draftwright 0.4.17__tar.gz → 0.4.18__tar.gz

{draftwright-0.4.17 → draftwright-0.4.18}/CHANGELOG.md

@@ -4,6 +4,116 @@
 
 ### Added
 
+- Report schema v1 now projects one recognition-owned physical requirement ledger shared with
+  completeness. Each bounded requirement identifies its accepted source occurrence(s), final IR
+  owner(s), semantic parameter where applicable, placement/satisfaction state, annotation names,
+  and representation rationale. The roster is evaluated once per report and fails closed when
+  exact same-run evidence or provenance is absent; it does not create a second recognition scan,
+  persistent topology identity, inferred manufacturing intent, or visual change (#1438, ADR 0017
+  Amendment 26).
+
+- Direct CLI rendering now writes `<output>.draftwright.json` by default, after successful visual
+  export, and prints its path with the requested artefacts. `--no-report` is the explicit opt-out;
+  library `Drawing.export()` remains free of report-writing side effects (#1438, ADR 0017
+  Amendment 25).
+
+- Generated Sheet Python now embeds a deterministic, versioned, JSON-compatible snapshot of
+  actionable accepted-occurrence gaps from the exact recognition run that produced its model.
+  The compact snapshot carries only unsupported, deferred, evidence-only, and unexpectedly
+  missing public records; STEP inputs are read once and recognition uses a private copy of the
+  exact bytes named by the embedded basename and SHA-256, with a best-effort final replay-target
+  check before writing. The hash is a later reconciliation boundary, not a lock on the mutable
+  source path. It is explicitly generation-time evidence, never current completeness authority
+  (#1438, ADR 0017 Amendment 24).
+
+- Accepted Plate occurrences now retain exact conditional ownership. Genuine multi-axis plates
+  keep their direct `PlateFeature`; other slabs are absorbed only when released same-run AAG
+  evidence joins them to the exact retained face-level, riser-shoulder, or slot-pattern owner.
+  Multi-owner claims are conjunctive, while whole-envelope, polygonal-boss, disconnected,
+  ambiguous, or malformed cases remain visibly `unexpectedly_missing` rather than borrowing an
+  owner from matching values or traversal order (#1438, ADR 0017 Amendment 23).
+
+- `Drawing.write_report(path)` now atomically writes the versioned report as deterministic,
+  strict UTF-8 JSON without coupling library export to sidecar generation. Report and filesystem
+  failures leave any existing destination untouched, with best-effort temporary cleanup that does
+  not mask the primary error (#1438, ADR 0017 Amendment 22).
+
+- Raw automatic drawings now expose `Drawing.report()`, a strict version-1 JSON-compatible
+  projection of every accepted recognition occurrence, its explicit consumer disposition and
+  report-local final IR owner(s), plus the existing lint summary. The published schema never
+  serializes provider references, topology IDs, object addresses, or transient member positions;
+  it makes requirement coverage an explicit field (subsequently populated by Amendment 26) and
+  calls a clean result `bounded-clear`, not manufacturing-ready. Paths without exact occurrence
+  ownership and raw paths with an unclassified accepted occurrence fail explicitly rather than
+  emit an empty or inferred report. Rendering remains unchanged (#1438, ADR 0017 Amendment 21).
+
+- Accepted `FaceLevel` and `RiserEvidence` occurrences now receive explicit `evidence_only`
+  outcomes. The released records remain shared substrate for the supported correlated height
+  ladder and shoulder projection, not independent inferred features or drawing requirements; no
+  IR, Sheet DSL, renderer, or recogniser API change is introduced (#1438, ADR 0017 Amendment 20).
+
+- Accepted round-boss occurrences now retain the exact existing consumer outcome: a singleton
+  `BossFeature`, one diameter-consolidated representative with transient member lineage, the exact
+  turned-step owner, or the exact groove owner reached through that step or the profile-gate
+  fallback. Ambiguous or missing correspondence fails closed without another scan, persistent ID,
+  generated artefact, or visual change (#1438, ADR 0017 Amendment 19).
+
+- Accepted through-step occurrences now retain either their exact `ThroughStepFeature` or every
+  exact final `StepLevelFeature`, `PlateFeature`, and `EnvelopeFeature` that jointly carries the
+  established compatibility projection. Plate claims are body-local, replacement is atomic, and a
+  missing or ambiguous selected owner fails closed rather than granting partial credit. One
+  occurrence can name multiple run-local IR owners without a second scan, persistent ID, generated
+  artefact, or change to rendering machinery (#1438, ADR 0017 Amendment 18).
+
+- Accepted turned-step occurrences now retain the conversion decision that already gives an
+  ordinary profile band its exact `StepFeature` or absorbs a groove-floor band into the exact
+  same-run `GrooveFeature`. The takeover must be one-to-one in both directions, so an ambiguous
+  groove, an adjacent band inside its positional tolerance, or any missing final owner fails closed
+  as `unexpectedly_missing`, with no second scan, persistent ID, generated artefact, or visual
+  change (#1438, ADR 0017 Amendment 17).
+
+- Accepted channel occurrences now receive an exact conditional consumer outcome. A genuine
+  multi-axis plate channel is `represented` by its `ChannelFeature`; a monolithic Z-depth rebate is
+  `absorbed` only when one retained body-local support carries its floor and both shoulders into the
+  final step ladder. Cross-axis, disconnected, or filtered channels fail closed as
+  `unexpectedly_missing` instead of inheriting an unrelated ladder, without a second scan,
+  persistent ID, generated artefact, or visual change (#1438, ADR 0017 Amendment 16).
+
+- Accepted countersink occurrences now receive an exact run-local outcome. A seat carried by a
+  `HoleRecord` is `absorbed` by that hole's final feature and transient member lineage; an unattached,
+  unowned, or duplicate nested occurrence fails closed without dimension matching, topology IDs, a
+  second scan, duplicate IR requirements, or visual changes (#1438, ADR 0017 Amendment 15).
+
+- Accepted angled-step, passage, prismatic-pocket, oriented-slot, and repeating-radial-profile
+  occurrences now receive exact run-local ownerless outcomes from Draftwright's existing consumer
+  capability declaration: `unsupported`, `deferred`, or report-facing `evidence_only`, with closed
+  reason codes and existing tracking issues. Equal-valued occurrences stay distinct; policy does
+  not invent IR owners, requirements, topology IDs, recognition scans, generated artefacts, or
+  visual output (#1438, ADR 0017 Amendment 14).
+
+- Accepted physical hole, slot, and pocket occurrences now retain explicit run-local ownership
+  when conversion represents a singleton or absorbs members into one same-spec hole group or
+  derived pattern feature. Family-specific reason codes and transient member lineage follow
+  member-specific AP242 splits to the exact final IR objects; missing lineage fails closed rather
+  than retaining a stale owner. Provider-derived patterns do not become invented occurrences, and
+  no report ID, extra recognition scan, generated artefact, or visual output is introduced (#1438,
+  ADR 0017 Amendment 13).
+
+- Raw automatic builds now capture run-local, conversion-time ownership from accepted provider
+  occurrences to their exact Draftwright IR features for unconditional one-to-one adapters. Equal
+  record values, order, coordinates and topology indices cannot create a binding; an omitted direct
+  binding fails closed as unexpectedly missing, while ownership rules not yet implemented remain
+  explicitly unclassified. The experimental read-only `Drawing.recognition_ownership()` exposes
+  the non-serializable ledger; it stays outside the IR/compiler and changes no generated or visual
+  artefact (#1438, ADR 0017 Amendment 12).
+
+- Raw automatic recognition now retains the provider's run-scoped `RecognitionEvidence` beside
+  its exact `RecognitionResult`, exposed through the experimental read-only
+  `Drawing.recognition_evidence()`. Declared drawings acquire the pair lazily on first physical
+  critique; bare and framed results remain evidence-less, and Draftwright never rescans merely to
+  backfill evidence. This is a non-visual reporting foundation and changes no rendered artefact
+  (#1438, ADRs 0015, 0017, 0020).
+
 - Automatic drawings can opt into the provider-owned local recognition frame with
   `framed_recognition=True`. The returned drawing exposes caller provenance, the exact downstream
   working solid, frame, and a copied framed/raw/refusal decision. PMI correlation, off-axis pattern
@@ -44,21 +154,25 @@
 
