draftwright 0.4.16__tar.gz → 0.4.17__tar.gz

{draftwright-0.4.16 → draftwright-0.4.17}/CHANGELOG.md

@@ -4,6 +4,13 @@
 
 ### Added
 
+- 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
+  placement, compilation, projection and physical lint remain in one coordinate system; typed frame
+  refusal performs one visible top-level raw fallback. Raw remains the rollout default (#1357,
+  ADR 0020).
+
 - Referential and measured Sheet dimensions accept optional `view=` / `side=` placement
   intent. Supported pairs survive IR/compiler and generated-script round trips, constrain
   ordinary corridor candidates without exposing page coordinates, and fail clearly when a
@@ -37,11 +44,283 @@
 
 ### Changed
 
+- Accepted schema-v1 convex `Blend` chains 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
+  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
+  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).
+
+- Rectangular blind slots recognised by `b123d-recognisers` 0.4.10 now cross a dedicated
+  Draftwright semantic path: `RectangularBlindSlotFeature`, the explicit
+  `Sheet.rectangular_blind_slot(...)` word, generated-code and role-specific tolerance round
+  trips, compiler-approved width/capped-run/depth measurements, and one solver-owned
+  `OPEN SLOT … DEEP` leader. Authored subsets retain every approved value with explicit
+  `WIDE`/`LONG`/`DEEP` roles, and leaders target proved cap/side material rather than the open
+  mouth. Native raw and rigidly moved framed builds retain the same three
+  measurement outcomes. A seven-case independently authored corpus now pins all six run/open-side
+  orientations plus an independently sized specimen: each occurrence contributes separate width,
+  capped-run and depth requirements followed by exact structural correspondence and measurement
+  provenance to placed, structured-note, suppressed, dropped, missing or unverifiable outcomes.
+  Removing recognition cannot shrink that 21-requirement benchmark (#1421; ADRs 0011,
+  0013–0017, 0020).
+
+- Round-bottom blind slots recognised by `b123d-recognisers` 0.4.10 now have their own complete
+  Draftwright path rather than borrowing through-slot, pocket or rectangular-blind-slot grammar.
+  `RoundBottomBlindSlotFeature` retains the released run/opening topology and exposes the three
+  independent manufacturing requirements: capped run length, straight bottom-flat width and equal
+  side radius. `Sheet.round_bottom_blind_slot(...)`, generated code, role-specific tolerances and
+  one solver-owned `ROUND-BOTTOM OPEN SLOT …` leader all preserve those exact identities; derived
+  total width and depth are not double-counted. A seven-case independently authored corpus covers
+  all six signed principal orientations plus a separately sized specimen (21 physical requirements),
+  proves raw/framed drawing parity, and follows exact source-to-IR correspondence to placed,
+  structured-note, suppressed, dropped, missing or unverifiable outcomes without labels, views or
+  page coordinates. The released 0.4.10 API is sufficient; no recogniser change is required (#1421;
+  ADRs 0011, 0013–0017, 0020).
+
+- Turned-step completeness is now backed by an independently authored, hash-pinned 11-case STEP
+  corpus with 26 physical outside-diameter bands, 52 parameter checks and 104 downstream checks.
+  Axis line and station preserve body-local occurrence identity across automatic IR, the existing
+  public `Sheet.step(...)` DSL word, executed-and-built generated code, and exact compiler-confirmed
+  length and OD drawing evidence. Grouped equal-length chains retain every adjacent member identity
+  and their exact compiler-approved multiplier and value; native and global OD labels are checked
+  in full. Generated declarations preserve the half-thousandth midpoint of odd-thousandth spans,
+  complete body-local profile ownership is compared at its full published precision, and malformed
+  or surplus records fail closed—including coercible or non-finite groove evidence that controls
+  band ownership. The root turned-step and groove inventories must remain immutable tuples; a
+  mutable or one-shot substitute is snapshotted once only to preserve its observable requirement
