openmapstack 0.4.0__tar.gz → 0.5.0__tar.gz

{openmapstack-0.4.0 → openmapstack-0.5.0}/PKG-INFO

@@ -1,14 +1,15 @@
 Metadata-Version: 2.4
 Name: openmapstack
-Version: 0.4.0
+Version: 0.5.0
 Summary: Validate, run, and inspect reproducible OpenMapStack projects
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: jsonschema<5,>=4
 Requires-Dist: PyYAML<7,>=6
+Requires-Dist: packaging>=23
 Provides-Extra: geo
-Requires-Dist: duckdb>=1.2; extra == "geo"
+Requires-Dist: duckdb>=1.5.3; extra == "geo"
 Provides-Extra: visual
 Requires-Dist: playwright>=1.40; extra == "visual"
 Provides-Extra: postgis
@@ -16,27 +17,33 @@ Requires-Dist: psycopg[binary]>=3.1; extra == "postgis"
 Provides-Extra: bigquery
 Requires-Dist: google-cloud-bigquery>=3.11; extra == "bigquery"
 Provides-Extra: motherduck
-Requires-Dist: duckdb>=1.2; extra == "motherduck"
+Requires-Dist: duckdb>=1.5.3; extra == "motherduck"
 Provides-Extra: all
 Requires-Dist: openmapstack[bigquery,geo,motherduck,postgis,visual]; extra == "all"
 Dynamic: license-file
 
 # openmapstack
 
-**Four AI agent skills for geospatial decisions, data discovery, spatial SQL and reproducible analysis projects.**
+**AI agent skills for geospatial decisions, data discovery, spatial SQL and reproducible analysis projects.**
 
-Install (assume you have git and nodejs with npx already installed):
+Explore the [project homepage](https://jaakla.github.io/openmapstack-skills/), with bundled NYC and Tartu examples in standalone HTML and Observable Framework. The [Observable city samples](examples/observable-cities/) include reactive filters, thematic maps, downloads, and delivery evidence. Browse more projects in the [examples gallery](https://jaakla.github.io/openmapstack-samples/). Build the complete Pages site with `python3 scripts/build_project_pages.py` (Python project dependencies, Node 20+ and npm required); no warehouse credentials or Observable account are needed.
+
+Install (assuming you have git and nodejs with npx already installed):
 ```bash
 npx skills@1.5.26 add jaakla/openmapstack-skills --skill open-map-stack -g
 ```
 
-OpenMapStack gives your favorite AI agent: Claude Code, Codex, Cursor, OpenCode, PI and 50+ other agents not just a one-off analysis, but a **production workflow** from authoritative data discovery through reusable analysis (pipeline.py) to interactive web (dashboard.html) and GIS deliverables (QGIS project). The workflow becomes **inspectable** and **repeatable** as well-defined projects, defined as a `yaml` file with pinned sources, explicit assumptions and CRS choices, deterministic processing in a python script. It allows also isolated overrides, machine-readable validation, and surfaced provenance.
+OpenMapStack gives your favorite AI agent: Claude Code, Codex, Cursor, OpenCode, PI and 50+ other agents not just a one-off analysis, but a well-defined **workflow project** (`project.yaml`) defining from data discovery through reusable analysis (`pipeline.py`) to interactive web (`dashboard.html`) and selected deliverables (built-in dashboard by default; optional QGIS or Observable views). The workflow becomes **inspectable** and **repeatable** as a clearly specified project with pinned sources, explicit assumptions and CRS choices and deterministic processing in a python script. It allows also isolated overrides, machine-readable validation, and surfaced provenance.
+
+It is open-first (both data and code-wise) and cloud-native by default, built on shoulders of the established Open GIS stack: STAC for discovery; GeoParquet, COG, and PMTiles for storage and delivery; DuckDB and PostGIS for compute; and QGIS, MapLibre, and Martin for presentation. It also teaches to deploy GDAL/OGR, GeoPandas, xarray/rioxarray, PDAL, routing engines, spatial SQL, and pragmatic hosted services when scale or reliability requires them.
+
+Dashboard basemaps honor user choices and appropriate local or regional services, then fall back to preconfigured goplex.ee Protomaps vector tiles with light/dark switching and visible provider credit. See [web-delivery.md](skills/open-map-stack/references/web-delivery.md#style-and-basemap-sources-open) for the configuration.
 
