OpenDataSuite (ODS) — Roadmap, Milestones & Release Plan¶
Status: proposed · Last updated: 2026-11-01
Plans, not promises. Target dates are proposals for a small team and are revisited at the end of each milestone. Scope, order and dates will change.
This document turns the initial backlog (issues #1–#157) into an ordered set of
milestones and releases. It is the planning source of truth until the vision
document (#103) lands; milestone membership is mirrored in
.github/milestones.json and applied to GitHub with
scripts/sync-milestones.sh.
1. Product thesis (one paragraph)¶
ODS is a Rust-first, provider-neutral control plane for analytics engineering.
A shared semantic core (ods-core) and a versioned plugin SDK power independently
useful modules — State (incremental, explainable "what needs to run"),
ERD, Usage, CI, LSP, Agent, Mesh and Synthetic. dbt is
the first project format and Databricks/Unity Catalog the first warehouse, but
neither is allowed to leak into core. Every decision ODS makes must be
explainable with evidence, and unknown evidence always resolves to the
conservative action (BUILD, deny, flag as inferred).
2. Guiding principles (release gates)¶
Every milestone must preserve these; reviewers reject PRs that break them.
- Core is provider-neutral — no
if warehouse == "databricks"in core; use capabilities (#3). - Contracts before implementations — providers depend on the SDK (#2), never the reverse.
- Conservative by default — insufficient evidence ⇒ BUILD / fail closed (#9, #20).
- Explainability — every plan/finding carries a reason chain and evidence (#21, #43).
- Failed runs never replace canonical state (#11, #24).
- Machine-readable output everywhere — human,
--json,--plain(#6, #108). - Public formats only — interoperate through public/OSS dbt contracts; no proprietary code or binaries (#10, #67).
- ERD ≠ lineage — separate domain types (#5).
3. Release train¶
| Release | Theme | Milestone(s) | Target | Exit criteria (summary) |
|---|---|---|---|---|
| pre-release (internal) | Foundations | M0 | 2026-10-30 | Workspace builds in CI; ADRs for architecture, Rust stack, CLI presentation; plugin SDK + semantic core compile with fake providers; demo dbt fixture project. |
| v0.0.1 — first public release | ODS State MVP (code- and data-aware) + first dashboard screens | M1 | 2026-10-30 | ods state plan / run / explain / diff / history on a dbt project using manifest.json, SQLite state, exact-selection dbt executor; skips models whose code and upstream data are unchanged (dbt source freshness + Delta table versions), with per-node staleness tolerance; failed runs leave prior state authoritative; JSON output; installable binary for Linux/macOS/Windows. ods serve dashboard: shell and Home (#310), State Plan and Why, Runs, Run (#311), Lineage with the State overlay (#312), Catalog and the Model page (#313). |
| v0.0.2 | ERD, Impact & hardening | R2 | 2026-11-27 | The ERD page (#64) and the Impact simulator (#347) on the dashboard, both over engines already built; failure explanations for agents and the CLI (#348); ods state savings (#210); Databricks error messages recorded from a real run (#349); the plugin conformance suite (#99) and property tests (#192). |
| v0.1.1 — released | ODS Dashboard complete | UX1 | 2026-11-01 (planned for 2027-01-29) | Every screen in the dashboard design is built (#309): the v0.0.1 screens plus Freshness evidence, Semantic layer placeholder, Impact simulator, ERD, Settings and dark mode. |
| v0.1.3 — released | First M2 work | M2 | 2026-11-04 | Failures as diagnostics (#447) and live run progress (#446) on a terminal (ADR-0003 §9, #442); an incremental model in the demo project (#435); the docs site on Zensical; secret providers (#126); Databricks sign-in, M2M and PAT, with ods doctor --connect (#297); producing-code references and code retrieval (#462); the crates on crates.io (ADR-0035); releases through rab. v0.1.2 was prepared but never published: its changes are in v0.1.3. |
