository should be migrated together in an isolated rehearsal before the live writer resumes.
• Common-basis braid validation, WARP DRIVE, WARP-TTD, offline bundles, live Echo/git-warp suffix exchange, and generic legacy compatibility remain outside v19 as stated by the milestone.
• Universal performance claims were not made. The release witnesses bounded, named workloads and memory envelopes.

Plan vs actual

Planned thesis: ship a boring Runtime/Lane/Intent/Observer/Observation/Reading/Receipt boundary and major performance improvements from corrected git-cas ownership.

Actual release: matched the thesis, but only after two release-blocking discoveries changed the implementation path.

1. A real Think probe proved that npm linking alone could not bridge retained v18 blob-backed state to the v19 asset-handle boundary. The release stopped, and the normal one-shot migration model was implemented under scripts/v18-to-v19/ instead of preserving legacy runtime code.
2. The first authentic v18 comparison showed only about 4–5% lower wall time. The release stopped again. Observation basis preparation, pinned coordinate optic reuse, and immutable captured-basis caching were optimized until the exact merged-main gate proved a material improvement.

The final train spans 290 commits, 51 first-parent merge commits, and 1,177 changed files (87,737 insertions, 47,334 deletions) from v18.2.1 to v19.0.0. The v19 milestone closed 56 issues and reached zero open release items before tagging.

Evidence

• Tag: v19.0.0
• Annotated tag object: a0769b73d6427d3449888eb833fe4b711b30fe51
• Tag/merge commit: f93f57fe5f241f326473b553e2a907f293339db1
• Release PR: Release v19.0.0 #802
• GitHub Release: https://github.com/git-stunts/git-warp/releases/tag/v19.0.0
• Guarded tag workflow: https://github.com/git-stunts/git-warp/actions/runs/30253840233
• Registry release workflow: https://github.com/git-stunts/git-warp/actions/runs/30254867021
• Exact merged-main performance workflow: https://github.com/git-stunts/git-warp/actions/runs/30253840231
• Performance goalpost and detailed witness: Goalpost: v19 bounded-memory and CPU performance contract #758
• Public API contract: Goalpost: v19 Runtime, Lane, and Observation public API #712 / PR Enforce one generated v19 vocabulary contract #799
• Migration implementation: PR Add one-shot v18-to-v19 retained-substrate migration #788
• Authentic migrated-read performance optimization: Prove and improve migrated v18-to-v19 read performance #800 / PR Gate v19 on migrated retained-read speed #801
• npm: @git-stunts/git-warp@19.0.0, dist-tag latest
• npm tarball: https://registry.npmjs.org/@git-stunts/git-warp/-/git-warp-19.0.0.tgz
• npm integrity: sha512-YiQOjaR5gDr3FMINdtBoNqrEpCX7HQLVJO18151fcJET3CpL9qw3fToEp9nAwDQWavg4Xp7wN4wh7ViLlYigQg==
• JSR npm-compat: @jsr/git-stunts__git-warp@19.0.0, dist-tag latest
• JSR tarball: https://npm.jsr.io/~/11/@jsr/git-stunts__git-warp/19.0.0.tgz
• JSR integrity: sha512-fXfbEREsrPIwAIjIbWn1DJZlZOFGeXnPrp6KNA3+Ml4Le23s5qskizjk6Mf+/GolNSp0RC5D+eZ+swmGuUfgLg==

Performance witness

Exact merged-main, Linux x64, Node 22.23.1, Git 2.54.0:

• Authentic approximately 2 MiB, 18-patch retained-v18 fixture; one-shot migration reported separately at 3,281.1 ms.
• Sixteen retained properties at one pinned coordinate, five measured cold and warm runs per runtime after warmup.
• Cold: 1,251.4 ms v18 → 861.9 ms v19 (31.1% lower wall); 657.5 ms → 525.7 ms operation CPU (20.1% lower).
• Warm: 1,239.0 ms v18 → 841.4 ms v19 (32.1% lower wall); 652.3 ms → 513.6 ms operation CPU (21.3% lower).
• Git commands: 322 → 172 (46.6% fewer) in both scenarios.
• Oversized proof: all 128 Readings / 256 MiB logical property data under a 64 MiB old-space cap, 35.2 MiB peak heap, zero whole-index scans, zero full materializations, hostile full-materialization control OOM as required.

Went well

• Real consumer evidence stopped the release instead of being dismissed as a linking problem.
• Legacy migration code stayed in the one-shot script boundary; the v19 product surface remained clean.
• The fixture stayed small: approximately 2 MiB, not a checked-in copy of a live or gigabyte-scale mind.
• The first weak benchmark result was accepted as RED evidence and drove a real runtime optimization rather than a weaker threshold.
• Exact-main performance, release preflight, tag identity, registry publication, and post-publication import/export checks all agreed.
• No amend, rebase, force, draft PR, or legacy compatibility shortcut was used.

Improve next time

• Curate the release synopsis before generated PR history. The generated v19 list surfaced superseded openWarp/Timeline prototypes and required a post-publication release-note prepend. Tracked by Publish curated release notes before generated PR history #803.
• Put the goalpost reference into the release PR body at creation time. PR Release v19.0.0 #802 had one metadata-only failed check before Related: #758 was added. Tracked by Reference the goalpost when opening release PRs #804.
• Keep the disposable registry smoke's working directory explicit. The post-publish smoke initially installed into /tmp; exact birth-time checks allowed safe removal of only the created paths, and no repository or user data was affected.
• Preserve the intentionally redundant final-main and tag verification gates, but investigate artifact reuse only if it can retain exact-SHA, package-content, and registry-publish guarantees.

Fallout issues

• Publish curated release notes before generated PR history #803: Publish curated release notes before generated PR history
• Reference the goalpost when opening release PRs #804: Reference the goalpost when opening release PRs
• COOL IDEA: generate post-release retrospective scaffolds #690: Generate post-release retrospective scaffolds
• COOL IDEA: validate the release state machine #691: Validate the release state machine

Next release recommendation

Do not start a v20 feature train immediately. First migrate one real graph-first consumer—Think is the obvious candidate—through both layers in an isolated copy:

1. stop all writers and preserve the live repository;
2. run the packaged v18-to-v19 rehearsal and inspect its recovery plan;
3. migrate the application from graph-first imports/patch callbacks to generated Intents and bounded Observers;
4. promote the retained substrate only after disposable reads, writes, receipts, doctor, and recovery tests pass;
5. resume the live writer only after compare-and-swap preconditions are rechecked.

Treat that as consumer validation, not a reason to reintroduce v18 runtime code into git-warp.

[flyingrobots]

Retrospective filed with release, performance, migration, registry, and fallout evidence. Closing this record as complete before any v20 train begins.

[coderabbitai]

🔗 Related PRs

#674 - Prepare v18.1.0 release [merged]
#710 - Release 18.2.1 [merged]
#788 - Add one-shot v18-to-v19 retained-substrate migration [merged]
#801 - Gate v19 on migrated retained-read speed [open]
#802 - Release v19.0.0 [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!