+  denominator, then remains unverifiable; a non-iterable substitute retains one explicit aggregate
+  contract outcome rather than masquerading as an empty family. A profile OD may
+  satisfy one uniquely identifiable equivalent largest-band diameter
+  without multiplying across disjoint bodies; and uniquely groove-owned narrow
+  bands are excluded once by a shared fail-closed Groove schema and ownership rule used by lowering,
+  lint and the evidence ledger. That join accounts for the provider's 0.001 mm Groove coordinate
+  quantum against its eight-decimal profile axis without widening either published identity
+  (#1374; ADRs 0010, 0011, 0013, 0015, 0017, 0020).
+
+- Framed recognition now carries all three 0.4.6 step families through their complete existing
+  consumer paths. `CircularBlindStep`, `PairedRampStep` and `ThroughStep` retain the same
+  manufacturing requirements from raw recognition through their public `Sheet` words, generated
+  code, compiler-owned drawing ink and lint; the provider-owned local records and Draftwright IR
+  remain invariant under a combined non-principal rotation and translation. Circular-step
+  validation now derives its bound from each independently published endpoint, centre, radius and
+  depth value's six-significant-figure decimal cell, while outside-cell geometric inconsistencies
+  still fail closed (#1382; ADRs 0011, 0013, 0015, 0017, 0018, 0020).
+
+- Countersink completeness is now backed by an independently authored, hash-pinned seven-case STEP
+  corpus covering functional seats, solid external-cone and deburr negatives, grouped equal seats,
+  and reverse-serialized topology. Detection identity and all four provider measurements are scored
+  separately from the two actual drawing requirements. Exact aggregate object ownership follows
+  each seat once, validates the provider-selected hole semantically, and fails closed on swapped,
+  duplicated, or canonically colliding owners. It then crosses automatic IR, public `Sheet.hole`,
+  generated code, and role-specific confirmed placed diameter/angle ink without recounting the parent bore. The known second seat of a two-sided bore
+  remains explicitly unverifiable rather than hidden or heuristically attached (#1370; ADRs 0010,
+  0013, 0015, 0017).
+
+- Double-D completeness is now backed by an independently authored, hash-pinned 11-case STEP
+  corpus with ten physical profile facts, 50 provider-parameter checks and 40 downstream checks.
+  Full-frame identity preserves disconnected coaxial bodies; exact multiplicity crosses automatic
+  IR, public `Sheet.double_d_bore`, generated code and role-specific compiler-confirmed
+  `⌀major THRU DOUBLE-D across A/F` ink. Round, blind, opposed-blind, malformed, duplicated and
+  foreign-owner mutations fail closed, while the provider's exclusive aggregate ownership keeps
+  each profile from being recounted as an ordinary hole (#1370; ADRs 0010, 0011, 0013, 0015,
+  0017).
+- Updated the immutable `b123d-recognisers` pin and fail-closed inspection/capability joins to
+  0.4.10. All 28 previously consumed family schemas remain unchanged. The two new blind-slot
+  families were adopted fail-closed under #1421 and remain visible as unscored recognised
+  inventories until each gains independent completeness evidence; Draftwright does not coerce
+  their depth, terminal/open-end, or round-floor evidence into its ordinary through-slot grammar.
+  The provider's additive evidence API is not consumed by
+  this bounded adoption slice. Recognition output does change under the unchanged schemas: the
+  release closes slot-depth, subdivided paired-ramp/AngledStep, and noisy stubby-pocket gaps; fixes
+  Double-D/Hole ownership, external-cone countersink false positives, Plate tie covariance, and
+  turned-step translation covariance; and can move an edge-open rectangular recess from `Pocket`
+  to the dedicated blind-slot inventory. Draftwright regression tests pin the consumer-visible
+  ownership changes while the provider release suite owns its private predicates
+  ([provider release](https://github.com/pzfreo/b123d-recognisers/releases/tag/v0.4.10); ADRs 0013,
+  0015, 0017, 0020).
+
+- Principal-profile lint now correlates its independent extremal-face evidence with the cached