-It is open-first (both data and code-wise) and cloud-native by default, built on shoulders of the awesome Open GIS stack: STAC for discovery; GeoParquet, COG, and PMTiles for storage and delivery; DuckDB and PostGIS for compute; and QGIS, MapLibre, and Martin for presentation. It also teaches to deploy GDAL/OGR, GeoPandas, xarray/rioxarray, PDAL, routing engines, spatial SQL, and pragmatic hosted services when scale or reliability requires them.
+Existing v1 manifests with a map must declare `presentation.map.basemap.kind` (`raster-xyz`, `raster-wms`, or `vector-style`). Vector basemaps require a style `url`; add the actual kind to older manifests before validation or regeneration. See the [basemap contract](skills/open-map-stack/references/project-spec.md#27-presentation-semantics).
 
 ## What's in this repo
 
-The 0.4.0 collection has four independently installable skills:
+The skill collection has four independently installable skills:
 
 | Skill | Use it for |
 |---|---|
@@ -62,21 +69,99 @@ Includes all the skills.
 - [references/formats-and-crs.md](skills/open-map-stack/references/formats-and-crs.md) - how to choose formats, conversions, projections, EPSG codes.
 - [references/processing.md](skills/open-map-stack/references/processing.md) - when and how to use GDAL/OGR, GeoPandas, xarray, DuckDB, PostGIS, PDAL and other open geo processing tools.
 - [references/analytics.md](skills/open-map-stack/references/analytics.md) — do vector/raster analytics, terrain, hydrology, network, point clouds, geocoding etc.
-- [references/web-delivery.md](skills/open-map-stack/references/web-delivery.md) — renderer selection for maps, PMTiles, MVT, Martin, TiTiler, MapLibre, deck.gl, kepler.gl, and lonboard formats and engines.
+- [references/web-delivery.md](skills/open-map-stack/references/web-delivery.md) — renderer selection for maps, tylertoo/Tippecanoe vector tiling, PMTiles, MVT, Martin, TiTiler, MapLibre, deck.gl, kepler.gl, and lonboard formats and engines.
 - [references/qgis.md](skills/open-map-stack/references/qgis.md) — QGIS desktop, plugins, PyQGIS, Processing, QGIS MCP.
 - [references/validation-and-ops.md](skills/open-map-stack/references/validation-and-ops.md) — validation, manifests, attribution, and deployment checks, including the machine-readable reproducible-project contract.
-- [references/project-spec.md](skills/open-map-stack/references/project-spec.md) — the specific`openmapstack-project/v1` schema: compiling any material analysis into a reproducible GIS project (`project.yaml`, pipeline, source provenance, overrides, validation, semantic presentation, QGIS output).
+- [references/project-spec.md](skills/open-map-stack/references/project-spec.md) — the specific`openmapstack-project/v1` schema: compiling any material analysis into a reproducible GIS project (`project.yaml`, pipeline, source provenance, overrides, validation, semantic presentation, selected delivery outputs).
 - [references/project-workflow.md](skills/open-map-stack/references/project-workflow.md) — the mandatory material-analysis workflow and delivery rules, loaded when a task needs a reproducible project.
 - [templates/](templates/) — ready scaffolds (`project.yaml`, `pipeline.py`, `presentation.yaml`, `validation.yaml`) for new projects.
 
