Activity

Aigarth Cloud Evolution PEP registered with command centre (3 docs). Status: draft. Pending architecture review.

Doc added: Aigarth Cloud Evolution PEP — From Neural Networks to Evolving Useful Intelligence

Doc added: Aigarth Cloud Evolution PEP — Executive Summary

Doc added: Aigarth Cloud Evolution PEP — Roadmap Impact

Aigarth Cloud Evolution PEP registered with command centre (3 docs). Status: draft. Pending architecture review.

Phase 25 — Aigarth CLI (agent-grade) registered. 8 tasks, ~8 SP. Decisions: Node + npm install (single-binary deferred), API key + Qubic wallet binding, foundation + discovery + 1 execution path, runtime schema introspection via aigarth schema. Wraps the existing SDK. Distributed as @aigarth/cli. Design: docs/proposals/phase-25-cli.md. The agent-grade framing is the differentiator: JSON by default, versioned schemas, idempotency keys, exit code taxonomy, no interactive prompts.

Phase 24 — Hardware node presale (tier 1) — closed. 8/8 tasks, 8/8 SP. Live in code: node_reservations + qubic_usd_rates tables, full lifecycle (24.1); node_escrow double-entry ledger + 4 routes (24.2); usd-price-oracle 60s worker (24.3); qearn-lock wrapper with 5% APY linear accrual (24.4); /nodes marketing page with 4 GPT Image 2 product images, hero + tier 1 spec + 10-item FAQ + Hetzner/AWS/Aigarth compare table + schema.org Product/PreOrder (24.5); /dashboard/nodes reservations page (24.6); operator script for the confirm window with --dry-run and --id flags (24.7); delivery doc (24.8). Vitest: 37/37 passing. Typecheck clean across compute, sdk, web, dashboard. Deferred to Phase 25+ / 20.6: real QPI multisig (escrow table is the audit mirror with synthetic tx_hashes), real Qearn lock (currently linear accrual stub), auto-trigger worker (operator script is the manual path), tier 2 + tier 3 specs (slots in DB only), service-role check on the internal endpoints. Tier 1 USD-anchored: $599 final, $30 deposit (5% ratio), yield opt-in (default off), per-tier sequenced drops (1 → 2 → 3, 14-day windows after mainnet). Source mechanic: Otto One presale, adapted to QUBIC-native two-stage payment.

Phase 24 — Hardware node presale (tier 1) — closed. 8/8 tasks, 0.5/0.5 SP. Live in code: node_reservations + qubic_usd_rates tables, full lifecycle (24.1); node_escrow double-entry ledger + 4 routes (24.2); usd-price-oracle 60s worker (24.3); qearn-lock wrapper with 5% APY linear accrual (24.4); /nodes marketing page with 4 GPT Image 2 product images, hero + tier 1 spec + 10-item FAQ + Hetzner/AWS/Aigarth compare table + schema.org Product/PreOrder (24.5); /dashboard/nodes reservations page (24.6); operator script for the confirm window with --dry-run and --id flags (24.7); delivery doc (24.8). Vitest: 37/37 passing. Typecheck clean across compute, sdk, web, dashboard. Deferred to Phase 25+ / 20.6: real QPI multisig (escrow table is the audit mirror with synthetic tx_hashes), real Qearn lock (currently linear accrual stub), auto-trigger worker (operator script is the manual path), tier 2 + tier 3 specs (slots in DB only), service-role check on the internal endpoints. Tier 1 USD-anchored: $599 final, $30 deposit (5% ratio), yield opt-in (default off), per-tier sequenced drops (1 → 2 → 3, 14-day windows after mainnet). Source mechanic: Otto One presale, adapted to QUBIC-native two-stage payment.

Phase 24 — Hardware node presale (tier 1) registered. 8 tasks, ~8.5 SP. 24.1 (schema + routes) shipped this run; 27 vitest passing, SDK wired, typecheck clean. 24.2-24.8 in backlog. Design: docs/proposals/phase-24-hardware-presale.md. Source mechanic: Otto One presale, adapted to QUBIC-native two-stage payment with yield opt-in.

Phase 15 — Knowledge layer v1 — /blog + /tutorials + /academy surfaces on the dashboard. All three reuse the existing Doc store (SQLite) and a shared hand-rolled markdown renderer extracted from doc-detail.tsx. Markdown files in docs/{blog,tutorials,academy}/ are the source of truth; the dashboard reads them on each render so editing a file is reflected on the next page load (no re-seed step needed). Frontmatter (title / date / author / tags) is parsed by a small YAML-ish reader; bare comma-separated list values are supported for academy lessons. Sidebar nav gets Blog + Tutorials + Academy. Seeded content: 3 blog posts (existing trinary launch + tweet thread + new observability GA post), 2 tutorials (first-stake, first-ann), 2 academy paths (Getting Started with Aigarth, Building ANNs). 22/22 monorepo typecheck clean. v2 scope (deferred): certifications, quizzes + completion tracking, search, authoring UI, versioning.

Doc added: Observability GA: every service, every request, one Grafana

Doc added: Trinary Intelligence is live on Aigarth Cloud

Doc added: Trinary Intelligence launch — tweet thread

Doc added: Your first stake on Aigarth

Doc added: Your first ANN — build, train, deploy

Doc added: Getting Started with Aigarth

Doc added: Building ANNs

Phase 15 v1 content registered: 7 docs (3 blog, 2 tutorials, 2 academy paths)

Phase 15 — Knowledge layer registered. 1 task, ~3 SP. v1 ships /blog + /tutorials + /academy surfaces on the dashboard, all backed by the existing Doc store + markdown renderer. Certifications are v2.