+  public `DoubleDBore` aggregate instead of importing provider-private profile helpers. Exact
+  axis/span, end-plane, centre, radius, A/F, chord, arc, and flat-direction checks fail closed;
+  occurrence mouths are consumed with matching multiplicity, and malformed records cannot earn
+  coverage. The exported low-level check now requires the aggregate projection explicitly so an
+  omitted inventory fails loudly; lint still owns its physical denominator and never reruns
+  recognition (#1411, ADRs 0013, 0015, 0017).
+
+- Whole polygonal-stock completeness is now backed by an independently authored, hash-pinned
+  13-case STEP corpus covering all principal axes, in-plane rotation, translation, topology
+  order and circular/rectangular/irregular/octagonal/attached/recessed/compound negatives. The
+  provider's exact six-side stock boundary, A/F, axial length and coupled physical flat supports
+  are scored separately from four downstream boundaries. A fail-closed two-requirement ledger
+  follows A/F and stock length through automatic IR, public `Sheet.polygonal_stock`, generated
+  code and solver-placed semantic ink; malformed or duplicate correspondence cannot borrow
+  credit from another support ring (#1371).
+
+- Multi-plate slab completeness is now backed by an independently authored, hash-pinned 11-case
+  STEP corpus covering all principal orientations, additive/cut channels, equal compound
+  occurrences, topology order and envelope/detached/thick/rotational negatives. A fail-closed
+  one-requirement ledger follows each body-local `thickness.length` through automatic IR, public
+  `Sheet.plate`, generated code and solver-placed Dimension provenance; the compiler-derived
+  opposite U-channel wall is represented explicitly without inventing duplicate ink. The corpus
+  owns 20 physical occurrences and cannot shrink when provider output is weakened or removed.
+  Exact step-level, slot-pattern, polygonal-boss and envelope ownership prevents derived material
+  spans from becoming duplicate Plate requirements while failing closed when that ownership cannot
+  be established (#1373).
+
+- Polygonal-boss completeness is now backed by an independently authored, hash-pinned STEP
+  corpus covering all three principal axes, in-plane rotation, equal compound occurrences,
+  topology order, and recess/detached/stock/other-protrusion negatives. Detection, the provider's
+  six-side schema invariant, across-flats, height, physical flat supports, and four downstream
+  boundaries are observed; a
+  two-requirement fail-closed ledger follows across-flats and boss height through the existing
+  IR/compiler/shared-solve path. The corpus owns the independent evaluation denominator; runtime
+  drawing lint remains explicitly denominated by the recognition inventory (#1372).
+
+- Updated the immutable `b123d-recognisers` pin and fail-closed manifest joins to 0.4.9. The
+  provider's body-local FaceLevel, RiserEvidence, and TurnedProfile schemas now cross an owned
+  prepared-frame boundary that normalizes, reuses one cylinder substrate, classifies locally,
+  and preserves the exact working solid/result pair. Every frame gauge and typed refusal has an
+  explicit conservative policy; refusal never hides a raw retry. Multiple physical turned
+  profiles fail explicitly where a singular adapter is required, while the unchanged raw
+  production route preserves its historical no-global-profile compound behavior with a visible
+  #1357 deferral. Production drawing remains on the explicit raw boundary until rigid-motion,
+  PMI, platform, and canary activation evidence is complete (ADR 0020, #1357).
+
+- Updated the immutable `b123d-recognisers` pin and lock to 0.4.8. Draftwright now names
+  its existing caller/world-coordinate aggregate boundary explicitly with
+  `build_raw_recognition_result`; framed recognition remains disabled until #1357's
+  frame-to-IR adapter is reviewed end to end. Installed-wheel tests exercise the public
+  framed fixes for rectangular pads, polygonal bosses, and nested multi-body plates without
+  making that API the production path (#1392).
+
+- Rectangular pads now consume `RaisedPad` schema v2 without assuming +Z. The IR, public