 ### Changed
 
-- Accepted schema-v1 convex `Blend` chains now cross a dedicated end-to-end consumer path:
+- Accepted schema-v3 straight or circular `Blend` chains, including concave occurrences, now cross
+  a dedicated end-to-end consumer path:
   `BlendFeature`, explicit `Sheet.blend(...)`, executable generated declarations, one
   compiler-approved `blend.radius` requirement, and a solver-owned grouped `n× R` leader. The
-  full canonical `axis_direction` remains in IR instead of being coerced to legacy Fillet
+  discriminated path kind, full canonical direction/normal, circular-path major radius,
+  rolling-ball radius, and material side remain in IR instead of being coerced to legacy Fillet
   semantics; the provider aggregate remains the sole exact Fillet/Blend precedence authority.
   Exact public-record/IR validation and completeness follow every accepted occurrence to placed,
   structured-note, authored-suppressed, dropped, missing, or unverifiable outcomes. Raw and
-  provider-framed arbitrary rigid motions retain the same radius meaning (#1433; ADRs 0011,
-  0013–0017, 0020).
-
-- Draftwright now exactly pins the immutable `b123d-recognisers` 0.4.12 release and advances its
-  fail-closed capability and inspection joins. This adoption first made the additive `Blend`,
-  `OrientedSlot`, `OrientedSlotArray`, and `OrientedSlotGrid` records visible and deferred under
-  #1430; Blend is now promoted by #1433 above, while oriented slots/patterns remain deferred and
-  recognised-but-unscored. All earlier record schemas remain compatible. The provider's newly
+  provider-framed arbitrary rigid motions retain straight-path radius meaning; framed recognition
+  also retains circular paths, while raw arbitrarily rotated circular recognition remains an
+  upstream limitation (b123d-recognisers#491) (#1433; ADRs 0011, 0013–0017, 0020).
+
+- Draftwright now exactly pins the immutable `b123d-recognisers` 0.4.14 release and advances its
+  fail-closed capability and inspection joins. `Blend` schema v3 replaces the old flat cylindrical
+  fields with exact `StraightBlendPath` and `CircularBlendPath` records and admits proved concave
+  and toroidal occurrences; Draftwright preserves those semantics through IR, `Sheet.blend(...)`,
+  generated replay, reporting, drawing, and completeness. Passage nested records advance to schema
+  v2 without changing Draftwright's unsupported passage policy. The provider's previously
   recognised shallow nonzero `PairedRampStep` keeps its existing
   two-requirement IR, public `Sheet.paired_ramp_step(...)` declaration, generated-code, compiler
   callout, solver placement, and completeness meaning (#1430; ADRs 0011, 0013–0017, 0020).