Phase 14 — Governance (off-chain) v1 — proposals subsystem shipped. Schema (proposals + proposal_votes + 3 enums: proposalKind, proposalStatus, voteChoice), service (createProposal / castVote / recomputeTally / finalise / executeProposal / listProposals / getProposal), 6 REST endpoints under /v1/proposals. 26/26 vitest cases pass. Phase 14 + Phase 22 are the off-chain + on-chain halves of the governance system: Phase 14 surfaces what should happen, Phase 22 enforces that it happens via multi-sig. Binding proposal kinds (treasury-spend, ecosystem-grant, parameter-change) push to AigarthPool.submitTreasuryTransfer; general is non-binding. Voting rule: simple-majority of stake weight, quorum gate. Stake snapshot at vote time (stable if user later unstakes). Default quorum = max(100k QUBIC, 1% of proposer stake) — prevents the 'tiny-proposer tiny-quorum' attack. Self-vote forbidden. Default voting window 7 days, max 30. Monorepo typecheck 22/22 clean. Combined with Phase 22, the full off-chain → on-chain governance loop is exercisable end-to-end (simulator path; AIO pending).

Phase 13 — Observability v1 — /metrics on every service, AigarthPool + Qearn business metrics, starter Grafana dashboard. Shared package @aigarth/observability (hand-rolled Prometheus text format, no prom-client dep, < 3 KB per service). registerHttpMetrics(app, serviceName) wires: service_info gauge + http_requests_total counter + http_request_duration_seconds histogram + correlation ID via X-Request-Id. All 10 services instrumented: identity, ann, qubic, economy, billing, marketplace, tissue, dataset, compute, gateway. AigarthPool business metrics (ann service) subscribe to the pool's event feed and update gauges on PositionOpened / PositionClosed / RewardsClaimed / PoolPaused / PoolUnpaused. Qearn business metrics (qubic service) source from the mirror table on each /metrics scrape (the watcher is a separate process). infra/observability/aigarth-cloud-overview.json: starter Grafana dashboard with 6 panels (request rate, p95 latency, error rate, process memory, AigarthPool totals, Qearn events observed). 22/22 monorepo typecheck clean. Distributed tracing (OpenTelemetry → Jaeger/Tempo) and alert routing (PagerDuty/Opsgenie) are deferred to v2.

Phase 14 — Governance (off-chain) registered. 1 task, ~3 SP. v1: proposals + vote tally + finalize + treasury-spend execution. The on-chain multi-sig (M1+M2) is Phase 22 — Phase 14 is the policy surface that drives it.

Phase 13 — Observability registered. 1 task, ~3 SP. v1: shared @aigarth/observability package + /metrics on every service + a starter Grafana dashboard. Distributed tracing + alert routing are v2.

Phase 23 — Pre-mainnet gates (M3 + M4 + p3) — closed. 3/3 tasks, 6/6 SP. M3 rate limits: MAX_STAKE_PER_TX=1M QUBIC + MAX_STAKE_PER_USER_PER_EPOCH=5M QUBIC enforced in stakeForAnn. M4 circuit breaker: pause/unpause procedures (governance-only) with new service route + dashboard banner + Pool status StatCard. p3 Qearn watcher: services/qubic/src/workers/qearn-watcher.ts + qubic_qearn_events mirror + NATS publish for services/economy. P3 stake-attribution audit: vitest harness + standalone script (apps/dashboard/scripts/p3-stake-audit.ts) — verified PASS against the simulator with 10 random (user, ann) positions. Aigarthpool package: 95/95 tests pass (was 66/66 after Phase 22). On-chain QPI build is hardware-gated behind the AIO Dev Kit (Phase 20.6); design + simulator land now. Remaining pre-mainnet items: M5 (migration plan, ~1 SP, doc-heavy), M6 (insurance fund, ops), I1–I4 (internal audit, 1 week with reviewer), E1–E3 (external audit via Qubic Foundation, 3–4 weeks), 20.6 (testnet deploy, hardware-gated), P1 (daily review post-mainnet), P2 (TVL cap for the first 7 days).

Phase 23 — Pre-mainnet gates (M3 + M4 + p3) registered. 3 tasks, ~6 SP. Closes the pre-mainnet gates from docs/aigarthpool/audit-checklist.md that don't require external coordination. M5 (migration plan) and M6 (insurance fund) stay in the tracker as next steps; 20.6 (testnet deploy) + I1–I4 + E1–E3 are out of scope (hardware + reviewer time).