+  `Sheet.pad`, generated code, compiler, semantic-view planner, shared placement solve, and
+  lint correspondence preserve the exact footprint, terminal-to-attachment height, two
+  in-plane locations, axis, direction, and world bounds for all six signed principal
+  orientations. Every height uses one solver-placed `… HIGH` leader with stable measurement
+  identity; side-normal pad footprints no longer leak into the unrelated Z-level grammar
+  (#1392).
+
+- Rectangular-pad completeness is now independently evidenced by a hash-pinned 12-case STEP
+  corpus covering all six signed principal orientations, equal compound occurrences, topology
+  order, and ledge/staircase/detached-body negatives. Detection, width/length/height fidelity,
+  and four downstream boundaries are scored separately; a five-requirement fail-closed ledger
+  follows both footprint sizes, height and directional location evidence through the shared
+  compiler and placement solve (#1372).
+
+- Circular blind steps recognised by `b123d-recognisers` now have a complete consumer path.
+  The oriented terminal-to-open centreline and transverse quarter-arc lower without rescanning
+  to an explicit-only IR/Sheet feature; generated code preserves the full correspondence.
+  Radius and stopped depth remain independently addressable and tolerance-aware while sharing
+  one solver-placed end-view `R… × … DEEP` leader, and a fail-closed two-requirement ledger
+  follows each outcome without parsing labels or guessing by order (#1382).
+
+- Paired-ramp steps recognised by `b123d-recognisers` now have one complete consumer path:
+  exact IR conversion, explicit `Sheet.paired_ramp_step(...)` declaration, generated-code
+  round trip, end-view solver-placed `2× angle × run RUN` leader, independent angle/run
+  tolerances and authored suppression, and a two-requirement fail-closed completeness ledger
+  (#1382).
+
+- Rectangular through steps recognised by `b123d-recognisers` now lower without rescanning to
+  an explicit-only IR/Sheet feature. Automatic and explicit lowering support every principal
+  axis; an X/Y occurrence stays with the established face-level and shoulder/plate grammar only
+  when that grammar proves both exact leg intervals, avoiding both silent loss and duplicate
+  dimensions. The two
+  orthogonal open-section legs are independently addressable, tolerance-aware, solver-placed in
+  the end view, preserved by generated code and scored by a fail-closed two-requirement
+  completeness ledger. The through extent stays owned by the envelope (#1382).
+
+- Hole completeness is now supported by a versioned, independently authored STEP corpus rather
+  than a Draftwright-derived census. Detection, parameter fidelity and downstream usefulness stay
+  separate, while the IR adapter, public Sheet declaration, executed generated code and placed
+  drawing are each observed through their real seam. Mutations prove that a missing provider hole
+  or a broken downstream boundary lowers the relevant evidence even when internal state remains
+  self-consistent (#1369).
+- Hole-pattern completeness is now supported by a separate independently authored STEP
+  arrangement corpus covering grids, bolt circles, linear arrays, false-positive controls,
+  compound cases and topology-order variants. It scores one grouping fact per pattern while the
+  hole corpus remains the sole owner of member diameter/depth/location requirements, and follows
+  each group through the IR, public Sheet declaration, generated code and placed pattern
+  dimensions (#1370).
+- Flat completeness is now supported by an independently authored STEP corpus covering lone,
+  opposed, slanted, parallel and coaxial stock plus negative and topology-order controls. It
+  counts one physical across-flats requirement per stock line and span, then observes automatic
+  IR, public `Sheet.flat`, executed generated code and placed measurement provenance independently
+  (#1371).
+- Lone-pocket completeness is now supported by an independently authored STEP corpus covering
+  off-centre, edge-anchored, equal, opposed-opening, side-opening, compound, ownership-negative
+  and topology-order cases. Provider `open_sign` now survives the IR, declaration and generated
