# Master milestone catalog

These are candidate achieved states, not approved commitments. Detailed cards
live in [`catalog/`](./catalog/).

| ID and title | Workstream | Achieved state and intended outcome | Primary deliverables | Dependencies | Acceptance evidence | Decision enabled | Candidate horizon | Evidence / confidence | Primary sources |
|---|---|---|---|---|---|---|---|---|---|
| [M-001 Corpus release basis explicitly decided](./catalog/M-001-corpus-release-basis-decided.md) | W-01 Corpus | An exact accept/hold record makes retry3's permitted internal/vector-source use and quarantine boundary unambiguous; O-01 | Scoped owner record; immutable identity receipt; downstream policy update | Named owner; exact retry3 hash/evidence | Authorization record matches exact identity and scope; no broader approval inferred | Freeze internal source or revise downstream assumptions | Immediate / months 0–1 | Confirmed; High | `CURRENT.md` — Blocking owner decision |
| [M-002 Full-corpus schema evidence completed](./catalog/M-002-production-schema-direction-proven.md) | W-01 Corpus/schema | The authorized schema-v6.1-r2 full A/B shadow passes its complete technical gate and is admitted truthfully as non-runtime, noncanonical evidence; O-01/O-02 | Sealed A/B shadow; exact receipts; reconciliation/consumer proof; full-shadow manifest | Consumed exact authorization; immutable inputs; ADR-0032 admission | Byte identity; hash/count/integrity/lineage/storage/query/reader/citation checks; strict bundle verification | Keep evidence-only, define a separate r2 runtime-adoption contract, or revise architecture | Current foundation / achieved | Confirmed; High | `CURRENT.md` Track A; ADR-0032; full-shadow receipt |
| [M-003 Retrieval and evaluation contract adopted](./catalog/M-003-retrieval-evaluation-contract-adopted.md) | W-02 Retrieval | Human-reviewed multilingual judgments, structural metadata, semantic-unit/source-edge contract, metrics, and thresholds are frozen before production selection; O-02 | Eval set/protocol; lexical baseline; metadata rules; ADR/config; reviewer receipts | Read-only corpus access; reviewer governance; M-002 schema fields inform filters | Independent adjudication; versioned strata; zero hard-gate violations; thresholds recorded, not invented | Authorize comparable model/unit/engine experiments and production-candidate criteria | Near / months 0–4 | Confirmed requirement; High | ADR-0026; retrieval research; `15-vb25...strategy.md` |
| [M-004 Trustworthy multilingual retrieval release certified](./catalog/M-004-multilingual-retrieval-release-certified.md) | W-02 Retrieval | A named corpus plus optional vector pack provides evaluated lexical/dense/hybrid behavior, exact citations, truthful fallback, and acceptable product performance; O-02 | Deterministic sidecar; model/engine/runtime; fusion/reranker; manifest; release eval | M-001, M-002, M-003 | A/B deterministic build; integrity/cardinality/source edges; per-stratum quality; latency/memory/size; failure modes | Permit semantic discovery in declared releases and downstream grounding | Early middle / months 3–7 | Confirmed direction; Medium | `04-search-and-embeddings.md`; `15-vb25...strategy.md` |
| [M-005 Coherent offline research experience integrated](./catalog/M-005-offline-research-experience-integrated.md) | W-03 Product/UX | In the routed ADR-0028 shell, users browse heterogeneous sources, read directly, search, cite, annotate, organize and write in Projects, and resume offline without false capabilities; O-03 | Generic nav/reader; search; notes/highlights; minimum Projects Writing/Board decision; settings; critical states | Generic schema/core contracts; design system; user-data migrations; corpus fixtures/candidate | Representative end-to-end journeys; Unicode/citation/restart; keyboard/accessibility; loading/empty/error/offline; real macOS runtime | Define the controlled-release feature set and retire compatibility branches where proven | Early middle / months 2–7 | Confirmed direction + Recommended product state; Medium | ADR-0021/0028; `07-desktop-app.md`; V3 reference |