 ### Reproducible GIS project
-Well-defined GIS project manifest and pipeline skills.
+
+Use [reproducible-gis-project](skills/reproducible-gis-project/SKILL.md) when the
+analysis method and stack are chosen and another analyst needs to inspect,
+execute and verify the result without the original conversation. It guides the
+agent to produce:
+
+- A `project.yaml` manifest with pinned sources, CRS, assumptions, ordered steps
+  and outputs, so the inputs and decisions are explicit. Its
+  [contract](skills/open-map-stack/references/project-spec.md#2-projectyaml-schema-openmapstack-projectv1)
+  permits project-specific metadata in open objects; extra fields have no built-in behavior.
+- One canonical pipeline with corrections and scenarios recorded as data or
+  executable logic, so derived results can be rebuilt without manual edits.
+- Machine-readable validation and run evidence, including a clean rerun, so
+  failures and reproducibility claims can be checked.
+- Selected web/desktop presentation derived from the same project, so maps and reports
+  reflect the validated analysis.
+
+The skill includes its own [project templates](skills/reproducible-gis-project/templates/)
+and these local references:
+
+- [project-workflow.md](skills/reproducible-gis-project/references/project-workflow.md) — required steps and delivery rules for a material analysis.
+- [project-spec.md](skills/reproducible-gis-project/references/project-spec.md) — the `openmapstack-project/v1` manifest and artifact contract.
+- [data-sources.md](skills/reproducible-gis-project/references/data-sources.md) — public data discovery and source assessment.
+- [user-data-sources.md](skills/reproducible-gis-project/references/user-data-sources.md) — warehouse access and reproducible snapshots.
+- [formats-and-crs.md](skills/reproducible-gis-project/references/formats-and-crs.md) — format choices and coordinate-system correctness.
+- [qgis.md](skills/reproducible-gis-project/references/qgis.md) — QGIS project delivery and desktop integration.
+- [validation-and-ops.md](skills/reproducible-gis-project/references/validation-and-ops.md) — checks, provenance and operational validation.
+- [installation.md](skills/reproducible-gis-project/references/installation.md) — CLI setup and local example requirements.
+- [companion-skills.md](skills/reproducible-gis-project/references/companion-skills.md) — optional product-specific help.
+
+Installing the other skills is optional. If the source, compute or delivery
+architecture is still undecided, start with `open-map-stack`.
 
 ### Geospatial data discovery
-Geodata discovery skills - open, where possible authoritative source data for global and local coverage.
+
+Use [geospatial-data-discovery](skills/geospatial-data-discovery/SKILL.md) when
+the data requirements are known and the task is to find and assess suitable
+public datasets or the user's own geospatial data. It guides the agent to:
+
+- Match coverage, time, feature meaning, attributes and resolution to the
+  question, so a convenient dataset is not mistaken for the right one.
+- Check current provider metadata, access and license terms, so availability
+  and permission claims have evidence.
+- Assess completeness and distinguish a sample or partial catalog from a full
+  extract, so missing features are not misread as absent features.
+- Report a reproducible version or snapshot strategy, retrieval details and
+  limitations, so a later analysis can use the same source.
+
+The standalone skill includes these local references:
+
+- [data-sources.md](skills/geospatial-data-discovery/references/data-sources.md) — public data providers, catalogs and discovery methods.
+- [user-data-sources.md](skills/geospatial-data-discovery/references/user-data-sources.md) — read-only warehouse discovery and snapshot rules.
+- [formats-and-crs.md](skills/geospatial-data-discovery/references/formats-and-crs.md) — formats, coordinate systems and spatial extents.
+- [companion-skills.md](skills/geospatial-data-discovery/references/companion-skills.md) — optional product-specific help.
+
+A source assessment needs no project templates, worked example or OpenMapStack
+CLI. Use `open-map-stack` when source selection depends on unresolved compute,
+storage or delivery choices; add `reproducible-gis-project` if the user wants to
+turn a chosen analysis into a full project.
 
 ### Spatial SQL
-Geospatial SQL specific skills only.
+
+Use [spatial-sql](skills/spatial-sql/SKILL.md) to write, review, debug or
+optimize a spatial query on an already chosen engine, such as PostGIS or DuckDB
+Spatial. It guides the agent to:
+
+- Check the engine, geometry types, CRS, input grain and expected result, so
+  the query uses supported functions and returns the intended entities.
+- Choose spatial predicates and boundary behavior deliberately, including how
+  null, invalid or multiply matched geometries affect the result.
+- Use appropriate metric or geodesic operations for distances, areas and
+  buffers, so a numeric threshold has the intended units.
+- Review indexes, candidate filters and query plans, then run small controls
+  when the engine is available, so performance changes preserve the result.
+
+The standalone skill includes these local references:
+
+- [spatial-sql.md](skills/spatial-sql/references/spatial-sql.md) — engine-specific spatial SQL patterns and query review.
+- [formats-and-crs.md](skills/spatial-sql/references/formats-and-crs.md) — coordinate, unit and format semantics.
+- [companion-skills.md](skills/spatial-sql/references/companion-skills.md) — optional product-specific help.
+
+A bounded query review needs no project templates, worked example or
+OpenMapStack CLI. Use `open-map-stack` when engine selection or cross-system
+architecture is undecided; add `reproducible-gis-project` if the user wants to
+compile a chosen analysis into a full project.
 