+  code boundaries; pocket sizes and both physical location ordinates are followed through exact
+  semantic placement evidence without counting pocket-pattern members twice (#1372).
+- Pocket-pattern completeness is now supported by a separate independently authored STEP corpus
+  covering axis-aligned and 30° linear/grid arrangements, threshold and unequal-spacing
+  negatives, a compound underside pattern and reverse-serialized topology variants. One grouped
+  count/size/lattice/location requirement reaches automatic IR, public `Sheet.pocket_pattern`,
+  generated code and exact placed ink—including approved pitch tolerances—without recounting its
+  member pockets. Diagonal pocket-pattern dimensions use exact label-ink clearance instead of a
+  false-blocking hull; generated code preserves the member opening side and grouped callouts
+  retain structured count provenance (#1372).
+- Groove completeness is now supported by an independently authored STEP corpus covering a
+  monotonic-step negative, all principal axes, a narrow circlip ownership contest, equal grooves
+  on separate bodies and reversed Boolean order. Each physical groove's width and floor diameter
+  pass independently through automatic IR, public `Sheet.groove`, executed generated code and
+  one exact compiler-backed `WIDE × ø` callout; mutations prove that recognition, identity,
+  parameters, provenance and printed syntax cannot disappear without lowering evidence (#1372).
+- Chamfer completeness is now supported by an independently authored STEP corpus covering all
+  principal axes, equal and asymmetric planar bevels, conical turned treatments, equal features
+  on separate bodies, topology-order variants, a plain negative and the AngledStep ownership
+  boundary. Each physical bevel is followed through exact IR, public `Sheet.chamfer`, executed
+  generated code, compiler measurement provenance, callout syntax and its physical leader target;
+  equal requirements may share ink without collapsing occurrence identity (#1374).
+- Fillet completeness is now supported by an independently authored STEP corpus covering all
+  principal axes, repeated equal rounds, turned toroidal treatments, separate bodies,
+  topology-order variants, a plain negative and the CircularBlindStep ownership boundary. Each
+  physical round is followed through exact IR, public `Sheet.fillet`, executed generated code,
+  compiler measurement provenance, `R`/`n× R` syntax and its physical leader target; grouped ink
+  retains every rounded edge's measurement identity (#1374).
+
+- Recognised angled blind steps now fail visibly instead of remaining an undecided inventory.
+  Each aggregate-reconciled `AngledStep` produces an actionable warning and one `unsupported`
+  completeness outcome. A frozen STEP corpus fixture pins the AngledStep/Chamfer ownership
+  boundary, while Draftwright deliberately invents no angle/run, IR, Sheet, generated-code,
+  section/detail, or annotation semantics (#1247).
+
+- Aggregate-reconciled `PrismaticPocket` occurrences now fail visibly instead of
+  remaining an unscored inventory. Each produces an actionable warning and one `unsupported`
+  completeness outcome, without repeating the provider's cross-family reconciliation or
+  inventing width/length, across-flats, IR, Sheet, generated-code, or annotation semantics (#1246).
+
+- Recognised prismatic passages now fail visibly instead of disappearing into an unscored
+  inventory. Each authoritative `SectionPassage` produces an actionable warning and an
+  `unsupported` completeness outcome; the legacy `Passage` compatibility projection is never
+  double-counted. Draftwright retains the family-level unsupported disposition rather than
+  claiming a regular-polygon callout represents the provider's complete line/arc schema (#1245).
+
 - Declared Double-D bore, countersink, chamfer, fillet, and groove geometry reads now use the
-  stable `b123d_recognisers.inspection` namespace from the exact 0.4.5 wheel. A separate
+  stable `b123d_recognisers.inspection` namespace from the exact 0.4.6 wheel. A separate
   Draftwright-owned format-1 contract fails closed on changes to the consumed signatures, result
   schemas, units, bevel rejection reasons, or cylindrical surface layout (#1362).
 