Phase 22 — Multi-sig treasury (M1+M2) — designed, simulated, and wired through the off-chain services + command centre. The audit checklist's blocker is unblocked: governance for setAnnSplits and a multi-sig treasury are now fully designed and exercised locally. Spec at docs/aigarthpool/governance.md. The C++/QPI side: AigarthPool.{h,cpp} now host initGovernance + treasury-transfer + signer-change procedures (each with submit/approve/execute, threshold = 1..N, PENDING_OP_TTL_EPOCHS=4), getGovernanceState + getSigners queries, 7 new events (GovernanceInitialized, TreasuryTransferSubmitted/Approved/Executed, SignerChangeSubmitted/Approved/Executed). setAnnSplits now accepts the ANN's creator OR a current signer (multi-sig fallback for M1). The TypeScript side: packages/aigarthpool/src/{types,simulator,client,index}.ts export the new governance surface. AigarthPoolSimulator mirrors every procedure with full event emission; 36 new vitest cases (66/66 package total, up from 30). services/ann adds 4 admin endpoints (POST /v1/aigarthpool/governance/{init,treasury-transfer,signer-change}, GET /v1/aigarthpool/governance/state — public read for the command centre). services/qubic watcher logs the new events generically. apps/dashboard gets /governance command centre page with signer set, pending-op cards, approval progress, endpoint health. Honest limit: QPI build is stubbed (AIO Dev Kit required); the design and simulator are tested, the on-chain version waits on hardware. Next: M3 (rate limits) and M4 (circuit breaker) — both cheap; M5/M6 are operational. The audit checklist's M1+M2 checkboxes are now ticked at 'designed' status.

Phase 22 created

Task p22-multisig-treasury created

Phase 22 — Multi-sig treasury (M1+M2) — designed, simulated, and wired through the off-chain services + command centre. The audit checklist's blocker is unblocked: governance for setAnnSplits and a multi-sig treasury are now fully designed and exercised locally. Spec at docs/aigarthpool/governance.md. The C++/QPI side: AigarthPool.{h,cpp} now host initGovernance + treasury-transfer + signer-change procedures (each with submit/approve/execute, threshold = 1..N, PENDING_OP_TTL_EPOCHS=4), getGovernanceState + getSigners queries, 7 new events (GovernanceInitialized, TreasuryTransferSubmitted/Approved/Executed, SignerChangeSubmitted/Approved/Executed). setAnnSplits now accepts the ANN's creator OR a current signer (multi-sig fallback for M1). The TypeScript side: packages/aigarthpool/src/{types,simulator,client,index}.ts export the new governance surface. AigarthPoolSimulator mirrors every procedure with full event emission; 36 new vitest cases (66/66 package total, up from 30). services/ann adds 4 admin endpoints (POST /v1/aigarthpool/governance/{init,treasury-transfer,signer-change}, GET /v1/aigarthpool/governance/state — public read for the command centre). services/qubic watcher logs the new events generically. apps/dashboard gets /governance command centre page with signer set, pending-op cards, approval progress, endpoint health. Honest limit: QPI build is stubbed (AIO Dev Kit required); the design and simulator are tested, the on-chain version waits on hardware. Next: M3 (rate limits) and M4 (circuit breaker) — both cheap; M5/M6 are operational. The audit checklist's M1+M2 checkboxes are now ticked at 'designed' status.

Phase 21 created

Task p21-metamask-snap-connect created

Phase 21 — MetaMask Qubic snap support — shipped. Option B (signTransaction as workaround) is end-to-end: services/identity has parseQubicTransaction + verifyQubicTransactionSignature (with reserved AIGARTH_AUTH_INPUT_TYPE 0x4147 / 'AG' to prevent piggybacking unrelated Qubic transactions), services/identity/src/services/walletAuth.ts has a discriminated WalletFinishSchema (kind: 'message' | 'transaction'). apps/web/lib/wallet/snap.ts gains buildQubicSelfTransfer (sender == receiver, amount = 0, inputType = 0x4147, input = canonical message) and signChallengeAsTransaction (which calls the snap's wallet_invokeSnap signTransaction RPC). apps/web/components/marketing/connect-qubic-wallet.tsx uses the new snap path end-to-end — the previous 'snap returns the address, then sign with the vault' workaround is no longer the primary path; submitFinish helper is split out and the signAndSubmit flow picks message vs transaction based on the path that produced the proof. invokeSnapSignMessage is reserved for Option A and throws a clear error pointing to Option B. Command centre (apps/dashboard/src/components/pages/wallet-auth.tsx) now shows: by-kind breakdown with side-by-side tiles (message vs transaction), snap_active_30d badge, and a transaction audit feed table (last 10 wallet-auth events with kind, address, label, IP hash, action). Bug fix: the result.reason reference in walletAuthFinish was outside its scope (would have thrown on the transaction kind path) — moved into the verifier blocks and returned in WalletFinishResult.verification.reason. Next: design the multi-sig treasury (audit-checklist M1+M2) as a separate phase before any testnet deploy of AigarthPool.

Phase 20 — 5 of 6 tasks done (20.1 AIO Dev Kit, 20.2 QPI contract, 20.3 splits procedure, 20.4 services integration, 20.5 watcher). 20.6 (testnet deploy + audit) stays in_progress — hardware-gated until AIO Dev Kit is available. Phase 20 → 92%.