| v0.1.4, … | Incremental releases | M2, and any later-milestone work that lands early | as ready | Fixes, and work from M2 onwards that is done before v0.2.0. |
| v0.2.0 | Databricks depth & reuse strategies | M2 | 2027-02-26 | Unity Catalog metadata; full Delta change providers (CDF, watermarks, partitions); remaining freshness/trigger policies incl. WAIT; REUSE/DEFER/CLONE strategies; PostgreSQL store with locking; secrets providers. |
| v0.3.0 | Schema intelligence: ERD + Usage | M3 | 2027-04-30 | ods erd generate/inspect/validate (dbt provider, Mermaid/DOT/PlantUML); ods usage … backed by Unity Catalog. |
| v0.4.0 | ODS CI | M4 | 2027-06-30 | Change-impact engine, selective CI planner, PR report (Markdown + JSON), usage-aware risk; SQL parser + column lineage. |
| v0.5.0 | Developer experience | M5 | 2027-08-31 | Clean-room LSP (index, completion, diagnostics, navigation, rename) + VS Code extension. |
| v0.6.0 | ODS Agent | M6 | 2027-10-29 | Agent runtime, BYOK LLM providers, policy/approval, patch engine, self-validation, first skills (tests/docs/investigate). |
| later | Platform & ecosystem | M7–M9 | — | Mesh, server mode/API/RBAC/audit, integrations, synthetic data, dev envs, ops/cost/governance. |
Versioning rule: releases stayed v0.0.x until every screen in the dashboard design
(docs/design/dashboard/) was built; v0.1 marks the complete dashboard (UX1,
309). 0.1.0 was prepared but never released, so v0.1.1 is the first 0.1 release. Each¶
milestone after it ships as the next minor version (M2 → v0.2.0, M3 → v0.3.0, …); fixes, and work that lands before its milestone's release, ship as v0.1.x patch releases.
Target dates are proposals for a small team and should be revisited at the end of M0.
Versioning follows SemVer with a 0.x "anything may break between minors" caveat; in
the v0.0.x series any release may break.
The formal policy (#101) is ADR-0019: one product
version, a version per interface, CHANGELOG.md and a deprecation window.
4. Milestones¶
Issue numbers in bold are on the critical path for the milestone's release.
M0 — Foundations (internal, before v0.0.1)¶
Goal: a compiling Rust workspace with the architectural skeleton every other module builds on.
| Issue | Title |
|---|---|
| #1 | Monorepo architecture & module boundaries (ADR) |
| #105 | Rust-first backend architecture (ADR) |
| #2 | Versioned plugin SDK |
| #3 | Provider capability negotiation |
| #4 | Shared semantic graph/domain core |
| #5 | ERD and lineage as distinct graph domains |
| #6 | Composable ods CLI framework |
| #7 | Unified configuration & profiles |
| #8 | Structured events & observability model |
| #108 | rs-rich-cli as the CLI presentation layer (ADR + PoC) |
| #100 | Representative dbt demo project & fixtures |
| #10 | Clean-room & third-party licensing policy |
| #101 | Release/versioning strategy |
| #103 | Vision/roadmap document in repo |
| #104 | Naming/package/domain availability research |
Definition of done: cargo build/test/clippy/fmt + cargo deny green in GitHub Actions;
ADR-0001 (module boundaries), ADR-0002 (Rust stack), ADR-0003 (CLI presentation) merged;
ods --version and ods state --help run; a fake provider passes a contract smoke test.
Status (2026-10-19): nothing left open. Done: #1, #105, #2, #3, #6, #7, #101; #100 and #108 with their last pieces moved to M2 (#435, #436). What was left of the others moved to the milestone that needs it: - M2: #4 the domain model, #183 entity identity, #10 licensing policy, #104 naming, #187 benchmarks, #189 roadmap drift in CI, #191 dbt version fixture matrix. - M3: #5, a test that lineage and ERD edges stay apart. - M7: #103 the vision document. M8: #8 the event sink and OpenTelemetry. - M10: #184 evidence and provenance, #186 presentation demo, #190 backlog hygiene.