{draftwright-0.4.17 → draftwright-0.4.18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: draftwright
-Version: 0.4.17
+Version: 0.4.18
 Summary: Automated technical-drawing generation for build123d
 Project-URL: Homepage, https://github.com/pzfreo/draftwright
 Project-URL: Repository, https://github.com/pzfreo/draftwright
@@ -684,7 +684,7 @@ Classifier: Programming Language :: Python :: 3.14
 Classifier: Topic :: Scientific/Engineering
 Requires-Python: <3.15,>=3.10
 Requires-Dist: antlr4-python3-runtime<4.10
-Requires-Dist: b123d-recognisers==0.4.12
+Requires-Dist: b123d-recognisers==0.4.14
 Requires-Dist: build123d-drafting-helpers>=0.14.2
 Requires-Dist: build123d<0.11,>=0.9.0; python_full_version < '3.13'
 Requires-Dist: build123d<0.12,>=0.11; python_full_version >= '3.13'
@@ -740,13 +740,14 @@ Point it at a STEP file — that's the whole workflow:
 
 ```
 draftwright my_part.step --title "Mounting Plate" --number DWG-001
-# writes my_part.pdf (the default)
+# writes my_part.pdf and my_part.draftwright.json (the defaults)
 ```
 
 Choose formats, scale, and page; or emit an editable drawing script:
 
 ```
 draftwright my_part.step --format pdf,dxf     # also: svg, all
+draftwright my_part.step --no-report          # opt out of the JSON sidecar
 draftwright my_part.step --scale 2 --page A3  # override the auto scale / page
 draftwright my_part.step --scale 1 --page A4 --scale-policy strict
 draftwright my_part.step --script             # write an editable declarative Sheet script
@@ -794,7 +795,7 @@ and removed in 0.5.0 (see [docs/deprecations.md](docs/deprecations.md)).
 Instead of relying on detection, **reference the build123d objects you built** and declare
 only the drawing intent — geometry supplies the sizes (⌀ read off the object), you supply
 tolerances, fits, datums, GD&T, and surface finish. The fluent `Sheet` skips detection and
-dimensions exactly what you declare (ADR 0011):
+dimensions exactly what you declare (ADR 4 (was 0011)):
 
 ```python
 from build123d import Box, Cylinder, Pos
@@ -810,7 +811,7 @@ hole = sheet.hole(bore).finish("1.6").note("M3x0.5 TAP")  # ⌀8 bore, Ra 1.6, t
 sheet.control(hole).position(0.1, to="A", diameter=True)  # ⌀0.1 position wrt A
 
 sheet.authored_dimensions()                               # THIS is the complete set
-sheet.dimension(env, "width.length")                      # (ADR 0016: a build says where
+sheet.dimension(env, "width.length")                      # (ADR 4 (was 0016): a build says where
 sheet.dimension(env, "height.length")                     #  its dimensions come from, and
 sheet.dimension(hole, "bore.diameter")                    #  omitting one means suppress it)
 
@@ -857,7 +858,7 @@ draftwright mymodule:thumbwheel --script
 
 The emitted values are *detected* off the geometry — honest, and a good starting point. Since
 you have the objects, swap each numbered line for a reference so the object stays the single
-source of truth (ADR 0011 — the size is read off the object, no numbers restated):
+source of truth (ADR 4 (was 0011) — the size is read off the object, no numbers restated):
 
 ```python
 # generated (detected):
@@ -991,7 +992,7 @@ generation, the lint→repair loop, and collect-then-solve placement).
 
 ## Architecture
 
-draftwright is structured as a **part-drawing compiler** (ADR 0015): recognised
+draftwright is structured as a **part-drawing compiler** (ADR 1 (was 0015)): recognised
 or declared features converge on a `PartModel` IR, then planner-fed and
 sanctioned model-routed render intents feed shared placement, projection, and
 export. Coverage lint independently compares recognised geometry with the

{draftwright-0.4.17 → draftwright-0.4.18}/README.md

@@ -40,13 +40,14 @@ Point it at a STEP file — that's the whole workflow:
 