| [M-006 macOS controlled release ready](./catalog/M-006-macos-controlled-release-ready.md) | W-03/W-04 Product release | A defined controlled population can safely install, use, update, recover, report problems, and understand the declared local feature set; O-03/O-06 | Signed/notarized package; bundle installer; onboarding; update/recovery; support/release runbook | M-001–M-005; scoped rights; release population and hardware matrix | Install/launch/upgrade/rollback; performance/accessibility; task success; support readiness; capability truth | Expand, narrow, iterate, or hold the macOS release | Middle / months 6–10 | Recommended; Medium | `11-roadmap.md` Gate 4; ADR-0028 |
| [M-007 macOS commercial release decision supported](./catalog/M-007-macos-commercial-release-decision-supported.md) | W-09 Commercial | Rights, packaging, pricing/entitlement hypotheses, onboarding, activation, support, privacy measurement, and controlled-release learning are sufficient for a public/commercial decision; O-08 | Rights matrix; offer tests; entitlement/account design; channel/support/measurement plans; decision memo | M-006; leadership/commercial/legal ownership | No invented thresholds: named evidence, willingness-to-pay/activation method, operational cost, rights and support sign-offs | Public release, limited release, revised offer, or defer | Middle / months 8–12, conditional | Recommended; Low/Medium | Overview legal posture; pricing and managed-cloud drafts |
| [M-008 Narrow grounded Atma ready for controlled use](./catalog/M-008-narrow-grounded-atma-controlled-use.md) | W-05 Atma | Users can ask bounded corpus questions and comparisons, inspect exact evidence and scope, receive truthful abstention/failure, and explicitly adopt any proposal; no persistent learning; O-04 | Neutral tools; bounded loop; identity/policy; citation gate; trust UI; eval/red team; problem report | M-001, M-003, M-004; M-005 trust surfaces; runtime-profile decision | Exact/clickable evidence; quote fidelity; supported-claim/abstention/override tests; user task validation; privacy receipts | Broaden controlled Atma workflows or revise/hold | Middle / months 7–12 | Confirmed intent + Recommended scope; Medium | `08-ai-agent.md` Gates 0–2 |
| [M-009 Governed project continuity proven](./catalog/M-009-governed-project-continuity-proven.md) | W-05/W-06 Atma/user data | Users resume multi-session projects with questions, hypotheses, artifacts and scoped, inspectable, correctable, exportable/deletable memory while Projects remain useful without Atma; O-04/O-05 | Session/run schema; memory policy/migrations; What I know; controls; longitudinal eval; recovery/export | M-005, M-008; memory product/privacy decisions | Migration/recovery/isolation; correct scope; correction/deletion/export; longitudinal usefulness without stale/cross-project drift | Enable or constrain proactive support, advanced skills, and optional sync | Later / months 11–18 | Confirmed architecture + Recommended horizon; Medium/Low | ADR-0029; `08-ai-agent.md` Gates 3–6 |
| [M-010 Hosted corpus evidence pilot operational](./catalog/M-010-hosted-corpus-evidence-pilot.md) | W-07 Hosted MCP | Invited authenticated clients receive bounded exact corpus evidence with truthful modes, anti-extraction controls, privacy-safe operations, and rollback; O-07 | Neutral tool DTOs; HTTP service; OAuth/entitlements; policy ledger; deployment/observability | M-001, M-004, M-006; rights and cloud owner | Deployed client interoperability; auth/isolation; exact citations; abuse/fail-closed tests; latency/reliability/cost receipts | Expand hosted MCP, revise limits, or stop service investment | Later / months 10–18 | Confirmed architecture + Recommended horizon; Medium | ADR-0019; `06-mcp-server.md` |
| [M-011 Windows supported release ready](./catalog/M-011-windows-supported-release-ready.md) | W-04 Desktop platform | A defined Windows population can install and use the declared parity set with platform-specific UX, Unicode/search, update, recovery, and performance proof; O-06 | Windows CI; signed installer; platform adapter/config; test matrix; release/support runbook | M-004, M-006; hardware/CI/signing; support matrix | Direct Windows install/runtime/search/copy/project/update/recovery/performance/accessibility evidence | Claim Windows support, narrow parity, or defer | Later / months 12–18, directional | Confirmed sequence; Medium/Low | ADR-0025; `CURRENT.md` Platform direction |