M1 — State MVP (→ v0.0.1, first public release)¶
Goal: prove the core value — "ODS decides WHAT runs, dbt decides HOW" — locally and explainably. v0.0.1 also carries the first dashboard screens (#310–#313, listed below and under UX1).
| Issue | Title |
|---|---|
| #11 | State domain model & lifecycle |
| #12 | dbt artifact provider (manifest/run_results/catalog) |
| #13 | Deterministic code fingerprinting |
| #14 | Provider-neutral RelationState & data fingerprints |
| #18 | DAG invalidation & impact propagation |
| #20 | State execution planner |
| #22 | ods state plan and dry-run |
| #21 | State explain/diff/history/why commands |
| #23 | dbt execution provider (exact node selection) |
| #25 | SQLite StateStore |
| #24 | ods state run end-to-end MVP flow |
| #26 | Filesystem JSON StateStore (debugging) |
| #16 | Source/upstream change detection contract; includes dbt sources.json freshness (max_loaded_at) as change evidence for any warehouse |
| #19 (M1 slice) | Freshness policies, minimal: AnyDependencyChanged (default), MaxStaleness (per-node/group tolerance, default 0: rebuild on any new data) and forced rebuild; the policy used appears in explain |
| #17 (M1 slice) | Delta change detection, minimal: latest table version + commit timestamp for sources, behind the relation_versions capability, read through dbt (ADR-0022) |
| #99 | Plugin conformance test suite (State-related contracts) |
| #168 | Read dbt State configs (state:, freshness.build_after, loaded_at_*) so existing dbt State projects work unchanged; dbt defaults (45m/any) only when the project already uses State, otherwise tolerance 0 |
| #211 | dbt executor: exact selection that can't widen (fqn:/selector file), no command-line length limit |
| #212 | Distribution: release binaries, Homebrew, pip-installable wrapper |
| #209 | Formatting-insensitive SQL fingerprints: comment/whitespace edits reuse |
| #214 | ods.toml settings for ods state run (dbt path, dirs, target, environment) |
| #220 | ods state run without tests by default; --test, ods state test, --full-refresh, --resource-type, --exclude, dbt args after -- |
| #227 | State scope includes the dbt target; dbt settings read from the environment (DBT_TARGET, DBT_PROFILE, DBT_FULL_REFRESH) are honoured or refused |
| #229 | dbt-shaped commands: ods state compile, run (models, dbt run), seed, snapshot, build (with tests); --full-refresh and --vars as in dbt |
| #230 | Don't reuse a node whose table no longer exists in the warehouse |
| #232 | Run source tests in build/test when the source has new or unknown data |
| #233 | CI job running the real-dbt integration tests (dbt + DuckDB) |
| #310 | Dashboard: app shell and Home (health overview) |
| #311 | Dashboard: State pages (Plan and Why, Runs, Run) |
| #312 | Dashboard: Lineage with State overlay (restyle ods serve lineage) |
| #313 | Dashboard: Catalog (models) and Model page |
| #73, #74 (preview) | SQL parser and open column-level lineage: ods lineage, ods serve, OpenLineage export, observed lineage from Unity Catalog (#164–#167). Delivered early; the CI integration stays in M4 |
Status (2026-10-04): released as v0.0.1.
- Closed on GitHub: #11, #18, #20, #21, #22, #23, #24, #74, #168, #181, #188, #209, #211,
#212, #214, #220, #227, #229, #230, #232, #233 and the four dashboard screens
#310–#313 (merged in #316, #318, #319, #320).
- Delivered beyond the table: run events, the run journal, the live view and run
playback (#322, ADR-0024,
ADR-0026), and failed nodes and tests explained (#323,
ADR-0025).
- Moved out of M1 on 2026-10-04, with their v0.0.1 slice merged and in use; what's
left of each goes to M2: #12 (the formal ArtifactProvider contract), #13 (the
pluggable FingerprintProvider), #14 (RelationState), #16 (the ChangeSet type
and ChangeProvider contract), #17 and #19 (the remainders below) and #25 (plan
history). #73's preview is done too, and its syntax-tree and symbol API with source
ranges goes to M5, which needs it.
