mostlyright-data 0.9.7__tar.gz → 0.9.8__tar.gz

{mostlyright_data-0.9.7 → mostlyright_data-0.9.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mostlyright-data
-Version: 0.9.7
+Version: 0.9.8
 Summary: Mostly Right CLI for reproducible, reviewed Parquet datasets
 Project-URL: Homepage, https://mostlyright.md/
 Project-URL: Documentation, https://mostlyright.md/docs/guides/cli/
@@ -126,7 +126,10 @@ The local execution lane runs only on Linux:
 python -m pip install "mostlyright-data[local]"
 ```
 
-It builds and checks worker inputs. It is not required for the hosted client.
+It is not required for the hosted client, and it is off by default: `mr-data` sends that work to
+the backend. What still runs on this computer is catalog authoring, the search index, and the
+commands that check a document or a run folder. A command with no backend lane refuses by name and
+says why.
 
 ## Documentation
 
@@ -138,11 +141,12 @@ It builds and checks worker inputs. It is not required for the hosted client.
 Use `mr-data --help` for the full command list and options. Commands that support
 `--json` write one JSON object.
 
-For local builds, `--input-root` and `--output` may use relative paths.
+`--input-root` and `--output` may use relative paths.
 
 ## Behavior and limits
 
-- `mr-data review --local RUN_DIR` exits 1 when a check fails.
+- `mr-data review RUN_DIR --local` runs the deterministic checks and writes no review decision. It exits 1 when a check fails. Write the flag after the command name; before it, `--local` is the lane selector.
+- `mr-data review RUN_DIR --report REPORT` seals a review decision from signed reviewer reports.
 - The weather reader accepts one extracted GRIB2 record. It refuses a multi-record GFS or HRRR file.
 - Hosted credential-free acquisition supports one single-address public HTTPS object. Bounded collections require both local attestations.
 

{mostlyright_data-0.9.7 → mostlyright_data-0.9.8}/README.md

@@ -45,7 +45,10 @@ The local execution lane runs only on Linux:
 python -m pip install "mostlyright-data[local]"
 ```
 
-It builds and checks worker inputs. It is not required for the hosted client.
+It is not required for the hosted client, and it is off by default: `mr-data` sends that work to
+the backend. What still runs on this computer is catalog authoring, the search index, and the
+commands that check a document or a run folder. A command with no backend lane refuses by name and
+says why.
 
 ## Documentation
 
@@ -57,11 +60,12 @@ It builds and checks worker inputs. It is not required for the hosted client.
 Use `mr-data --help` for the full command list and options. Commands that support
 `--json` write one JSON object.
 
-For local builds, `--input-root` and `--output` may use relative paths.
+`--input-root` and `--output` may use relative paths.
 
 ## Behavior and limits
 
-- `mr-data review --local RUN_DIR` exits 1 when a check fails.
+- `mr-data review RUN_DIR --local` runs the deterministic checks and writes no review decision. It exits 1 when a check fails. Write the flag after the command name; before it, `--local` is the lane selector.
+- `mr-data review RUN_DIR --report REPORT` seals a review decision from signed reviewer reports.
 - The weather reader accepts one extracted GRIB2 record. It refuses a multi-record GFS or HRRR file.
 - Hosted credential-free acquisition supports one single-address public HTTPS object. Bounded collections require both local attestations.
 

{mostlyright_data-0.9.7 → mostlyright_data-0.9.8}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "mostlyright-data"
-version = "0.9.7"
+version = "0.9.8"
 description = "Mostly Right CLI for reproducible, reviewed Parquet datasets"
 readme = "README.md"
 requires-python = ">=3.13,<3.14"
@@ -54,8 +54,10 @@ mostlyright-studio-client-v3 = { path = "vendor/wheels/mostlyright_studio_client
 [project.scripts]
 # ⚠ NOT `cli:main`. `cli` imports `hosted_worker`, which imports pyarrow and cryptography at module
 # scope, so `cli` cannot even be IMPORTED on a thin install. The router is the one module that can
-# run on a bare interpreter and decide which lane this invocation belongs to; with the `local`
-# extra present it hands every ordinary invocation straight to `cli.main` unchanged.
+# run on a bare interpreter and decide which lane this invocation belongs to. The local product
+# lanes are off by default, so with the `local` extra present it hands the names that still run on
+# this computer to `cli.main` unchanged, and every other name is answered by the hosted lane or
+# told, by name, that this computer no longer runs it.
 mr-data = "mostlyright.data_harness.thin.router:main"
 # ⚠ AND NOT THE MODULES THEMSELVES, for the same reason. Each of these seven is defined in a
 # module that imports pyarrow or cryptography at module scope -- or, for the catalogue job, reaches