+- Updated the exact `b123d-recognisers` production lock to 0.4.6. At adoption time, circular
+  blind, paired ramp, and through-step inventories were carried from the one aggregate and
+  surfaced as unscored completeness evidence pending the #1382 semantic slices now recorded
+  above. Rectangular-grid pitch placement also uses a
+  deterministic outer witness line across plan, front, and side views despite the release's
+  coordinate-rounding differences.
+
 - `annotation_overlap` now reports the whole defect when the same pair also draws
   line-work through one of the two labels: the message names the crosser, the label
   crossed and the millimetres, and replaces the "use `label_offset_x`" remedy, which

{draftwright-0.4.16 → draftwright-0.4.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: draftwright
-Version: 0.4.16
+Version: 0.4.17
 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==0.4.5
+Requires-Dist: b123d-recognisers==0.4.12
 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'
@@ -703,11 +703,15 @@ Description-Content-Type: text/markdown
 </h1>
 
 [![CI](https://github.com/pzfreo/draftwright/actions/workflows/ci.yml/badge.svg)](https://github.com/pzfreo/draftwright/actions/workflows/ci.yml)
+[![Docs](https://github.com/pzfreo/draftwright/actions/workflows/docs.yml/badge.svg)](https://pzfreo.github.io/draftwright/)
 [![codecov](https://codecov.io/gh/pzfreo/draftwright/branch/main/graph/badge.svg)](https://codecov.io/gh/pzfreo/draftwright)
 [![PyPI](https://img.shields.io/pypi/v/draftwright.svg)](https://pypi.org/project/draftwright/)
 [![Python](https://img.shields.io/pypi/pyversions/draftwright.svg)](https://pypi.org/project/draftwright/)
+[![Downloads](https://static.pepy.tech/badge/draftwright/month)](https://pepy.tech/project/draftwright)
 [![License: AGPL v3](https://img.shields.io/badge/License-AGPL_v3-blue.svg)](LICENSE)
 [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)
+[![mypy](https://img.shields.io/badge/mypy-checked-2a6db2.svg)](https://mypy-lang.org/)
+[![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json)](https://github.com/astral-sh/uv)
 
 **[API reference](https://pzfreo.github.io/draftwright/)** ·
 **[Examples and source](https://github.com/pzfreo/draftwright#readme)**

{draftwright-0.4.16 → draftwright-0.4.17}/README.md

@@ -3,11 +3,15 @@
 </h1>
 
 [![CI](https://github.com/pzfreo/draftwright/actions/workflows/ci.yml/badge.svg)](https://github.com/pzfreo/draftwright/actions/workflows/ci.yml)
+[![Docs](https://github.com/pzfreo/draftwright/actions/workflows/docs.yml/badge.svg)](https://pzfreo.github.io/draftwright/)
 [![codecov](https://codecov.io/gh/pzfreo/draftwright/branch/main/graph/badge.svg)](https://codecov.io/gh/pzfreo/draftwright)
 [![PyPI](https://img.shields.io/pypi/v/draftwright.svg)](https://pypi.org/project/draftwright/)
 [![Python](https://img.shields.io/pypi/pyversions/draftwright.svg)](https://pypi.org/project/draftwright/)
+[![Downloads](https://static.pepy.tech/badge/draftwright/month)](https://pepy.tech/project/draftwright)
 [![License: AGPL v3](https://img.shields.io/badge/License-AGPL_v3-blue.svg)](LICENSE)
 [![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)
+[![mypy](https://img.shields.io/badge/mypy-checked-2a6db2.svg)](https://mypy-lang.org/)
+[![uv](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json)](https://github.com/astral-sh/uv)
 
 **[API reference](https://pzfreo.github.io/draftwright/)** ·
 **[Examples and source](https://github.com/pzfreo/draftwright#readme)**