Phase 20 — Aigarth on-chain settlement (AigarthPool) — substantively complete. Built for the on-chain contract, simulated locally until AIO Dev Kit is available. 20.1: AIO Qubic Dev Kit — docker-compose qubic-devnet profile (AIO image, AVX2 + 24GB RAM requirement documented, healthchecks), pnpm stack:up:devnet script, docs/aigarthpool/README.md. 20.2: AigarthPool QPI contract — services/aigarthpool/contract/AigarthPool.{h,cpp}, CMakeLists.txt, full on-chain spec: stakeForAnn, extendLock, unlock, claimRewards, setAnnSplits, plus all queries. events: PositionOpened, PositionClosed, RewardsClaimed, SplitsChanged. docs/aigarthpool/contract.md is the spec. 20.3: Splits procedure (applySplits) is part of the contract — 30/60/10 default, truncation rule documented (user absorbs remainder, contract never mints/burns). 20.4: workspace package @aigarth/aigarthpool with TypeScript simulator (default backend) + Qubic RPC client (qpi mode, hardware-gated, throws until Phase 20.6). 30 vitest cases for the simulator (procedures, queries, events, error codes, splits arithmetic, yield model). services/ann wired up: createAnn auto-registers with AigarthPool, new endpoints POST /v1/anns/:idOrSlug/{stake,extend,unlock,claim} + GET /v1/anns/:idOrSlug/position. 20.5: services/qubic worker aigarthpool-watcher subscribes to AigarthPool events and logs them to the audit log. 20.6: docs/aigarthpool/audit-checklist.md is the pre-flight gate (G1-G5 internal, I1-I4 code review, E1-E3 external audit, M1-M6 mainnet gates, P1-P3 post-mainnet) — ready to walk through once AIO lands. Test counts: ann 150/150, training 42/42, aigarthpool 30/30 (new). Monorepo typecheck 22/22 clean. Honest limit: qpi mode is stubbed (constructor throws) — full mainnet path requires AIO Dev Kit hardware. Next: governance + multi-sig treasury (M1+M2) as a separate phase before any testnet deploy.

Phase 18 — Off-Chain Economy — DONE. Final 10% shipped: the wallet_address join in settleRun. New services/identity endpoint GET /v1/internal/wallets/by-user/:userId (guarded by INTERNAL_TOKEN, returns verified + non-revoked wallet addresses for a user). New services/identity service helper listVerifiedWallets (filters on verified_at IS NOT NULL AND revoked_at IS NULL). New services/economy/src/lib/identity-client.ts with lookupWalletsForUser (typed fetch wrapper, 404 → empty array, 401 → IdentityLookupError, 5xx → IdentityLookupError). settleRun now (1) dedupes recipient userIds, (2) calls lookupWalletsForUser for each, (3) persists the resolved wallet_address on each payout_recipient row, (4) then broadcasts via services/qubic. New env var: INTERNAL_TOKEN (default 'internal_dev_token_change_me', 16+ chars) on services/identity and services/economy. Audit action 'payout.wallet_lookup_failed' for transport / token errors so dashboard surfaces are clean. Monorepo typecheck 21/21 clean. Splits smoke (38 assertions) still green. Phase 19 (Datasets & Training) and Phase 20 (Aigarth on-chain settlement) follow.

Phase 19 — Datasets & Training Orchestration — DONE. All 22 tasks complete (31/31 SP). Garden UX shipped (19A, 3.5 SP). Dataset service shipped (19B, 9.5 SP, services/dataset on port 7009, MinIO + multipart + connector framework + SDK). Training orchestration shipped (19C, 11 SP, services/training on port 7011, recipe catalog, compute bridge, SSE progress, auto-publish new ANN version). Feedback loops shipped (19D, 6 SP — outcomes, auto-retrain, A/B shadow, actual-vs-predicted Garden UI). Test counts: services/ann 150/150 (added 19 versionRouter), services/training 42/42, monorepo typecheck 21/21 clean. Brand voice: em-dash-free Garden copy, no abstract nouns, tagline visible. Closeout delivery report: docs/deliveries/phase-19-delivery.md (SP velocity, sub-phase breakdown, build time). Phase 20 (Aigarth on-chain settlement) registered for the next push — 6 tasks, 6 SP, build order: AIO Dev Kit → AigarthPool QPI contract → splits → services integration → watcher → testnet audit.

19D.4 — Garden UX: actual success rate next to predicted accuracy. AnnCard now renders a two-line accuracy cell: '94%' (predicted, lab benchmark) on top, 'actual 89% (1,247 calls)' underneath in muted text. The actual line only renders when the ANN has at least one labeled decision outcome (the /v1/anns/:slug/stats endpoint from 19D.1). When there are zero outcomes, the card keeps the pre-19D.4 single-line look so old ANNs don't suddenly look bare. SDK: new Anns.stats() method returning the snake_case shape (total_outcomes, success_rate, recent_success_rate, avg_confidence_in_label, last_recorded_at). Garden page now fetches stats for the top-6 ANNs in parallel, tolerates per-ANN failures (one bad row doesn't blank the whole card), and prefers ann.accuracy (0-100 lab benchmark) over ann.rating_average (0-5 community stars) for the predicted line. New fmtPct100 helper for the 0-100 scale. Phase 19 → 92% (28.5 of ~31 SP). Phase 20 (AigarthPool) registered separately.

Phase 19 → 92% (28.5 of ~31 SP). 19D.4 done. Phase 19 substantively complete — only the 0.5 SP Phase 19 closeout delivery report remains.