- Moved out of M1 with no work merged, since v0.0.1 doesn't need them: #26 (a JSON
store for debugging) goes to M2, and #99 and #192 (the conformance suite and property
tests) go to R2.
- The release itself (#356): v0.0.1 is on GitHub Releases and PyPI (#360), and the
ADRs that shipped are accepted (#357, merged in #359).
Why data awareness moved into M1: skipping only on code changes still rebuilds models whose upstream data hasn't changed. Skipping on unchanged code and data is the minimum useful behaviour for the State MVP.
What the #17 slice pulls in: table history is read through dbt's own connection
(dbt show --inline), with the Delta query supplied by ods-provider-databricks
(ADR-0022). ODS still handles no
warehouse credential, so neither a Databricks SQL connection (#15) nor a secret
resolver (#126) is needed for M1. Both stay in M2.
Staleness is propagated through views: a view is only as fresh as its inputs, so
18 carries upstream freshness through view-materialised models.¶
Out of scope for v0.0.1: the remaining dashboard screens (UX1, #309), Unity Catalog metadata beyond table history, Change Data Feed, watermark/partition triggers, WAIT decisions, clone/defer, distributed locking, server mode. The M1 slices keep #17 and #19 open; their remainder ships in M2.
R2 — ERD, Impact & hardening (→ v0.0.2)¶
Status (2026-10-06): released as v0.0.2.
Goal: put on the dashboard what ODS already computes but only shows in the CLI, and harden what v0.0.1 ships. The two screens are part of the v0.1.0 design (#309) and come first because their engines exist.
| Issue | Title |
|---|---|
| #64 | Dashboard: the ERD page, over ods-erd (from M3) |
| #347 | Dashboard: the Impact simulator, over ods lineage impact |
| #348 | Explain a failure from MCP and the CLI (explain_failure, ods state explain-failure) |
| #210 | ods state savings: builds skipped and time avoided (from M2) |
| #349 | Record the dbt catalogue's Databricks error messages from a real run |
| #99 | Plugin conformance test suite |
| #192 | Property and fuzz tests for the graph, SQL lineage and State invariants |
UX1 — ODS Dashboard (→ v0.1.1)¶
Goal: build every screen of the dashboard design in
ods serve (server-rendered, read-only, ADR-0009).
v0.1.0 ships only when every screen in the design is done; until then releases are
v0.0.x.
Status (2026-11-01): released as v0.1.1 (0.1.0 was skipped). Every screen of the design is built; #309 and its screens are closed.
| Issue | Title |
|---|---|
| #310 (in v0.0.1, tracked in M1) | App shell and Home (health overview) |
| #311 (in v0.0.1, tracked in M1) | State pages (Plan and Why, Runs, Run) |
| #312 (in v0.0.1, tracked in M1) | Lineage with State overlay |
| #313 (in v0.0.1, tracked in M1) | Catalog (models) and Model page |
| #309 | Remaining screens for v0.1.0 (the umbrella; one issue per screen below) |
| #64, #347 (in v0.0.2, tracked in R2) | ERD, Impact simulator |
| #350 | Freshness evidence (Catalog → Sources) |
| #351 | Settings (read-only), with the Server mode placeholders |
| #352 | Semantic layer (Catalog → Metrics), read-only placeholder |
| #353 | Dark mode checked on every screen |
| #354 | Health badges and coverage from real signals (the dashboard slice of #117) |
| #355 | Run timeline: a per-thread Gantt view (playback follow-up, ADR-0026) |
| #388 | Freshness evidence: the source versions the last dry run read, without the server connecting |
| #392 | Configurable and pluggable health checks: [health] config, declarative, probe and script checks, plugins (ADR-0030) |
310–#313 are in the M1 GitHub milestone because they ship in v0.0.1, and #64 and #347¶
are in R2 because they ship in v0.0.2. #309 closes when every screen is built.