{draftwright-0.4.16 → draftwright-0.4.17}/docs/adr/README.md

@@ -33,9 +33,10 @@ architecture** table; open retired or superseded records only for design history
 | [0014](0014-collect-then-solve-annotation-placement.md) | Collect-then-solve annotation placement (as built) | Collect, select, assign, and deterministically solve annotations per corridor before rendering; compatible sparse ordinary side/plan hole and post-drain machined leaders share one bounded late inventory. | Accepted; supersedes 0009; #740 Amendment 1; #1166 Amendment 2 | `test_carve_free_position_callers.py`, `test_strip_layout.py`, `test_layout_property.py`, `test_issue_740_leader_assignment.py`, `test_issue_1166_cross_pass_feature_leaders.py`, `test_import_boundaries.py` |
 | [0015](0015-part-drawing-compiler-as-built.md) | The part-drawing compiler, as built | Use one detected-or-declared feature IR and planner-fed dimension groups as the compiler waist. | Accepted; supersedes 0008 | `test_part_model.py`, `test_detect_once.py`, `test_import_boundaries.py` |
 | [0016](0016-declared-dimensioning-intent.md) | Declared dimensioning intent: capture what to measure, let the engine place it | Declare which measurements matter as scale-independent intent routed through the planner and corridor solve; never hardcode dimension geometry. | Accepted; epic #867 complete; phase 6 landed (#940) | `test_compiled_plan_boundary.py`, `test_issue_1215_no_approved_tolerance_is_dropped.py`, `test_label_provenance.py`, `test_sheet_emit.py`, `test_add_dimension.py` |
-| [0017](0017-recognition-inventory-correspondence-and-measurement-provenance.md) | One recognition result per run; correspondence is evidence-gated | Produce one external immutable recognition result held by Draftwright's per-run cache; require vertical-slice evidence before generalising correspondence, identity, requirements, outcomes, or reconciliation. | Accepted; external cache ownership clarified, extensions gated by #1018 | `test_external_recognition_boundary.py`, `test_recognition_manifest.py`, `test_declared_recognition_gate.py` |
+| [0017](0017-recognition-inventory-correspondence-and-measurement-provenance.md) | One recognition result per run; correspondence is evidence-gated | Produce one external immutable recognition result held by Draftwright's per-run cache; require vertical-slice evidence before generalising correspondence, identity, requirements, outcomes, or reconciliation. | Accepted; external cache ownership clarified, extensions gated by #1018 | `test_external_recognition_boundary.py`, `test_detect_once.py`, `test_declared_recognition_gate.py` |
 | [0018](0018-requirement-driven-view-planning-and-editable-sheet-layout.md) | Requirement-driven view planning and editable sheet layout | Use one `ViewSpec` vocabulary and planner with distinct authored `ViewConstraints` and immutable `ResolvedViewPlan`; jointly validate views, typography, convention, scale, paper and layout against requirement survival. Supersedes ADR 0004's fixed-topology assumption while retaining compose-then-pack. | Accepted 2026-08-16; partially implemented (`ViewSpec`, `ResolvedViewPlan`, `ViewCoverage`, arrangement choice and requirement gate); tracked by #1130/#1259-#1262 | Required guards are listed in the ADR |
 | [0019](0019-display-complete-labels-and-dimension-outcomes.md) | Display-complete labels and a dimension-outcome ledger | The compiled plan carries the full label text (tolerance and collapse wording included) so renderers render and never compose; dimension outcomes reconcile at one seam on both build routes; ladder rungs get per-mark identity. Finishes the ADR 0016 Amdt 1 boundary; supersedes Amdt 6's enforcement mechanism. | Proposed | `test_issue_1215_no_approved_tolerance_is_dropped.py` reduces to plan-equality when implemented |
+| [0020](0020-provider-owned-frame-boundary.md) | Provider-owned frame boundary for detected compilation | Prepare, classify, and recognise against one exact local working solid; retain source provenance, propagate typed refusals, and gate semantic direction claims by frame gauge. | Accepted; explicit opt-in activation, raw rollout default | `test_issue_1357_framed_boundary.py`, `test_issue_1357_framed_activation.py`, `test_recogniser_capabilities.py`, `test_import_boundaries.py` |
 
 ## Historical records