 ```
 draftwright my_part.step --title "Mounting Plate" --number DWG-001
-# writes my_part.pdf (the default)
+# writes my_part.pdf and my_part.draftwright.json (the defaults)
 ```
 
 Choose formats, scale, and page; or emit an editable drawing script:
 
 ```
 draftwright my_part.step --format pdf,dxf     # also: svg, all
+draftwright my_part.step --no-report          # opt out of the JSON sidecar
 draftwright my_part.step --scale 2 --page A3  # override the auto scale / page
 draftwright my_part.step --scale 1 --page A4 --scale-policy strict
 draftwright my_part.step --script             # write an editable declarative Sheet script
@@ -94,7 +95,7 @@ and removed in 0.5.0 (see [docs/deprecations.md](docs/deprecations.md)).
 Instead of relying on detection, **reference the build123d objects you built** and declare
 only the drawing intent — geometry supplies the sizes (⌀ read off the object), you supply
 tolerances, fits, datums, GD&T, and surface finish. The fluent `Sheet` skips detection and
-dimensions exactly what you declare (ADR 0011):
+dimensions exactly what you declare (ADR 4 (was 0011)):
 
 ```python
 from build123d import Box, Cylinder, Pos
@@ -110,7 +111,7 @@ hole = sheet.hole(bore).finish("1.6").note("M3x0.5 TAP")  # ⌀8 bore, Ra 1.6, t
 sheet.control(hole).position(0.1, to="A", diameter=True)  # ⌀0.1 position wrt A
 
 sheet.authored_dimensions()                               # THIS is the complete set
-sheet.dimension(env, "width.length")                      # (ADR 0016: a build says where
+sheet.dimension(env, "width.length")                      # (ADR 4 (was 0016): a build says where
 sheet.dimension(env, "height.length")                     #  its dimensions come from, and
 sheet.dimension(hole, "bore.diameter")                    #  omitting one means suppress it)
 
@@ -157,7 +158,7 @@ draftwright mymodule:thumbwheel --script
 
 The emitted values are *detected* off the geometry — honest, and a good starting point. Since
 you have the objects, swap each numbered line for a reference so the object stays the single
-source of truth (ADR 0011 — the size is read off the object, no numbers restated):
+source of truth (ADR 4 (was 0011) — the size is read off the object, no numbers restated):
 
 ```python
 # generated (detected):
@@ -291,7 +292,7 @@ generation, the lint→repair loop, and collect-then-solve placement).
 
 ## Architecture
 
-draftwright is structured as a **part-drawing compiler** (ADR 0015): recognised
+draftwright is structured as a **part-drawing compiler** (ADR 1 (was 0015)): recognised
 or declared features converge on a `PartModel` IR, then planner-fed and
 sanctioned model-routed render intents feed shared placement, projection, and
 export. Coverage lint independently compares recognised geometry with the

