wdbx

Ecosystem Parallel Completion Implementation Plan

For agentic workers: use superpowers:subagent-driven-development for owned implementation slices and superpowers:executing-plans for coordinator work. Independent repository work proceeds concurrently. Steps use checkboxes.

Goal: Complete WDBX, ABI, Abbey, Abbey Bot, MLAI/Quesar and WebPress’s release requirements, including the WebPress refactor, measured demonstrations, full brand assets, and rendered animated trailers.

Architecture: Preserve the existing canonical repositories and public contracts. WDBX owns durable episodic semantics; ABI and Abbey consume sibling crates, while Abbey Bot retains its independent contract implementations. Qualify independent changes concurrently, then verify dependent integration.

Tech stack: Repository-pinned Rust, Zig, Bun/TypeScript, React, native Apple tooling, and existing animation/rendering tools. Each repository’s instructions and executable gate take precedence over copied commands.

Spec: Donald’s approved in-thread ecosystem scope and immediate multi-repo, multi-agent direction on 2026-10-03, together with each repository’s approved designs. This plan does not turn reference subsystems into production claims.

Global constraints

• Four agent slots total, including the coordinator. Delegation must remain within that limit. Assign disjoint file ownership and preserve others’ work.
• Work in canonical checkouts; do not switch a shared checkout’s branch.
• Do not commit, push, publish, deploy, or restart live services without the corresponding authorization. Completion of local work does not imply rollout.
• Use scratch data for tests. Never use private live stores as test fixtures.
• No new production dependency without approval. Use relevant installed skills; discover additional skills only for a concrete capability gap.
• Keep WDBX’s frozen v2 and v3 commitment contracts compatible. Specimen Studio remains a separate workspace with a pinned git dependency.
• Keep heavy gates resource-bounded. Do not run Abbey’s Rust and Zig gates simultaneously or build against actively changing shared dependency files.
• Existing dirty work is a baseline, not this team’s accomplishment.
• No calendar delay is built into this plan. Start available independent work immediately; synchronize on tests and interfaces, not estimated weeks.

Review focus

1. Recovered state after interrupted writes must remain valid and attributable.
2. Consent, deletion, quarantine, and contradictory evidence must retain their documented meaning across consumers, persistence and exported projections.
3. Cancellation, unavailable providers, and partial output must terminate or recover explicitly without silent success.
4. Saved projects, accepted proposals, undo and publication must preserve identity and revision authority across reopen and export.
5. Public numerical and visual claims must match reproducible evidence, while private data and restricted assets remain excluded from launch artifacts.

Task 1: Establish and preserve current state

• Record each repository’s HEAD, branch, dirty paths, active owners, current spec/plan/ledger and authoritative gate.
• Map every explicit release requirement to source, test and runtime acceptance evidence; classify Current, Partial, Proposed or Blocked.
• Verify MLAI’s recorded integration into Quesar; avoid duplicating a retired standalone application.
• Select concrete independent slices; record ownership and test commands before editing. Inspect current files rather than trusting old task ticks.

Task 2: WDBX and Specimen Studio

Owner: coordinator initially. Interfaces: existing canonical episode encoder/store/receipts and unchanged consumer contracts.

• Qualify the current cross-language test changes and run bash tools/check.sh on the pinned toolchain.
• Reconcile stale contract documentation with implemented signatures, memory edges and gateway boundaries using source/test citations.
• Close remaining approved episode/retrieval/recovery requirements with regression-first slices, preserving existing golden commitments.
• Qualify Studio separately with bun run check, bun run lint, bun run test:native, bun run check:native and applicable browser/native journeys. Resume the P2 shell plan from live evidence.
• Run optional FHE checks for affected feature work; record their separate outcome without claiming production cryptography or multi-host evidence.

Task 3: ABI and Abbey

Owner: runtime lane. Interfaces: existing sibling WDBX crate APIs, gateway/CLI/MCP contracts, provider and assistant interfaces.