M2 — Databricks & data-aware State (→ v0.2.0, current)¶
| #15 Databricks/UC metadata provider · #12 dbt ArtifactProvider contract · #13 FingerprintProvider contract · #14 RelationState · #16 ChangeSet and ChangeProvider · #25 plan history in the SQLite store · #26 JSON StateStore for debugging · #169 ods mcp read-only MCP server (ADR-0010) · #173 ODS skills pack for existing coding agents · #170 live Unity Catalog lineage reader · #17 Delta change providers (remainder: CDF, commit history, watermark, partition arrival) · #19 freshness & trigger policies (remainder: AllDependenciesChanged, MinInterval, WatermarkReached, PartitionAvailable, WAIT) · #29 REUSE/DEFER/CLONE abstraction · #30 Databricks shallow clone · #27 PostgreSQL StateStore · #28 locking/leases/fencing · #126 secrets & external config · #31 column-aware invalidation (research) · #215 public benchmark of ods state run · #221 reuse hooked models safely (keyed digests of run-time values) · #224 dbt selector parity (-s, --selector, methods, graph operators; ADR-0015) · #225 ods dbt … front-end, also run as dbt (ADR-0015) · #231 rebuild or adopt seeds by comparing the table with the CSV (row count + hash) · #387 pluggable source-version (freshness) providers: a registry instead of the hard-coded Databricks mapping, external providers, author guide, per-warehouse research (Databricks stays the only built-in) · #297 Databricks authentication (U2M, M2M; ADR-0021) · #437 Databricks integration tests gated by ODS_TEST_DATABRICKS · #438 ods state import and the favor-state scenario on Databricks (ADR-0020) · #398 configurability audit · #408 health probes' login-check gaps (ADR-0030 §4e) · #4 shared domain model · #183 canonical entity identity · #191 dbt version fixture matrix · #435 an incremental model in the demo project · #187 performance benchmarks · #10 clean-room and licensing policy · #104 naming research · #189 roadmap drift in CI · #436 rs-rich-cli gaps filed upstream · #441 bump rs-rich (done by #450: rs-rich and rs-rich-ext from git) · #442 ADR-0003 amendment for rs-rich-ext and rs-rich-diagram · #446 ods state run progress with rs-rich-ext workflow and LiveCoordinator · #447 failures rendered as rs-rich-ext diagnostics |
M3 — ERD & Usage (→ v0.3.0)¶
| #172 ODS metadata index (queryable lineage/State/ERD/usage) · #60 ERD domain model · #61 dbt ERD provider · #62 relationship inference · #63 render/export · #65 ods erd CLI · #55 Usage domain & UsageProvider · #56 Unity Catalog usage provider · #59 ods usage CLI · #66 warehouse-native ERD providers (stretch) · #5 a test that lineage and ERD edges stay apart · #443 the ERD in the terminal with rs-rich-diagram · #444 ods lineage as a graph in the terminal |
M4 — ODS CI (→ v0.4.0)¶
| #75 change-impact engine (on the column lineage delivered in M1 preview) · #171 column lineage for Python models · #84 selective CI planner · #85 PR report/check output · #58 usage in CI risk · #109 data diff · #110 data diff in CI · #213 GitHub Action PR impact comment · #242 ods plan as the one change-planning command · #243 one impact engine for code, columns and consumers (builds on #75) |
M5 — LSP & VS Code (→ v0.5.0)¶
| #67 clean-room LSP architecture · #73 syntax-tree and symbol API with source ranges (the parser's M1 preview is done) · #68 indexing · #69 completion · #70 diagnostics · #71 navigation/hover · #72 semantic rename · #107 VS Code extension · #116 SQL scratch/REPL |
M6 — ODS Agent (→ v0.6.0)¶
Core: #32 architecture & tool runtime · #33 LLMProvider/BYOK · #34 context planner · #9 policy & approval (pulled forward if State needs it) · #50 patch engine · #51 self-validation · #52 Git-aware review · #53 skills SDK. First skills: #36 missing tests (#251 the evidence-driven recommendation engine) · #37 test review · #38 test priority · #57 usage-driven tests · #39/#40 docs · #41 contracts · #35 investigate. Stretch: #42–#49 review skills, #54 explore mode, #123 diagnostics.