 
 ### Additional materials:
@@ -91,8 +176,9 @@ Some my local Estonia-specific guidance (Maa- ja Ruumiamet, ETAK, EPSG:3301 / L-
 
 ## Install
 
-The 0.4.0 collection provides four independently installable skills and a
-separately installed, matching CLI. From a checkout, install the generalist:
+The 0.5.0 release candidate provides four independently installable skills. The
+OpenMapStack CLI is a separate installation for project execution and validation.
+From a checkout, install the generalist and CLI for project work:
 
 ```bash
 npx skills@1.5.26 add . --skill open-map-stack -a codex -y
@@ -101,18 +187,21 @@ python -m pip install '.[geo]'
 
 Use `-a claude-code` for Claude Code, `-g` for global installation and `--copy`
 for independent copies. Select additional skills by repeating `--skill NAME`;
-`--skill '*'` selects the full collection. Each skill includes local templates,
-project schema, a trimmed worked Tartu example and CLI setup instructions.
-No installed sibling is required. Generated example outputs and downloaded
-source data are omitted; its pipeline needs network access and its documented
-GIS environment.
-
-Once the release tag and Python package are published, pin both parts of the
-coordinated release:
+`--skill '*'` selects the full collection. `open-map-stack` and
+`reproducible-gis-project` include local project templates, the project schema,
+a trimmed worked Tartu example and CLI setup instructions. The two bounded
+specialists ship their task guidance and local references; they need no CLI or
+installed sibling. Generated example outputs and downloaded source data are
+omitted; the example pipeline needs network access and its documented GIS
+environment.
+
+The following 0.5.0 commands are unavailable until the release tag and Python
+package are published. Before publication, use the matching checkout or a wheel
+built from it. After publication, pin both parts of the coordinated release:
 
 ```bash
-npx skills@1.5.26 add https://github.com/jaakla/openmapstack-skills/tree/v0.4.0/skills/open-map-stack -a codex -y
-python -m pip install 'openmapstack[geo]==0.4.0'
+npx skills@1.5.26 add https://github.com/jaakla/openmapstack-skills/tree/v0.5.0/skills/open-map-stack -a codex -y
+python -m pip install 'openmapstack[geo]==0.5.0'
 ```
 
 For a floating install use `npx skills@1.5.26 add jaakla/openmapstack-skills
@@ -143,9 +232,10 @@ and scope before changing anything. Legacy `open-gis` is a different identifier:
 remove it explicitly if it is an unwanted duplicate, rather than silently
 rewriting its lock entry.
 
-The CLI is separate from skill installation. Verify `openmapstack --version`
-against the installed skill's `metadata.version`. Before publication, a wheel
-or pinned Git commit from the matching checkout is the supported alternative.
+The CLI is separate from skill installation and is needed only for project
+execution or validation. When installing it, verify `openmapstack --version`
+against the project skill's `metadata.version`. Before publication, a wheel or
+pinned Git commit from the matching checkout is the supported alternative.
 
 ## Use
 
@@ -164,8 +254,72 @@ does not require a project; material multi-stage analyses retain the complete
 reproducibility contract. Casual place lookups and ordinary non-spatial coding
 are outside the collection's scope.
 
-See [0.4.0 release preparation](docs/release-0.4.0.md) for executed installation
-checks, consumer migration and remaining release acceptance.
+See the [0.5.0 release preparation](docs/release-0.5.0.md) and
+[candidate evidence](docs/acceptance-0.5.0-rc.json) for current checks and pending
+acceptance. The published [0.4.0 release notes](https://github.com/jaakla/openmapstack-skills/releases/tag/v0.4.0)
+retain the previous release's validation and installation evidence.
+
+## Use your established stack
+
+OpenMapStack supplies GIS methodology, source provenance and an inspectable
+project contract around existing tools. It does not replace loaders, schedulers
+or BI platforms. Preserve the user's stack and give each transformation and
+dependency one authoritative owner.
+
+The [focused and composed examples](examples/established-stack/README.md) run
+dlt file ingestion/cursor state, dbt + DuckDB Spatial models/tests, Dagster assets
+and failure handling, real Observable Framework builds, and optional native QGIS
+delivery. The default example requires only the base spatial runtime and produces
+a dashboard. Choosing one integration does not install all tools. Composed runs
+use Dagster → dlt → dbt → Observable, with optional QGIS from the same validated
+outputs. Observable hosted notebook publication is not implemented.
+
+Read [the integration boundary](skills/open-map-stack/references/established-stack.md)
+for who benefits, ownership, optional versioned bindings, clean reruns, capability
+checks and future enterprise surfaces. MotherDuck/BigQuery source connectors
+already exist; remote compute, Snowflake and Esri integrations remain extensions.
+
+## Selectable delivery profiles
+
+New project templates explicitly select the built-in dashboard using
+`delivery.schema: openmapstack-delivery/v1`. Users can replace that default with
+QGIS-only or Observable-only delivery, or choose multiple targets. Each target
+binds declared document/evidence outputs to the same validated analytical data.
+Unselected integrations have no artifact/runtime obligation; source provenance,
+CRS correctness, immutable overrides and clean reruns remain required.
+
+```yaml
+delivery:
+  schema: openmapstack-delivery/v1
+  targets:
+    - id: dashboard
+      kind: dashboard
+      output: dashboard
+      inputs: [candidate_parcels]
+      evidence: dashboard_evidence
+```
+
+`dashboard` and `dashboard_evidence` must be declared document outputs. The
+canonical pipeline embeds input-bound metadata, generates the view and writes
+its evidence before run hashes. Missing selected artifacts fail; unavailable
+browser/QGIS checks are `not_testable`. The [established-stack examples](examples/established-stack/README.md) add
+native optional QGIS generation and a real Observable Framework static build.
+Hosted Observable targets additionally retain an export, build/configuration,
+URL and retrieval identity. A URL alone is insufficient; verification does not
+certify current remote state or publish anything.
+
+Try the offline [delivery-profile example](examples/delivery-profiles/README.md),
+which demonstrates default, external-only and combined selections without
+cloud credentials. The Tartu worked example retains legacy combined delivery.
+
+**Migration is explicit:** existing v1 manifests without `delivery` retain their
+legacy presentation/QGIS obligations. Add the selection, update output/evidence
+bindings and canonical generation, deliberately remove obsolete target-specific
+dependencies/checks, and regenerate run evidence before `verify --rerun`. The CLI
+never silently migrates or deletes old outputs. Use a CLI whose check catalog
+contains `delivery.declaration_valid`; older tools that ignore the extension
+cannot certify it. Full semantics and migration instructions are in
+[project-spec.md section 2.0](skills/open-map-stack/references/project-spec.md#20-selectable-delivery--openmapstack-deliveryv1).
 
 ## Project CLI
 
@@ -210,6 +364,30 @@ openmapstack run project.yaml --json
 openmapstack inspect project.yaml --json
 ```
 