draftwright-0.4.18/docs/adr/README.md

@@ -0,0 +1,48 @@
+# Architecture decision records
+
+Five live records, one per core aspect of Draftwright. Read them before changing anything they
+govern. Each states a decision, then a numbered list of invariants an agent must not violate,
+each naming the test that fails if it is — that list is the reason the document exists.
+
+| ADR | Owns | Guard tests (representative) |
+| --- | --- | --- |
+| [1 — The compiler pipeline](0001-compiler-pipeline.md) | One engine, the `PartModel` IR waist, the module DAG, single-owner build state, the compiled plan fed to placement | `test_import_boundaries.py`, `test_drawing_encapsulation.py`, `test_detect_once.py`, `test_compiled_plan_boundary.py` |
+| [2 — Sheet layout and view planning](0002-sheet-layout-and-view-planning.md) | Requirement-driven view planning, compose-then-pack, collect-then-solve placement, Policy B, scale-before-sheet | `test_carve_free_position_callers.py`, `test_layout_cleanliness.py`, `test_issue_1146_scale_completeness.py`, `test_issue_1260_view_constraints.py` |
+| [3 — The recognition boundary](0003-recognition-boundary.md) | External geometry-only recognition, one run per build, the fail-closed provider join, occurrence ownership, the framed boundary | `test_recogniser_capabilities.py`, `test_declared_recognition_gate.py`, `test_detect_registry.py`, `test_issue_1357_framed_boundary.py` |
+| [4 — Declared intent](0004-declared-intent.md) | The IR as public input, authored dimension sets, suppression by omission, the compiled-plan boundary, one declarative script | `test_declare.py`, `test_label_provenance.py`, `test_parameter_id.py`, `test_sheet_identity_invariant.py` |
+| [5 — Trust and honest failure](0005-trust-and-honest-failure.md) | Determinism, lint as an independent judge, provenance, reports that refuse rather than shrink, absences reported | `test_export_reproducible.py`, `test_render_seam.py`, `test_issue_1215_no_approved_tolerance_is_dropped.py`, `test_issue_1438_report_projection.py` |
+
+`archive/` holds the twenty records these replaced, frozen, with their filenames and numbers
+unchanged. Each is stamped with the live record that now owns its decision. They are the
+why-trail; nothing in them is a work instruction.
+
+## Rules of engagement
+
+These exist because the previous corpus reached 21 files, 8,787 lines and one record with 28
+amendments, and its own "at roughly four amendments, write a successor" rule was a convention
+that nothing enforced. Each rule below is enforced by `tests/test_adr_corpus.py` where a test can
+enforce it.
+
+1. **A record changes only when a boundary or invariant changes.** Adopting a provider version,
+   adding a feature family, recording ownership for one more record type, adding a report field:
+   that is a PR body, not an amendment. If the change does not alter the Invariants section, it
+   does not touch the record.
+2. **No record text is written without the maintainer's sign-off.** An agent or contributor who
+   believes a record needs to change says so in the PR in two sentences and waits. A reviewer
+   recommending an ADR is a recommendation, not an authorisation.
+3. **Each live record is at most 200 lines; the five together at most 1,000.** Enforced.
+4. **No live document cites an archived record as its authority.** Code, tests and docs cite
+   ADR 1–5. A pointer into the archive is written `ADR n (was 00NN …)` so the history stays
+   findable while the live number is the one a reader acts on. A bare `ADR 00NN` outside
+   `archive/` fails the guard.
+5. **Every invariant names its guard.** One without a test is listed under *Unguarded* with the
+   reason, never mixed into the numbered list. The guard checks that every named test module and
+   test function exists. That list is expected to shrink.
+6. **Cite symbols and test names, never source line numbers.**
+
+## Reading order
+
+Pipeline first (1), then whichever aspect the change touches. Layout (2) and declared intent (4)
+each assume the pipeline; recognition (3) and trust (5) each assume the other four's boundaries.
+
+The consolidation itself is recorded in `docs/plans/adr-consolidation-plan.md`.

{draftwright-0.4.17 → draftwright-0.4.18}/docs/reference/recogniser-capabilities.md

@@ -188,7 +188,7 @@ directional datum locations. Edge-anchored pockets intentionally require only th
 because the adjacent stock edges establish position. Pattern members are excluded rather than
 double-counted. Each fact must survive automatic IR, public `Sheet.pocket`, executed generated
 Sheet code and placed size/location provenance. The IR now retains provider `open_sign`, and the
-shared location renderer records which physical X/Y ordinate landed while preserving ADR 0016's
+shared location renderer records which physical X/Y ordinate landed while preserving ADR 4 (was 0016)'s
 single public `location` authoring unit.
 
 ## Pocket-pattern completeness evidence
@@ -485,11 +485,12 @@ physical corpora.
 ## Historical: recognisers 0.4.8 raw boundary and RaisedPad v2
 
 Draftwright pinned the published `b123d-recognisers==0.4.8` wheel in the preceding release.
-The default route deliberately calls `build_raw_recognition_result`; records remain in the
-caller/world coordinate system. #1357 now provides a reviewed explicit framed-result activation
-added with 0.4.9. The 0.4.8 tests used the public framed API as release evidence for upstream #331, #332,
-and #334, comparing each aggregate inventory with the corresponding public family call on the exact
-returned local solid. Production added no fallback, second aggregate, or family rescan. Face levels,
+At that time the default route deliberately called `build_raw_recognition_result`; records
+remained in the caller/world coordinate system. #1357 now provides a reviewed explicit
+framed-result activation added with 0.4.9. The 0.4.8 tests used the public framed API as release
+evidence for upstream #331, #332, and #334, comparing each aggregate inventory with the
+corresponding public family call on the exact returned local solid. Production added no fallback,
+second aggregate, or family rescan. Face levels,
 risers, and turned profiles remained outside framed production pending their subsequently released
 upstream fixes.
 
@@ -552,9 +553,30 @@ with exact full-record correspondence and parameter/outcome provenance under the
 rules as the rectangular family. It now participates independently in `audited_score`; ordinary
 slots, pockets, channels and rectangular blind slots do not share ownership.
 