| [M-012 Cloud project continuity pilot validated](./catalog/M-012-cloud-project-continuity-pilot.md) | W-06/W-09 Cloud continuity | Opted-in users can back up, restore, and—if validated—share one project under explicit roles, conflicts, history, source entitlements, privacy, and portability; O-05/O-07/O-08 | Identity; sync engine; encryption; restore; roles; version/conflict model; exports; ops | M-009; commercial/runtime decision; rights; account/retention model | Local/offline transitions; conflict/restore drills; tenant/role isolation; deletion/export; user value and cost evidence | Expand backup only, collaboration, or defer cloud state | Later / months 12–18, conditional | Open/Recommended; Low | ADR-0015; pricing/managed-cloud drafts on HOLD |
| [M-013 Paramatma selected-text evidence workflow validated](./catalog/M-013-paramatma-selected-text-workflow.md) | W-08 Paramatma adapter | An explicit selected text or project claim yields a citation-bound evidence report with accepted review states and human confirmation through a thin adapter; O-04/O-10 | External contract/repo dependency; adapter; report DTOs; review UI; evals | M-004, M-008; Paramatma package/release contract; reviewers | Exact claim/evidence round trip; isolation; accepted statuses; human review; failure/abstention | Expand the adapter or keep Paramatma separate from active product scope | Later / months 12–18, conditional | Confirmed boundary; Low/Medium | ADR-0017; `08-ai-agent.md` Gate 7 |
| [M-014 Mobile investment decision ready](./catalog/M-014-mobile-investment-decision-ready.md) | W-04/W-10 Mobile | Leadership has enough user-demand, mobile-journey, corpus-pack, offline/cloud, device, cost, and platform evidence to fund, defer, or reject a mobile product and record a new ADR; O-06 | Research; mobile product brief; pack experiments; bounded device spikes only after authorization; ADR proposal | M-006 learning; stable portable contracts; explicit authority for spikes | Named mobile jobs; device/package feasibility; storage/performance/accessibility/privacy; investment case | Start iOS/Android planning, choose one platform, or remain deferred | Late / months 15–18 or outside horizon | Confirmed deferral + Recommended decision state; Low | ADR-0025; historical platform alternatives |
| [M-015 Language or edition expansion decision ready](./catalog/M-015-language-edition-expansion-decision-ready.md) | W-10 Language/edition | A named additional language/edition has demand, rights, source acquisition, corpus/retrieval/citation/typography/localization/QA/maintenance evidence sufficient to fund or defer it; O-09 | Demand/partner evidence; source audit; localization/retrieval plan; maintenance and cost model | Stable release process; named source/rights/reviewers; M-004/M-006 learning | No universal metric: full acceptance plan and representative evidence for the named edition | Fund one edition, gather evidence, partner, or defer | Outside horizon unless prerequisites emerge | Open/Recommended; Low | Corpus invariants; no named current expansion program |
| [M-016 Narration launch decision ready](./catalog/M-016-narration-launch-decision-ready.md) | W-05/W-09 Atma/TTS | A bounded narration pilot preserves frozen text, source/synthesis distinctions, alignment, synthetic disclosure, pronunciation review, rights, privacy, reliability, and sustainable cost well enough to launch or reject; O-04/O-09 | Render contract; provider/voice profiles; alignment verifier; audio receipt; human review; cost/rights evidence | M-008; runtime/commercial decision; rights; qualified language reviewers | Text/audio fidelity; alignment; pronunciation strata; no impersonation; privacy/cost/retry/failure proof | Include narration in a later offer, revise, or keep out of program | Outside horizon / conditional | Open/conflicting; Low | `08-ai-agent.md` TTS; overview non-goal; managed-cloud draft |

## Catalog interpretation

- M-002 is an already achieved technical-evidence foundation. M-001 and
  M-003–M-006 form the recommended first decision/product-state sequence. None
  of the future states is an approved delivery commitment.
- M-007 and M-008 require product decisions in addition to technical evidence.
- M-009 through M-013 are later or conditional expansion tracks.
- M-014 through M-016 primarily preserve decision readiness; they prevent
  mobile, language expansion, or narration from becoming invented commitments.