• Recover the live training/runtime and assistant plans, preserving their currently dirty implementation and instruction files.
• Complete the first unqualified approved slice in each repo with explicit source ownership and regression evidence.
• Exercise orchestration, memory provenance, cancellation, provider failure, restart and streaming acceptance required by the release checklist.
• Run ABI’s ./tools/check.sh, Abbey’s ./check.sh, and retained Zig qualification separately as required by their current instructions.
• Recheck consumers after the current additive WDBX interface change: ABI full gate and Abbey’s 12 focused WDBX consumer tests passed.

Task 4: Abbey Bot and MLAI/Quesar

Owner: application lane. Interfaces: existing adapter and learning contracts; Quesar authentication, publication and sidecar boundaries.

• Recover Learning Tasks 4–5 without overwriting existing erasure/reset work.
• Close remaining required consent, routing, memory, voice, reconnect and shutdown behavior; qualify the bot’s source and live service separately.
• Resume Quesar’s current approved plans from actual source/review evidence.
• Verify public and authenticated journeys, persistence, responsive states, provider configuration, server build and static publication.
• Run each repo’s authoritative gate and relevant runtime acceptance tests.

Task 5: WebPress refactor and complete authoring journey

Owner: WebPress lane. Interfaces: existing project model, proposal host, accepted checkpoints, rendering, publication and runnable export contracts.

• Preserve and qualify the current refactor before rebuilding any completed task. Resolve the first remaining runtime or source failure from evidence.
• Complete create/edit/preview/save/reopen/publish/export acceptance, including undo, revision safety, malformed input and recovery.
• Verify remaining required CMS, templates, offline, collaboration and commerce behavior against approved scope and fixtures.
• Run WebPress’s current full gate and browser/native acceptance separately.

Task 6: Full brand system and rendered trailers

Owner: release/media lane, scheduled in a free slot alongside implementation.

• Inventory existing brand assets, typography, licenses, motion components, cinematic work and webpress/examples/abbey-trailer.
• Complete ecosystem and individual product identities: vector assets, colors, typography, icons, motion rules, accessible usage and export kit.
• Build on existing launch material; produce editable timelines, source assets and reproducible render commands, not only HTML storyboards.
• Render one 90–120 second ecosystem trailer and 45–60 second product cuts; deliver 15/30 second derivatives, thumbnails, captions and audio.
• Export 4K masters and 1080p web versions, with deliberate 16:9 and 9:16 compositions. Inspect actual rendered frames, pacing, audio and captions.
• Use real product behavior and reproducible measurements. Label conceptual illustrations accurately and exclude private or unlicensed material.

Task 7: Recursive review, integration and completion

• For each slice: reproduce failure, implement, focused test, independent standards review and independent spec review, repair, then repo gate.
• Pin the review baseline and source diff; never review an empty range as evidence for uncommitted work. Report standards/spec findings separately.
• Retest cross-repo contracts after integration, then run essential journeys.
• Record benchmark hardware, revisions, data, configuration, repetitions, correctness, latency distribution, throughput and resource use.
• Audit every release requirement and launch artifact against direct evidence. Keep unresolved requirements open; do not equate a local gate with full product, provider, hardware or deployment acceptance.

Execution record

The companion progress file records evidence, task ownership, commands, tool handles, decisions and remaining work. It is updated as results arrive. The program is complete only when all tasks above satisfy their acceptance scope.

Qualified slices and immediate continuation

These bounded deliveries do not close the wider unchecked release requirements. The evidence index is ../../verification/2026-10-03-delivery-index.md.