19D.3 — A/B shadow deployment shipped. New ann_deploy_mode enum (active/canary/shadow/deprecated) + deployMode + trafficWeight columns on ann_deployments. New shadowVersion + shadowEnvelope columns on ann_decisions (the shadow's envelope is stored for audit/replay). New service services/ann/src/services/versionRouter.ts: pickByWeight (deterministic weighted pick, defensive against negative/zero weights), pickServingDeployment (active+canary only, falls back to ANN's currentVersionId when no deployment), pickShadowDeployment (shadow only, 50/50 split to keep noise down). services/ann/src/services/trinary.ts decideAnn now consults the router — the served result comes from the weighted pick, and the shadow (if picked) runs in parallel, signs its own envelope, and stores both. Migration drizzle/0005_gorgeous_lockheed.sql. 19 new vitest cases (versionRouter.test.ts) covering pickByWeight distribution, active+canary picking, 50/50 shadow split, version-not-found fallback, status filter, weight edge cases. ann suite 150/150 green (was 131). Monorepo typecheck clean. Next: 19D.4 (Garden UX: show actual success rate + call count alongside predicted accuracy, 0.5 SP).

Phase 19 → 90% (28 of ~31 SP). 19D.3 done. Last bit: 19D.4 (Garden actual vs predicted UI, 0.5 SP).

19D.2 — Auto-retrain trigger shipped. New ann_retrain_configs + ann_retrain_events tables (per-ANN opt-in config: optIn, baselineSuccessRate, driftThreshold, outcomeFloor, cooldownHours, lastRetrainAt, preferredDatasetVersionId, notifyMode). 5 new endpoints: GET/PUT /v1/anns/:id/retrain-config, GET /v1/anns/:id/retrain-events, POST /v1/anns/:id/retrain (manual trigger, 60s rate-limit). 2 internal endpoints guarded by TRAINING_INTERNAL_TOKEN: POST /v1/internal/anns/retrain-scan (cron target), GET /v1/internal/anns/retrain-eligible. Service layer: getRetrainConfig, upsertRetrainConfig (owner-only, captures baseline on first opt-in), listRetrainEvents, triggerManualRetrain (60s rate-limit, calls services/training), runRetrainScan (walks opted-in ANNs, evaluates cooldown + drift + low-volume triggers, records events). Training service: AnnClient.triggerRetrain (stub + http), services/training/src/services/retrainTrigger.ts, workers/retrain-cron.ts (setInterval, default 6h, fires on start so devs don't wait). 21 new vitest cases (retrain.test.ts), ann suite 131/131 green, training suite 42/42 green (no regressions). Monorepo typecheck 21/21 clean. Phase 20 (Aigarth on-chain settlement) registered.

Phase 19 → 84% (26 of ~31 SP). 19D.2 done. 19D remaining: 19D.3 (A/B shadow deploy, 2 SP), 19D.4 (Garden actual vs predicted UI, 0.5 SP).

19D.1 — Decision outcome tracking shipped. New ann_decision_outcomes table (id, decision_id, ann_id, ann_version_id, outcome enum [success|failure|reverted|unknown], confidence_in_label, notes, recorded_by, source, outcome_at). 4 new endpoints: POST/GET /v1/anns/:id/decisions/:decisionId/outcome, GET /v1/anns/:id/outcomes, GET /v1/anns/:id/stats. Service layer: recordOutcome (idempotent upsert, 201 on first / 200 on subsequent), getOutcome, listOutcomesForAnn (paginated, filterable), computeOutcomeStats (successRate, recentSuccessRate last 100 graded, avgConfidenceInLabel). 30 new vitest cases, ann suite 110/110 green. Monorepo typecheck 21/21 clean. Phase 20 also registered (AigarthPool on-chain settlement, 6 tasks, 6 SP, not started).

Phase 19 → 77% (24 of ~31 SP). 19D.1 done. 19D remaining: 19D.2 (auto-retrain, 2 SP), 19D.3 (A/B shadow deploy, 2 SP), 19D.4 (Garden actual vs predicted, 0.5 SP). 19D.2 unblocks now that outcomes are tracked.

Phase 20 — On-Chain Settlement (AigarthPool) registered. 6 tasks, ~6 SP. Punted from Phase 19C so 19C could ship off-chain first. Build order: 20.1 (AIO Dev Kit) → 20.2 (AigarthPool contract) → 20.3 (splits) → 20.4 (services integration) → 20.5 (watcher) → 20.6 (testnet + audit). The contract is the only piece that turns the off-chain economy into a verifiable on-chain protocol.

19C — Training Orchestration fully shipped (5/5 remaining sub-tasks closed). New services/training on port 7011. 5 built-in recipes (mlp_classifier, cnn_classifier, text_classifier, gradient_boost_tabular, trinary_classifier). Recipe catalog + job submission + worker pool + compute bridge (stub + http modes) + SSE progress via NATS (in-memory fallback) + auto-publish new ANN version on success. 42 new vitest tests (recipes 12, jobs 13, computeBridge 8, progress 9). Monorepo typecheck 21/21 green. No regressions in dataset (18/18) or ann (80/80).

Phase 19 → 73% (22.5 of ~31 SP). 19C complete (6/6 sub-tasks, 11 SP). 19A + 19B + cross-cutting already done (10/10, 12.5 SP). Only 19D (Feedback loops, ~6 SP, 4 tasks) remains before Phase 19 closes. Next: 19D.1 (decision outcome tracking) — small, unblocks the rest of 19D.