-## Recognisers 0.4.12 adoption and additive-family boundary
-
-Draftwright exactly pins `b123d-recognisers==0.4.12`, including the immutable 0.4.11 `Blend`
+## Recognisers 0.4.14 adoption and path-complete Blends
+
+Draftwright exactly pins `b123d-recognisers==0.4.14`; the provider family count remains 33 and
+the inspection API remains format 1 / major 1. The fail-closed record join now accepts `Blend`
+schema v3 with nested `StraightBlendPath` and `CircularBlendPath` schema-v1 records. It also
+accepts `PassageEnds`, `PassageSection`, and `SectionPassage` schema v2. Passage remains explicitly
+unsupported under #1245: accepting its released structural schema does not invent an IR feature,
+Sheet declaration, or completeness credit.
+
+The supported Blend adapter preserves the complete discriminated path. Straight occurrences keep
+their canonical line direction and anchor; circular occurrences keep their centre, canonical
+normal, and major path radius; both keep rolling-ball radius and proved convex/concave material
+side. `BlendFeature`, `Sheet.blend(...)`, generated replay, report projection, renderer, and
+completeness all carry those facts. The provider aggregate remains the only Fillet/Blend precedence
+authority, and Draftwright does not rescan or reinterpret detached faces. Circular leaders combine
+the two radii with analysis-owned coaxial cylinder candidates, then use exact same-run
+occurrence/IR ownership and the provider-issued defining face to select the physical surface
+tangency instead of the rolling path or its centre. A declaration without that face authority
+refuses competing support radii rather than choosing another body's cylinder; the shared placement
+solve still owns the page position.
+
+## Historical 0.4.12 adoption and additive-family boundary
+
+Draftwright initially pinned `b123d-recognisers==0.4.12`, including the immutable 0.4.11 `Blend`
 addition and the 0.4.12 `OrientedSlot`, `OrientedSlotArray`, and `OrientedSlotGrid` records. The
 provider capability manifest therefore contains 33 families. Every family and public record has an
 explicit consumer declaration and registry home; the inspection API remains format 1 / major 1.
@@ -578,9 +600,33 @@ recognition to shallow nonzero ramp pairs. A released shallow specimen now cross
 supported consumer path without changing meaning: exact record-to-IR lowering, the existing
 `Sheet.paired_ramp_step(...)` declaration, executed generated code, two compiler-approved angle/run
 requirements, one solver-owned compound leader, and the established completeness ledger. Existing
-raw and provider-framed contract suites continue to guard the ADR 0020 boundary. Draftwright uses no
+raw and provider-framed contract suites continue to guard the ADR 3 (was 0020) boundary. Draftwright uses no
 provider-private geometry or sibling checkout for any of these claims.
 
+#1438 first consumed the release's public raw `b123d_recognisers.evidence` acquisition. One
+`RecognitionEvidence` owns the accepted physical occurrence/face references and projects the same
+`RecognitionResult` the existing compiler and lint consumers already read. Draftwright retains
+both in its per-drawing `RecognitionCache`, including across scale/view retries; declared builds
+remain recognition-free until physical critique, and a bare result is never rescanned merely to
+backfill evidence. Recognisers 0.4.14 publishes the framed-evidence contract requested by
+[`b123d-recognisers#463`](https://github.com/pzfreo/b123d-recognisers/issues/463), but Draftwright's
+framed route remains evidence-less until the independent #1438 consumer slice adopts it; it does
+not infer cross-frame occurrence identity in the interim.
+
+Round-boss ownership is now explicit at that same run-local boundary. Each accepted `BossRecord`
+retains the exact consumer-selected final owner: its singleton `BossFeature`, the one existing
+same-diameter representative, its body-local turned step, or the groove that absorbs that step or
+the profile-gate fallback. The occurrence's own geometry remains available even when the drawing
+uses a consolidated owner. Ambiguous selection fails closed as `unexpectedly_missing`; it cannot be
+certified by equal values, ordering, rendered labels, or topology identity.
+
+`FaceLevel` and `RiserEvidence` occurrences now receive explicit `evidence_only` outcomes. This
+follows the released provider manifest: both are role=`evidence`, deliberately absent from the
+feature census, and consumed as shared substrate for Draftwright's correlated `StepLevelFeature`
+height/shoulder projection. The aggregate face-level and riser families remain supported; the
+outcome says only that each raw evidence record is not itself an independent inferred feature or
+completeness requirement. It adds no Sheet word, IR adapter, drawing ink, or provider API.
+
 ## Recognisers 0.4.9 prepared frame boundary
 
 Draftwright's 0.4.9 boundary accepted `RiserEvidence` v2,
@@ -588,7 +634,7 @@ Draftwright's 0.4.9 boundary accepted `RiserEvidence` v2,
 same-solid levels and physical turned-profile membership in compounds. The fail-closed capability
 join rejects an older, future, missing, or malformed schema instead of treating it as equivalent.
 