Slice	Exact seam / deliverable	Gate / next acceptance	State
Core ML fixture portability	abi/crates/abi-gpu/native/metal_dot.swift	21 focused tests plus ABI full gate; independent review	Current local
Studio component behavior	wdbx/specimen-studio/components/ui/, hooks/use-mobile.ts, component tests and browser command	web/native/lint/Clippy/browser plus five instrumented native tests passed; packaging/signing remains	Current local / Partial runtime
Bot offline quality	abbey-bot/src/brain/quality_evaluation.rs, DQN import/replay, CLI, 100-case fixture	fix3 regressions, independent reviews and source-stable strict gate pass (2,087 Rust tests)	Current local / Partial release
WebPress saved publication	webpress/src/export/publication.ts and tests	full gate, actual browser-saved ZIP and independent review passed	Current local
Quesar films	quesar.cloud/scripts/export-films.ts, static byte validation, narrated-film player	six complete movies; source/static/browser gates and review passed	Current local
Bot yanked-package repair	abbey-bot/Cargo.lock, yoke-derive 0.8.4	strict 2,087 Rust / 339 Python / Swift 12+16 gate; five accepted vulnerabilities and three unmaintained warnings remain	Current local / Partial release
Quesar claims correction	src/routes/quesar.tsx, OG source/card and versioned brand kit	744 unit / 129 static pages / 226 browser; independent Standards+Spec PASS	Current local
Quesar brand distribution	quesar.cloud/scripts/package-brand.ts, pinned brand inputs and local archive	reproducibility,71 payload hashes, 54 references, full 744-test gate and review passed	Current local
Abbey concept delivery	webpress/examples/abbey-trailer/render.py, retained metadata regression	30/15 second cuts in 1080p, native 4K and recomposed portrait pass	Current concept-format set
ABI callback cancellation	abi-agent-host/src/host.rs and local contract tests	Six RED regressions; 25 focused pass; independent reviews; fresh ABI 996 Rust +144 Python gate	Current local
Ecosystem concept film	quesar.cloud/notes/launch/ecosystem-trailer-2026-10-03/	90/30/15-second 1080p movies, original score, captions, full decode and Chromium playback pass; receipts, checksums and desktop/mobile review-gallery playback pass	Current concept set / Partial format program
WDBX/ABI resolver read	Canonical per-edge digest, additive gateway field 9, CLI text/JSON	Independent reviews, WDBX 647 and ABI 988 Rust + 144 Python full gates passed	Current local
Release timing observations	Existing 500-vector scalar HNSW harness, five runs	All runs retained; source/binary hashes and resource use; recall unmeasured	Current local measurement
Abbey desktop	abbey/desktop/check.sh with frozen source and locked dependencies	49 Rust + 13 Python desktop gate plus seven native Routes UI scenarios and 21 AX snapshots pass; broader desktop journeys separate	Current local / Partial runtime

The installed Zig compiler used for qualification is 0.17.0-dev.2338+b46a7f3a2, newer than the documented minimum 0.17.0-dev.2320+1e770dbef; exact historical compiler qualification remains unverified. Bot’s recorded 2026-10-03 07:10 UTC service observation was ready; that historical observation predates the current candidate. The installed binary differs from the qualified candidate, so source completion is not rollout proof.

Closed since the prior checkpoint: Bot fix3 strict gate, all six Abbey concept exports, per-edge WDBX/ABI resolver provenance with fresh full gates, and five release-profile timing observations. Abbey native Routes UI also passed seven scenarios. Studio instrumented native journeys passed five tests/six scenarios. WebPress native authored publication reached a verified isolated build but awaits ownership resolution after external edits to its acceptance window. All implementation and review work remains capped at four active agents and resource-bounded local commands.

The admitted abi-agent-host callback cancellation slice is now locally qualified: six regressions proved the defect, 25 focused tests pass, both independent reviews pass, and the root full gate passes 996 Rust plus 144 Python on frozen source. The wider runtime/training plan is not reopened merely because historical checklist marks lag verified implementation. WebPress native authored export remains Partial pending ownership of the externally edited acceptance window. Remaining product stories, ecosystem 4K/portrait formats, full brand program and platform/provider/release acceptance remain explicit unfinished work.

This site is open source. Improve this page.