+`run` reads Python dependencies from `runtime.environment` in `project.yaml`
+and installs them into a reusable project virtualenv, including when invoked
+through `uvx`. It warns if the CLI's Python version differs from the declared
+version: `python: '3.12'` accepts any 3.12 patch release. Select a matching
+interpreter with `uvx --python 3.12 openmapstack run project.yaml`.
+
+Native dependencies such as GDAL, PROJ, QGIS, PostgreSQL, and PostGIS require
+external setup. `run` emits installation and verification guidance for declared
+native dependencies; their availability remains **unverified**. A GDAL installation
+may need particular drivers and matching Python bindings in the pipeline's runtime.
+PostGIS must be installed on the database server and enabled in the target database;
+installing a Python client is insufficient. Use a prepared system, Conda, or container
+runtime via `runtime.implementation.command` when needed. The declaration
+`gdal: DuckDB spatial extension` does not request a system GDAL installation:
+the pipeline must install/load DuckDB's `spatial` extension. See the
+[runtime contract](skills/open-map-stack/references/project-spec.md#28-runtime-runs-warnings)
+for dependency syntax and execution behavior.
+
+Runtime advisories appear on stderr, or in the JSON `warnings` array, including
+for `--dry-run`. They are non-blocking and separate from artifact validation;
+`--strict` continues to apply to validation, not these advisories. An explicit
+launch command owns its interpreter, so the CLI does not compare that interpreter
+against its own Python version.
+
 ### Sampled runs — nail it before you scale it
 