Phase 19B (Dataset service) closed out. 6/6 tasks done. Shipped: services/dataset on port 7009, schema per ADR 004 (5 tables), MinIO upload pipeline with SHA-256 + schema sniffer, public catalog browse UI at /datasets, dataset detail page at /datasets/[slug], studio listing at /dashboard/datasets, marketing + dashboard nav updated, SDK Datasets resource (8 methods, multipart upload), and a minimal connector framework with 1 working kind (http_api). Roadmap updated. 20/20 typecheck green. 18/18 dataset tests pass. 80/80 ANN tests still green. Next: 19C.1 (services/training skeleton on port 7010). Full delivery: docs/deliveries/phase-19b-dataset-delivery.md

19B complete. 6/6 tasks done. Shipped: SDK Datasets resource (client.datasets.{list,catalog,retrieve,listVersions,create,update,changeStatus,uploadVersion}), public catalog browse UI at /datasets with detail page at /datasets/[slug], dataset studio page at /dashboard/datasets, plus a minimal connector framework (1 kind: http_api) with manual sync endpoint. 20/20 typecheck green. 18/18 dataset tests pass. 80/80 ANN tests still green. Next: 19C.1 (services/training skeleton on port 7010).

19B complete. 6/6 tasks done. Shipped: SDK Datasets resource (client.datasets.{list,catalog,retrieve,listVersions,create,update,changeStatus,uploadVersion}), public catalog browse UI at /datasets with detail page at /datasets/[slug], dataset studio page at /dashboard/datasets, plus a minimal connector framework (1 kind: http_api) with manual sync endpoint. 20/20 typecheck green. 18/18 dataset tests pass. 80/80 ANN tests still green. Next: 19C.1 (services/training skeleton on port 7010).

19B.1–3 (services/dataset skeleton, schema, MinIO upload pipeline) shipped. 3/3 tasks done. 20/20 monorepo typecheck green. 12/12 dataset tests pass. 80/80 ANN tests still green (no regressions). Next: 19B.4 (public catalog UI) + 19B.6 (SDK resource), or 19C.1 (services/training skeleton).

Phase 19 closeout. Status: in_progress at 24% (7.5 / 31 SP). 7 tasks done: 19A.1–4 Garden UX, 19C.3 Real LLM invocation, ADR 004 (dataset licensing), brand-voice rewrite of the Phase 3 Vision. 15 tasks deferred to next session. Roadmap updated (ROADMAP.md). 19/19 typecheck green. No regressions. Next session: 19B.1 + 19B.2 + 19B.3 (services/dataset).

Cross-cutting tasks done. ADR 004 (dataset ownership + licensing) and brand-voice rewrite of the Phase 3 Vision proposal both shipped. 2/2 tasks done.

19A Garden UX complete. 4/4 tasks done. New /dashboard/garden home view with 4 widget zones (ANNs, tissues, training queue placeholder, marketplace revenue). /dashboard now redirects to /dashboard/garden. Nav restructured — Garden is the new top-level item; existing pages remain at their paths.

19C.3 — Real LLM invocation in ANN /decide (close the trinary stub gap). Plug-in TrinaryBackend interface (stub + OpenAI-compatible). 38 new tests. Monorepo typecheck 19/19 green. All ANN tests 80/80 green.

Phase 19 — Datasets & Training Orchestration. 19C.3 done (real LLM invocation). Phase 19 status → in_progress at 10% (3 of 31 SP). Next: 19A.1–3 Garden UX (3 SP).

Doc added: Phase 3 Vision — Intelligence Foundry (proposal)

Doc added: Phase 3 Vision — Evaluation

Phase 19 — Datasets & Training Orchestration registered. 4 sub-phases (19A Garden UX, 19B Dataset registry, 19C Training orchestration, 19D Feedback loops), 20 tasks, ~25 SP. Build order: 19A (cheap win, recommended first sprint) → 19B → 19C → 19D. Composes with the existing tissue primitive. 2 docs (proposal + evaluation) registered.

Doc added: Services/economy — Checkpoint 1 (Phase 18 build state)

Doc added: Phase 18 ANN Catalog Seed (6 real ANNs)

Doc added: Stake-to-Access Landing Page (/stake-access)

Phase 18 — Off-Chain Economy marked in_progress at 90%. 12 tasks done. services/economy shipped (port 7008) with splits, contributors, locks, bundles, payouts. Qearn watcher built. 6-ANN catalog seeded. /stake-access marketing page live. ADR 002 (use Qearn as-is, do not clone) recorded. 38-assertion splits smoke green. ~15 SP remaining for full prod wiring (wallet_address join in settleRun, K12 sig verification, real TCP client).

Doc added: Original proposal — Aigarth Cloud Material Science Intelligence ANN

docs/use-cases/material-science-proposal.md — original proposal registered

Doc added: ADR 002 — Staking Contract Strategy (Use Qearn As-Is)

Doc added: ADR 002 — Staking Contract Strategy (Use Qearn As-Is)

Task created: Qearn watcher: observe Qearn locks, drive services/economy grant (3 SP, p0)

Phase 3 follow-up: Qearn-aware stub shipped. 4 tasks done (~5 SP). Stub now models the real Qearn contract semantics: week-bucket early-unlock penalty, ~12.5% APY accrual, 50/50 burn/redistribution split on penalty. Dev seed runs on startup so treasury + showcase validators + Qearn contract alias are available out of the box. Real TCP client swap is the remaining gap to live.