Direction: ODS won't build its own chat interface. The engines answer and the model
proposes:
- M2 makes ODS agent-ready in the agents teams already use (#169, #173).
- M6 adds a headless agent for CI and review whose output is a proof-carrying evidence
bundle (#50–#52), governed by declarative policy (#9, #98).
- The working product name is ODS Steward (ods steward as an alias of ods agent),
pending the #104 name check.
M7 — Platform: Mesh & Server (unscheduled)¶
86 Mesh resolver · #89 registry/contracts · #90 change propagation · #91 cross-platform mapping · #95 server mode · #96 REST API · #97 RBAC/OIDC · #98 audit log · #106 language bindings · #102 docs site · #124 multi-repo discovery · #103 vision document · #249 data product model · #254 one project graph for dbt, SQLMesh and raw SQL · #257 plugin ecosystem (the registry is built, ADR-0031; left: installing plugins). Idea, no issue yet: one federated build graph across project formats (dbt + SQLMesh first), with Mesh as the same mechanism for dbt only (ADR-0027). Open data environments for any build tool, unifying #29, #30, #114, #120 and #195 (ADR-0028).¶
Studio (ideas, unscheduled). ODS Studio, an analytics engineering IDE over the control plane (#255), with MetaForge Community as its visual modelling surface: #260 the integration, #261 its contract, #262 #263 diagrams and views, #264 #265 validating visual edits, #266 #267 #268 test, relationship and contract suggestions, #269 Open Data Studio, #270 identity alignment (after #183), #271 the metadata adapter, #272 the change workflow. They wait on a decision about Studio as a product.
M8 — Ecosystem integrations & operations (unscheduled)¶
92 OpenMetadata · #93 Elementary · #94 MetricFlow · #111–#113 cost & optimisation · #117 health scoring · #118 observability store · #119 alert routing · #120 deployment/promotion · #121 package governance · #122 ownership/SLA governance · #125 snapshot/SCD review · #114 dev env cloning · #115 dev sampling · #226 SQLMesh project provider and ods sqlmesh front-end (ADR-0015) · #8 event sink and OpenTelemetry · #247 warehouse intelligence and optimisation recommendations.¶
M9 — ods-synthetic (unscheduled)¶
127 module & ADR · #147 engine contract · #129 profiling · #130 planner · #131 ERD-aware generation · #132 dbt constraints · #133 statistical engine · #134 copula engine · #135 rule/Faker engine · #136 SDV adapter · #137 engine selection · #148 sensitive-column policy · #139 privacy-risk eval · #141 quality eval · #142 validate with dbt tests · #143 writers · #144 CLI · #146 realistic mode · #150 row counts · #151 edge-case mode · #152 ods-dev integration · #153 ods-ci integration · #154 usage-prioritised fidelity · #155 explainability · #156 benchmarks · #149 differential privacy · #157 licensing & privacy docs.¶
Synthetic depends on M3 (ERD, usage) and M1 (core/state). It is a strong v1.x candidate and should get its own ADR (#127) before any engine work.
5. Backlog hygiene — suspected duplicates¶
The backlog was created in several passes and contains duplicates. Recommend closing the
newer/older copy listed on the right as duplicate of the canonical issue on the left
(canonical = the copy with the richer description or lower number used above):
| Canonical | Duplicate(s) | Topic |
|---|---|---|
| #147 | #128 | SyntheticEngine contract |
| #148 | #138 | Sensitive-column policy |
| #149 | #140 | Differential privacy |
| #150 | #145 | Scale/row-count generation |
| #85 | #88 (closed) | CI PR report |
| #84 | #87 (closed) | Selective CI planner |
| #60 | #83 (closed) | ERD domain model |
| #59 | #82 (closed) | ods usage CLI |
| #58 | #78, #81 (closed) | Usage in CI risk |
| #57 | #77, #80 (closed) | Usage-driven test recommendations |
| #56 | #76, #79 (closed) | Unity Catalog usage provider |
| #184 | #246 | Explaining across capabilities (ods why) |
| #114 | #253 | Ephemeral development environments (ADR-0028) |
| #32 | #256 | Evidence-grounded agent |
| #204 | #258 | Portable metadata bundle |
| #103 | #259 | Vision and north star |
Note: several closed copies (#76–#88) look like they were closed as duplicates without
state_reason: duplicate; the open copies remain the canonical ones above.