-ADR 0020 adds `prepare_framed_detection`, the one Draftwright-owned intake for the public
+ADR 3 (was 0020) adds `prepare_framed_detection`, the one Draftwright-owned intake for the public
 `prepare_framed_part` seam. A successful unit retains the caller-space source as provenance while
 pairing the exact local working solid, provider frame, prepared cylinders, Draftwright's local
 rotational classification, and aggregate. Classification reuses the prepared cylinder objects;
@@ -649,12 +695,12 @@ profile warning only for that occurrence, so an unrelated unrecognised profile r
 
 ## Passage compatibility boundary
 
-The installed `b123d-recognisers==0.4.8` release contains the `passages` family introduced
-in 0.2.6. Version 0.4.0 makes `SectionPassage` the authoritative physical output and retains
-schema-v1 `Passage` as its compatibility projection. Draftwright declares all six public and nested
-record schemas exhaustively but deliberately keeps the family `unsupported`, with the drafting
-decision recorded by issue #1245. This is a truthful consumer disposition: both aggregate
-inventories remain visible, but only authoritative `section_passages` contributes an explicitly
+The installed `b123d-recognisers==0.4.14` release contains the `passages` family introduced
+in 0.2.6. Version 0.4.0 made `SectionPassage` the authoritative physical output; 0.4.14 publishes
+its nested schema v2 and retains `Passage` as a compatibility projection. Draftwright declares all
+six public and nested record schemas exhaustively but deliberately keeps the family `unsupported`,
+with the drafting decision recorded by issue #1245. This is a truthful consumer disposition: both
+aggregate inventories remain visible, but only authoritative `section_passages` contributes an explicitly
 `unsupported` completeness requirement. Draftwright does not invent an IR feature, DSL
 declaration, generated code, or drawing annotation for either inventory.
 
@@ -667,7 +713,7 @@ The 0.4 contract is explicit:
 - rich split-junction passages can supersede a Slot claim, moving physical ownership to the
   unsupported Passage family through `SLOT_SUPERSEDED_BY_PASSAGE`.
 
-The exact 0.4.8 pin, manifest-v2 validator and explicit unsupported inventories make that limitation
+The exact 0.4.14 pin, manifest-v2 validator and explicit unsupported inventories make that limitation
 fail-visible rather than silently treating rich passages as supported. Draftwright deliberately
 does not claim that a regular-polygon `HEX … A/F THRU` callout covers the complete line/arc section
 schema. Each authoritative `RecognitionResult.section_passages` occurrence therefore emits

{draftwright-0.4.17 → draftwright-0.4.18}/docs/reference/sheet.md

@@ -122,27 +122,35 @@ callout. More general ordered operation stacks remain tracked by issue #1360. Un
 requirement has a structured parameter, use a feature-linked `note(..., satisfies=(...))` only
 for parameter ids the handle actually exposes; free prose does not satisfy coverage.
 
-### Convex Blend chains
+### Straight and circular Blend chains
 
-When recognition accepts a complete schema-v1 convex Blend chain that is not superseded by a
-dimension-worthy Fillet, it becomes one radius requirement. Declare the same meaning explicitly
-with the dedicated word:
+When recognition accepts a complete schema-v3 straight or circular rolling-ball Blend path that is
+not superseded by a dimension-worthy Fillet, it becomes one radius requirement. Declare the same
+meaning explicitly with the dedicated word:
 
 ```python
 blend = sheet.blend(
-    axis="z",                         # dominant component only
+    axis="z",                         # canonical dominant component (x/y/z tie-break)
     axis_direction=(0.321394, 0.383022, 0.866025),
     radius=0.2,
-    at=(12.345, -4.5, 6.789),         # physical leader anchor in part coordinates
+    at=(12.345, -4.5, 6.789),         # straight anchor / circular centre in part coordinates
     side="convex",
 )
 sheet.dimension(blend, "blend.radius")
 ```
 
-`axis_direction` is authoritative and may be non-principal; it is preserved by generated Sheet
-code. `at` identifies feature geometry, not a page position. The annotation placement solve owns
-the final leader and label coordinates. The word is explicit-only because a detached cylindrical
-face cannot prove a complete chain or the provider aggregate's Fillet precedence.
+For a straight path, `at` is a point on its analytic line and `axis_direction` is the canonical
+line direction. A circular path uses `path_kind="circular"`, treats `at` as its centre and
+`axis_direction` as its normal, and requires `path_radius=` for the centre-line circle's major
+radius. `radius` remains the rolling-ball radius; `side` may be `"convex"` or `"concave"`.
+`axis` must match the same first-maximum `x`/`y`/`z` tie-break used by automatic conversion, so an
+explicit declaration has the same exact occurrence identity and view routing as its released
+record.
+Generated Sheet code preserves every field.
+
+These are part-space feature coordinates, never page positions. The annotation placement solve
+owns the final leader and label coordinates. The word is explicit-only because detached surface
+geometry cannot prove a complete chain or the provider aggregate's Fillet precedence.
 
 ## View handle
 

{draftwright-0.4.17 → draftwright-0.4.18}/docs/research/1062-repeating-radial-profile-evidence.md

@@ -77,17 +77,17 @@ discovery does not invent them.
 