Doc added: Phase 17 — Material Science Intelligence ANN Delivery Report

Phase 17 delivery report registered: docs/deliveries/phase-17-material-science-delivery.md

Phase 17 delivery report written

Doc added: Material Science Intelligence ANN — Architecture Evaluation

Doc added: Use case — Aigarth Cloud as a collaborative ANN lab for material science

Phase 17 (Material Science Intelligence ANN) created — 4-dimension architecture compatibility assessment complete (ANN 8/10, Compute 6/10, Continuous learning 5/10, Qubic 9/10 blocked on K12).

docs/use-cases/material-science-eval.md — architecture evaluation registered

docs/use-cases/material-science.md — blog article registered

Dashboard /material-science page live: 8-role architecture diagram, 4-dimension compatibility matrix, 5-phase roadmap, per-stage cost model (DFT dominates at 87%), 8-row risk register. Same look-and-feel as /video.

Doc added: Phase 16 — Video Synthesis ANN Delivery Report

Phase 16 delivery report registered: docs/deliveries/phase-16-video-synthesis-delivery.md

Phase 16 delivery report written — ~95 min, 13 SP, 7.3 min/SP

Doc added: Phase 16 — Video Synthesis ANN Delivery Report

Phase 16 delivery report registered: docs/deliveries/phase-16-video-synthesis-delivery.md

Phase 16 delivery report written — ~95 min, 13 SP, 7.3 min/SP

Doc added: Video Synthesis ANN — Architecture Evaluation

Doc added: Use case — Aigarth Cloud as a coordinator for AI video production

Phase 16 (Video Synthesis ANN) created — 4-dimension architecture compatibility assessment complete (ANN 8/10, Compute 7/10, Continuous learning 5/10, Qubic 9/10 blocked on K12).

docs/use-cases/video-synthesis-eval.md — architecture evaluation registered

docs/use-cases/video-synthesis.md — blog article registered

Dashboard /video page live: 7-role architecture diagram, 4-dimension compatibility matrix, 5-phase roadmap, illustrative cost model, 8-row risk register. Same look-and-feel as /services.

Genesis Document published: Aigarth Cloud — The Genesis Document

Doc added: Aigarth Cloud — The Genesis Document

Doc added: ADR 001 - Intelligence Economy Layer

Doc added: Phase 9 — apps/web Wiring Delivery Report

Phase 9 (apps/web wiring) marked complete — 38 SP delivered in ~180 min (4.7 min/SP). Auth + 10 dashboard pages + chat playground all backed by real APIs. 7/7 services now consumable from a real browser session.

Phase 9 delivery report written: docs/deliveries/phase-9-delivery.md

Cookie-based auth flow shipped: signup → auto-login → session cookie → /v1/me → dashboard. Middleware guards /dashboard/*. Login + signup forms + logout round-trip clean.

Chat playground wired to the real gateway via /api/chat SSE proxy. Streaming end-to-end, model picker, system prompt, stop button, clear.

Hard-won: workspace packages + Next.js + .js extensions = import the SDK's compiled dist via a relative path, not the @aigarth/sdk alias. The alias follows the symlink to src/ which Next can't resolve. Documented in lib/server/aigarth.ts.

Doc added: Phase 8 — Developer Platform Delivery Report

Phase 8 (Developer Platform) marked complete — 28 SP delivered in ~135 min (4.8 min/SP). Multi-service SDK + aigarth CLI + node-chat sample app shipped. All 7 services now consumable from one client.

Phase 8 delivery report written: docs/deliveries/phase-8-delivery.md

@aigarth/sdk now covers all 7 services: identity, qubic, compute, gateway, billing, ann, marketplace. 8 typed resources, ~75 methods, error class hierarchy, custom request() helper.

aigarth CLI shipped: login, whoami, chat (one-shot + REPL), keys (list/create/revoke), anns (list/deploy), usage, init. 7 subcommands, no external deps.

examples/node-chat shipped: Express + SSE chat demo, ~200 LOC. Open at http://localhost:8787.

Doc added: Phase 6 — Compute Marketplace Delivery Report

Phase 6 (Compute Marketplace) marked complete — 22 SP delivered in ~80 min (3.6 min/SP). Listings, offers, auctions (Dutch/English/sealed-bid), bids, reviews. 3.6 min/SP is the new velocity floor — patterns are fully locked. Phase 8 (Developer Platform) or Phase 9 (Dashboard) is next.

Phase 6 delivery report written: docs/deliveries/phase-6-delivery.md

Phase 5+ gap-closing (Stripe-style credit gating, fail-closed deploy, pg_trgm search, idempotency, rate limit, analytics) shipped earlier this session: ~50 min, 11 SP, 4.5 min/SP.

Doc added: Phase 5+ — Gap-Closing Delivery Report

Phase 5+ gap-closing complete: 7 gaps closed, 11 SP, ~50 min, 4.5 min/SP. E2E extended to 15 sections, all green. Phase 6 (Compute Marketplace) unblocked.

Phase 5+ delivery report written: docs/deliveries/phase-5-plus-delivery.md

Phase 5 stable. Phase 6 (Compute Marketplace) starting: listings, offers, auctions (Dutch/English/sealed-bid), bids, purchases, reviews.

Doc added: Phase 5 — ANN Platform Delivery Report

Phase 5 (ANN Platform) marked complete — 24 SP delivered in ~145 min (6.0 min/SP). Registry + marketplace: 9 tables, 5 enums, 23 endpoints, 10-section E2E all green. Slug auto-gen, semver versioning, denormalized aggregates, deploys to services/compute, license grants, Qubic signing (format-only).