6. Proposed labels¶
area:core area:sdk area:cli area:state area:erd area:usage area:ci
area:lsp area:vscode area:agent area:mesh area:server area:synthetic
area:dev area:ops area:docs · provider:dbt provider:databricks
provider:postgres · type:adr type:feature type:research type:docs
type:chore · priority:critical-path · good-first-issue.
These are defined in .github/labels.json and created by the sync script.
7. Target crate layout (see ADR-0001, #1)¶
crates/
ods-core/ # semantic graph, domain types, capability model — no I/O, no vendors
ods-sdk/ # plugin contracts (traits) + versioning + conformance harness
ods-events/ # event schema, EventSink, tracing glue
ods-config/ # layered config, profiles, secret references
ods-policy/ # allow/deny/require-approval evaluation
ods-state/ # fingerprints, invalidation, planner, StateStore trait
ods-health/ # health checks, findings and badges, configured by [health] (ADR-0030)
ods-erd/ ods-usage/ ods-ci/ ods-lsp/ ods-agent/ ods-mesh/ ods-synthetic/
ods-cli/ # `ods` binary; module subcommand registration; presentation boundary
providers/
ods-provider-dbt/ # ArtifactProvider, Executor, ERD provider
ods-provider-databricks/ # Metadata/Change/Clone/Usage providers
ods-store-sqlite/ ods-store-fs/ ods-store-postgres/
ods-provider-fake/ # reference fake used by conformance tests
ods-provider-git/ # approved local immutable source retrieval (#462)
fixtures/
dbt/jaffle-ods/ # #100 demo project + artifacts for multiple dbt versions
Dependency direction: ods-core ← foundation ← ods-sdk ← {modules, providers} ← ods-cli.
Modules and providers never depend on each other; the CLI wires concrete providers in at the edge.
8. Immediate next steps (first two weeks)¶
- Merge this plan; run
scripts/sync-milestones.shto create milestones/labels and assign issues. - Close the duplicates in §5.
- Write ADR-0001 (#1) and ADR-0002 (#105) — use the
adrskill. - Scaffold the Cargo workspace + CI (fmt, clippy
-D warnings, test, cargo-deny) —new-crateskill. - Land
ods-coregraph types (#4) + SDK traits (#2) + capability model (#3) with a fake provider. - Build the fixture dbt project and commit its
manifest.jsonfor dbt 1.8/1.9/1.10 (#100).
M10 — Advanced Intelligence & Developer Experience (unscheduled)¶
Cross-cutting developer intelligence and operational ergonomics that build on the core metadata/evidence platform without changing the committed M0–M9 release train.
194 ods drift · #195 environment parity / ods env diff · #196 dead asset detection · #197 lineage confidence/evidence strength · #198 contract compatibility · #199 ods reproduce · #200 warehouse query-plan/performance inspection · #201 metadata-aware search · #202 policy-as-code · #203 local metadata daemon/watch mode · #204 portable metadata snapshots · #205 PII/sensitive-data propagation · #206 mutation testing · #207 migration/upgrade assistant · #208 interactive ods explore · #184 evidence and provenance model · #244 evidence graph · #245 ods debug · #248 architecture fitness rules · #250 architecture time travel · #252 test effectiveness · #186 presentation demo · #190 backlog hygiene checks · #445 an interactive terminal UI on rs-rich intuiTUIve (ADR).¶
This milestone is intentionally unscheduled. Individual capabilities can be pulled forward when they directly unblock an earlier milestone, but the source-of-truth grouping remains M10 until the roadmap is rebaselined.