 A wide-area analysis can run for hours before a late step fails. A sampled run
@@ -244,8 +422,11 @@ See `references/project-spec.md`.
 library in `openmapstack/checks/` against what the pipeline actually produced:
 geometry read back through DuckDB Spatial, dataset CRS read from the artifact
 rather than the manifest's claim, validation evidence recomputed from the
-geodata it summarises, and QGIS project structure and runtime loading where
-PyQGIS is available.
+geodata it summarises, non-spatial `kind: table` outputs read back as tables
+(declared columns and types, unique key, CSV/XLSX downloads holding the same
+rows), QGIS project structure and runtime loading where PyQGIS is available, and
+— for projects that declare a design-language version — every dashboard or
+report page operated in a real browser.
 
 ```bash
 openmapstack verify path/to/project.yaml
@@ -255,6 +436,13 @@ openmapstack verify path/to/project.yaml --json --output validation/verify-repor
 openmapstack verify path/to/project.yaml --strict    # warnings and not-testable also return 1
 ```
 
+DuckDB Python pipelines also receive a warning-only static axis-order check.
+Reprojection or GDAL export with no explicit setup evidence produces a warning;
+every other outcome is `not_testable`, never `passed`. Static analysis cannot
+prove axis order: an independent runtime coordinate-order assertion supplies
+that guarantee. The shared Spatial setup uses GIS x/y order on each connection. See the
+[spatial SQL setup](skills/open-map-stack/references/spatial-sql.md#duckdb-spatial).
+
 These checks require no repository-owned golden answer, so they work on data
 neither this repository nor the model has seen. They establish bounded
 structural, provenance, artifact, and reproducibility predicates; they do not
@@ -273,13 +461,46 @@ checked output. A check whose dependency is missing reports `not_testable` and
 is counted separately — never a silent pass. A mixture of executed and
 `not_testable` checks has aggregate status `warning`, and every report includes
 `applicable`, `executed`, and `execution_rate` coverage. Install
-`openmapstack[geo]` for the DuckDB-backed geodata checks; PyQGIS comes from a
-system QGIS install.
+`openmapstack[geo]` for the DuckDB-backed geodata checks and for reading XLSX
+(DuckDB `excel` extension) and Parquet tables; CSV and JSON tables need nothing
+extra. PyQGIS comes from a system QGIS install. For the browser check, install
+`openmapstack[visual]` and run `python -m playwright install chromium`.
+
+The browser check runs when the manifest sets `presentation.design_language`
+(see `references/design-language.md`). It opens every declared page and
+operates each declared control through the language's state hooks, including
+controls on inactive tabs: a control must change the page, an exploratory one
+must show the exploratory label, and the reset must restore the published
+state. Projects without a language version keep their existing plan: selected
+web delivery targets receive smoke checks, and selected map-primary dashboards
+also receive the existing visual check.
+
+New material projects also generate `findings.md` from pipeline results before
+output hashing. Its readable Markdown and YAML frontmatter share findings,
+question/scope, run identity, assumptions, validation status and cited output
+definitions/hashes. CSV/Parquet/geodata remain the underlying results; the view
+need not display them all. `validate` and `verify` check a declared findings
+artifact for stale/missing evidence and consistency with the current run.
+Evidence inventories hash and link every declared table download as well as the
+primary result, so changing or removing an XLSX/JSON export invalidates findings.
+The offline example also exposes a portable findings reference in QGIS metadata.
+Try the [offline example](examples/delivery-profiles/README.md);
+see [the findings contract](skills/open-map-stack/references/project-spec.md#findings)
+for the writer API and format. Embedded chat is a separate optional enhancement.
+
+Delivered project views credit the declared author, free OpenMapStack toolkit
+and canonical analysis time. `validate` and `verify` check the metadata and
+actual HTML/QGIS credit values and links. Existing v1 projects missing both
+identity fields receive an explicit legacy warning; declaring either requires
+the complete pair and rendered credits. See the [credits contract](skills/open-map-stack/references/project-spec.md#credits-header).
 
 See [the applicability reference](docs/verify-applicability.md) for the exact
 plan conditions, dependencies, current regression evidence, and deliberate
-exclusions. In particular, browser/dashboard checks are not yet part of the
-automatic `verify` plan.
+exclusions. Explicit delivery targets add input-bound view evidence and web
+smoke checks; selected map-primary built-in dashboards also run existing visual
+checks. Legacy projects retain their previous delivery and rendering plan,
+with the explicit credits migration warning described above. These checks establish
+bounded properties, not every analytical or cartographic claim.
 
 Project-specific known answers can be declared under
 `validation.expectations[]`. The five allowlisted checks cover row count,