Phase 5 delivery report written: docs/deliveries/phase-5-delivery.md

Extracted entity serializers to lib/serialize.ts as a new convention (used by routes/anns.ts and routes/me.ts). Replaces the per-route local serializers that caused typecheck friction in earlier phases.

Doc added: Phase 4 — Billing Delivery Report

Phase 4 (Billing) marked complete — 22 SP delivered in ~130 min (5.9 min/SP). Stripe-style credit model: invoice total = subtotal, credits debited at pay time, preview shows would-be credit. 4 plans seeded, 21 endpoints, 11-section E2E all green.

Phase 4 delivery report written: docs/deliveries/phase-4-delivery.md

Re-designed credit application mid-flight: auto-apply at generation → apply at pay time (Stripe model). Avoids the '0 credit balance after auto-apply, can't pay-with-credits' failure mode. Pro plan signup credit bumped 5 Qu → 30 Qu for first-month-free.

Doc added: Phase 7 — AI Gateway Delivery Report

Phase 7 (AI Gateway) marked complete — 15 SP delivered in ~100 min (6.7 min/SP). OpenAI-compatible surface live. Stub backends; SDK consumes it. Real provider integration is next.

Phase 7 delivery report written: docs/deliveries/phase-7-delivery.md

@aigarth/sdk now consumes the gateway end-to-end (list, chat, stream, embed).

Doc added: Phase 2 — Aigarth Core (Compute) Delivery Report

Phase 2 (Aigarth Core) marked complete — 19 SP delivered in ~100 min (5.3 min/SP). Compute broker over Qubic tick execution. Biggest decision: don't reinvent Qubic's scheduler.

Phase 2 delivery report written: docs/deliveries/phase-2-delivery.md

Doc added: Phase 3 — Qubic Integration Delivery Report

Phase 3 (Qubic Integration) marked complete — 19 SP delivered in ~140 min (7.4 min/SP). Stub-backed service + E2E test green. Real Qubic TCP client is the last gap to live.

Phase 3 delivery report written: docs/deliveries/phase-3-delivery.md

Phase 1 (Identity & Access) complete. Sprint 2 shipped: orgs, members, teams, API keys, TOTP MFA, WebAuthn stub, Qubic wallet linking (stub verifier), audit log reads. End-to-end test in services/identity/scripts/e2e-sprint2.ts passes 30+ assertions.

Phase 0 (Foundation) complete: monorepo, design system, SDK, local infra, CI all built and verified.

Doc added: Delivery Report Template (Time-to-Ship Standard)

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents

Phase 1 (Identity & Access) complete. Sprint 2 shipped: orgs, members, teams, API keys, TOTP MFA, WebAuthn stub, Qubic wallet linking (stub verifier), audit log reads. End-to-end test in services/identity/scripts/e2e-sprint2.ts passes 30+ assertions.

Phase 0 (Foundation) complete: monorepo, design system, SDK, local infra, CI all built and verified.

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents

Phase 1 (Identity & Access) complete. Sprint 2 shipped: orgs, members, teams, API keys, TOTP MFA, WebAuthn stub, Qubic wallet linking (stub verifier), audit log reads. End-to-end test in services/identity/scripts/e2e-sprint2.ts passes 30+ assertions.

Phase 1 (Identity & Access) — Sprint 1 complete: identity service scaffolded, schema applied to local Postgres, full auth flow (signup/login/verify/reset/me/logout) tested end-to-end with audit log verified.

Doc added: Phase 1 — Identity & Access Delivery Report (Sprint 1)

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents

Phase 1 (Identity & Access) — Sprint 1 complete: identity service scaffolded, schema applied to local Postgres, full auth flow (signup/login/verify/reset/me/logout) tested end-to-end with audit log verified.

Phase 0 (Foundation) complete: monorepo, design system, SDK, local infra, CI all built and verified.

Doc added: Phase 0 — Foundation Delivery Report

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents

Phase 0 (Foundation) complete: monorepo, design system, SDK, local infra, CI all built and verified.

Doc added: Product Requirements Document

Doc added: Business Requirements Document

Doc added: System Architecture

Doc added: Technology Stack

Doc added: Data Model

Doc added: API Specification

Doc added: Security & Compliance

Doc added: Sprint Plan

Doc added: Risk Register

Doc added: Team and Roles

Doc added: Governance

Doc added: Glossary

Doc added: Brand & Voice

Doc added: Contributing

Doc added: Developer Guide

Doc added: Documentation Index

Doc added: Endgame Proposal (Qubic Incubation)

Doc added: Engineering Roadmap

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents

Doc added: Product Requirements Document

Doc added: Business Requirements Document

Doc added: System Architecture

Doc added: Technology Stack

Doc added: Data Model

Doc added: API Specification

Doc added: Security & Compliance

Doc added: Sprint Plan

Doc added: Risk Register

Doc added: Team and Roles

Doc added: Governance

Doc added: Glossary

Doc added: Brand & Voice

Doc added: Contributing

Doc added: Developer Guide

Doc added: Documentation Index

Doc added: Endgame Proposal (Qubic Incubation)

Doc added: Engineering Roadmap

Tracker seeded with 16 phases, 8 starter tasks, and 18 documents