 ## ADR Fit
 
-- **ADR 0013:** the proven repeat is lower-tier, geometry-only evidence. `gear` is an
+- **ADR 3 (was 0013):** the proven repeat is lower-tier, geometry-only evidence. `gear` is an
   application semantic and must not be smuggled into the recognition record.
-- **ADR 0015:** any future automatic record must enter through the compiler's existing
+- **ADR 1 (was 0015):** any future automatic record must enter through the compiler's existing
   recognition waist. A declared gear requirement belongs on the declared path.
-- **ADR 0017:** if productionised, one orchestration owns the correspondence evidence and
+- **ADR 3 (was 0017):** if productionised, one orchestration owns the correspondence evidence and
   consumers project from it. The mutation precedes trust in the guard, and a new identity
   scheme is not justified by this single case.
-- **ADR 0011:** explicit author intent is the appropriate source for requirements that the
+- **ADR 4 (was 0011):** explicit author intent is the appropriate source for requirements that the
   B-rep cannot establish.
 
-No ADR amendment is required. No placement decision is made, so ADR 0014 is unaffected.
+No ADR amendment is required. No placement decision is made, so ADR 2 (was 0014) is unaffected.
 
 ## Follow-up Slices
 

{draftwright-0.4.17 → draftwright-0.4.18}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "draftwright"
-version = "0.4.17"
+version = "0.4.18"
 description = "Automated technical-drawing generation for build123d"
 readme = "README.md"
 license = { file = "LICENSE" }
@@ -18,7 +18,7 @@ dependencies = [
     "build123d>=0.9.0,<0.11 ; python_full_version < '3.13'",
     "build123d>=0.11,<0.12 ; python_full_version >= '3.13'",
     "build123d-drafting-helpers>=0.14.2",
-    "b123d-recognisers==0.4.12",
+    "b123d-recognisers==0.4.14",
     "numpy>=1.24",
     # PNG raster export (SVG→PDF→PNG). Both permissively licensed (Pillow HPND; pypdfium2
     # Apache-2.0 wrapping Google PDFium BSD-3) and ship pre-built wheels with NO native system
@@ -162,6 +162,7 @@ docs = [
 dev = [
     "codespell>=2.3.0",
     "hypothesis>=6",
+    "jsonschema>=4,<5",
     "mypy>=2.1.0",
     "pytest>=8",
     "pytest-cov>=5",

{draftwright-0.4.17 → draftwright-0.4.18}/src/draftwright/__init__.py

@@ -28,6 +28,11 @@ _LAZY = {
     "make_drawing": "draftwright.builder",
     "Drawing": "draftwright.drawing",
     "FeatureInfo": "draftwright.drawing",
+    "ReportUnavailableError": "draftwright.reporting",
+    "INSPECTION_SCHEMA": "draftwright.inspection",
+    "INSPECTION_SCHEMA_VERSION": "draftwright.inspection",
+    "InspectionUnavailableError": "draftwright.inspection",
+    "inspect_step": "draftwright.inspection",
     "Sheet": "draftwright.sheet",
     "lint_feature_coverage": "draftwright.linting",
     "PmiExtractionReport": "draftwright.pmi",
@@ -85,6 +90,12 @@ if TYPE_CHECKING:  # static analysers / IDEs — no runtime import, no kernel co
     from draftwright.builder import ScaleIncompatibilityError, build_drawing, make_drawing
     from draftwright.compose import choose_scale
     from draftwright.drawing import Drawing, FeatureInfo
+    from draftwright.inspection import (
+        INSPECTION_SCHEMA,
+        INSPECTION_SCHEMA_VERSION,
+        InspectionUnavailableError,
+        inspect_step,
+    )
     from draftwright.linting import lint_feature_coverage
     from draftwright.pmi import (
         PmiExtractionReport,
@@ -93,6 +104,7 @@ if TYPE_CHECKING:  # static analysers / IDEs — no runtime import, no kernel co
         extract_pmi,
         extract_pmi_report,
     )
+    from draftwright.reporting import ReportUnavailableError
     from draftwright.sheet import Sheet
     from draftwright.view_plan import ViewConstraints, ViewPlanIncomplete, ViewSpec
 
@@ -105,12 +117,16 @@ def __dir__():
 
 
 __all__ = [
+    "INSPECTION_SCHEMA",
+    "INSPECTION_SCHEMA_VERSION",
     "Drawing",
     "SoftDeprecationWarning",
     "FeatureInfo",
+    "InspectionUnavailableError",
     "PmiExtractionReport",
     "PmiRecord",
     "PmiSourceEntity",
+    "ReportUnavailableError",
     "Sheet",
     "ScaleIncompatibilityError",
     "ScaleCompletenessWarning",
@@ -121,6 +137,7 @@ __all__ = [
     "choose_scale",
     "extract_pmi",
     "extract_pmi_report",
+    "inspect_step",
     "lint_feature_coverage",
     "make_drawing",
 ]