{mostlyright_data-0.9.7 → mostlyright_data-0.9.8}/src/mostlyright/data_harness/cli.py

@@ -2260,7 +2260,29 @@ def _require_verify_url(value: str) -> str:
 
 
 def _agent_workflow_surface() -> dict[str, Any]:
-    """Return the versioned discovery surface for external data agents."""
+    """Return the versioned discovery surface for external data agents.
+
+    ``implementation_status`` says where a stage runs, and the three values partition the stages
+    rather than describing them: exactly one is true of any stage, and which one is decided by the
+    surface beside it rather than by how the stage feels.
+
+    * ``available`` -- every command the surface names runs from this command line, with no
+      backend: the kept lane-``local`` names and the portable ones.
+    * ``hosted_only`` -- the surface names only hosted-lane commands. The stage still happens, and
+      the named commands are how it is asked for and reported on; what is not true of it is that
+      this computer performs it. A hosted-lane command is not a skipped stage.
+    * ``backend_internal`` -- a backend operator step: it has a name, and nobody runs it from here.
+
+    ⚠ THE STATUS IS THE PROMISE, SO IT IS NOT DECORATION. An agent chooses its route from this
+    payload, and a stage that says ``available`` about a command which exits 2 sends the agent to
+    a refusal it will read as a defect and retry. The word has to be the answer the command would
+    actually give.
+
+    ⚠ AND A SURFACE NAMED TWICE IS CLASSIFIED ONCE. ``view`` appears under two stages; both say
+    ``hosted_only``, because the same command cannot run here in one row and on the backend in the
+    next. ``tests/test_agent_workflow_cli.py`` derives both halves of the partition from the live
+    lane table, so a command that changes lane fails there rather than drifting.
+    """
 
     return {
         "schema_version": "mr-data-agent-workflow.v4",
@@ -2269,9 +2291,9 @@ def _agent_workflow_surface() -> dict[str, Any]:
         "stages": [
             {
                 "stage": "viewer_observability",
-                "implementation_status": "available",
+                "implementation_status": "hosted_only",
                 "judgment_owner": "external_agent",
-                "deterministic_surface": "mr-data view",
+                "deterministic_surface": "mr-data watch/view",
             },
             {
                 "stage": "question_and_requirements",
@@ -2299,15 +2321,15 @@ def _agent_workflow_surface() -> dict[str, Any]:
             },
             {
                 "stage": "recipe_approval",
-                "implementation_status": "available",
+                "implementation_status": "hosted_only",
                 "judgment_owner": "human",
                 "deterministic_surface": "mr-data recipe-approve",
             },
             {
                 "stage": "acquire_clean_join_validate_package",
-                "implementation_status": "available",
+                "implementation_status": "hosted_only",
                 "judgment_owner": "deterministic_worker",
-                "deterministic_surface": "mr-data recipe-run/recipe-verify",
+                "deterministic_surface": "mr-data run/status",
             },
             {
                 "stage": "independent_review",
@@ -2317,19 +2339,19 @@ def _agent_workflow_surface() -> dict[str, Any]:
             },
             {
                 "stage": "independent_verification",
-                "implementation_status": "available",
+                "implementation_status": "hosted_only",
                 "judgment_owner": "deterministic_verifier",
                 "deterministic_surface": "mr-data verify/status/inspect",
             },
             {
                 "stage": "handoff",
-                "implementation_status": "available",
+                "implementation_status": "hosted_only",
                 "judgment_owner": "external_agent",
                 "deterministic_surface": "mr-data notebook/view",
             },
             {
                 "stage": "hosted_export",
-                "implementation_status": "available",
+                "implementation_status": "backend_internal",
                 "judgment_owner": "external_agent",
                 "deterministic_surface": "mr-data export-hosted-candidate",
             },

{mostlyright_data-0.9.7 → mostlyright_data-0.9.8}/src/mostlyright/data_harness/thin/propose.py

@@ -2560,8 +2560,8 @@ def _proposal_receipt(
     and ``mr-data build --hosted`` queues one; it is approved and the confirmation STARTED the
     Build in the same act -- Studio #124 -- and the run is named here because ``mr-data build``
     would refuse it rather than replay it; or it was declined (exit 2). Both commands carry
-    ``--hosted``, because on a full install the bare names are the local lane's. Nothing here
-    decides anything.
+    ``--hosted`` because it names the lane explicitly and remains valid: the bare names reach the
+    backend on every install. Nothing here decides anything.
     """
 
     payload = _common(client, journal, stage=stage)