draftwright 0.4.27__tar.gz → 0.4.29__tar.gz

{draftwright-0.4.27 → draftwright-0.4.29}/CHANGELOG.md

@@ -2,6 +2,259 @@
 
 ## Unreleased
 
+### Added
+
+- Circular-seat recognition, declared features, and generated-script replay (#1613).
+  Diameter, open run, arc sweep, and three datum-to-axis coordinates retain each physical
+  seat's measurement identities, including counted coaxial callouts and shared axis marks.
+- Opt-in independent sheet margins and exact title-block width on `Sheet`, `build_drawing`,
+  `make_drawing`, and the CLI, including generated-script replay (#1612, #1589). Sergio's
+  25/10/10/10 mm margins and 175 mm block are explicit options; existing defaults remain.
+
+### Fixed
+
+- Direct and generated builds use the same final IR step ladder for page layout (#1592).
+  Raw face levels owned by another feature no longer reserve a phantom step corridor at
+  the title-block clearance threshold.
+
+- Location intents now accept `.format(decimals=...)` for all selected directions (#1610),
+  including plan locations, side-drilled holes and slots. Compiler-owned labels reach
+  placement footprints, live location edits and generated scripts. Coincident marks with
+  incompatible explicit display text are refused instead of overriding an intent.
+
+- Explicit dimension precision, including `.format(decimals=0)`, now reaches geometric
+  label critique (#1620). Correct rounding is reported as information; labels outside
+  the requested rounding interval still produce `label_vs_measured`. Height and step
+  ladders retain the same compiler policy as ordinary dimensions.
+
+- **`label_vs_measured` no longer reports the drawing's own rounding as a topology error
+  (#1600).** A sheet at one decimal place prints 4.450 as `4.5`; the check then compared
+  `4.5` against 4.450 and said "possible axis swap or wrong endpoint". The comparison was
+  *relative*, so the same 0.05 mm of rounding passed silently on a 100 mm dimension and was
+  an error on a 4 mm one — it fired on smallness, not on wrongness.
+
+  A label now states its own precision: `4.5` shows one place, `4.45` two, `12` none. A
+  difference within half of the last displayed place is the label doing its job; anything
+  beyond it is still reported, and an axis swap or wrong endpoint misses by far more.
+
+  This is a prerequisite for #1602's proposed CI gate: `label_vs_measured` sits in the
+  register of codes meaning *the sheet says something untrue*, and a code that fires on
+  correct rounding cannot be hard-failed on.
+
+### Added
+
+- **`nominal_rounded` says where the sheet's precision loses the model value.** Silencing
+  the false alarm above left nothing reporting it at all, and `4.5` for 4.450 is still a
+  real 0.05 mm between drawing and model — on a part carrying 0.05 mm clearances, the whole
+  clearance. One `info` per sheet, naming the worst case:
+
+  > 7 dimension(s) print a nominal the model does not have, rounded to the sheet's
+  > precision; the largest is '15.7' for 15.6500 (0.0500 mm). Within a general tolerance
+  > this is ordinary; where a fit depends on it, raise that dimension's places with
+  > `.format(decimals=...)`
+
+  `.format(decimals=…)` cannot yet reach a **location** dimension (#1610), and on the
+  fixture this was measured against six of the seven offenders are locations — so the
+  message says so rather than prescribing a remedy that raises `ValueError`.
+
+  It does **not** guess, and does not change any rounding. `4.450` locating a hinge axis is
+  design intent; `71.595` as an overall envelope is parametric fallout no drafter would
+  print in full — and telling them apart needs to know what the part is *for*, which a STEP
+  file does not carry (the same gap as #1597). So the engine reports and the author decides:
+  `.format(decimals=…)` already sets places per dimension and survives the `--script` round
+  trip.
+
+  Measured: 7 of 25 numeric dimension labels on `whistle_frame_reference.step`; silent on
+  every golden fixture, all of which are built from round numbers.
+
+### Fixed
+- Adopt published Quiddity 0.2.9 (#1619). The frame document now retains six separate
+  hinge bores instead of a single path spanning air gaps. Three cylindrical seat occurrences
+  are reported explicitly; their drawing grammar remains pending #1613.
+- Refused hole patterns retain their members in the hole-completeness ledger (#1607).
+  The adapter records the refusal against the exact recognition run, so a six-hole group
+  is no longer reported as an unverifiable four-hole pattern and two loose holes. Declared
+  pattern mismatches still undergo the independent correspondence checks.
+
+- **A required dimension with nowhere to go now re-plans the sheet instead of being
+  reported and left (#1590).** The automatic recovery ladder — larger scales on the
+  selected page, then dropping the optional pictorial, then a larger page — already
+  existed and already refused any candidate that had not fixed the problem. It ran on
+  two symptoms: a turned part missing an axial station, and an **authored** intent that
+  failed to place. A required *envelope* dimension that found no room was neither, so
+  these drawings never entered the ladder at all — `scale_decision["attempts"]` came
+  back empty and an `overall_dim_withheld` error was reported on the first sheet tried.
+
+  `overall_dim_withheld` is deliberately not a scale blocker, and that is right:
+  reporting it through `placement_unsatisfiable` once made `build_drawing(part,
+  scale=...)` raise on parts that had always built. But that argument is about refusing
+  the scale a **caller asked for**. Choosing one automatically is a different question,
+  and one predicate was answering both.
+
+  What it changes, measured:
+
+  | part | before | after |
+  |---|---|---|
+  | stepped `Box(80, 60, 30)` | A4 1:1, overall depth withheld (error) | A4 1:1, **clean**, pictorial dropped |
+  | 80 × 60 plate, two close bores | A4 1:1, `incomplete`: unlocated hole, dropped location, withheld depth | A2 2:1, **clean**, every dimension placed |
+
+  **ADR 2 invariant 13 is amended** to match. It read "the optional isometric yields only
+  to close missing axial coverage", which had been one trigger behind the code since
+  2026-08-23 — the record was consolidated twelve days later and transcribed the narrower
+  claim. It now states the principle the triggers share: the isometric yields only to
+  preserve manufacturing completeness, never for space, crowding or preference. The list
+  itself is `builder._ISO_YIELD_TRIGGERS`, and a guard fails if it changes, so the next
+  widening is caught by a test rather than by a reader.
+
+  Dropping the optional pictorial is the first lever, which is the order #443/#1299
+  already argue for — a pictorial cannot outrank a dimension needed to manufacture the
+  part. Pin `page=` or `scale=` to opt out; a pinned sheet still reports honestly rather
+  than replanning.
+
+- **A generated `--script` reproduces a re-planned layout.** A declared build does what
+  it is told and never enters the recovery ladder, so `generate_sheet_script` pins the
+  settled page, scale and view set into the script. It used to do that only for the two
+  families whose replan could be predicted from the model; nothing predicts the new
+  trigger, because whether a mark fits is a fact about a measured sheet. The reference
+  build is now unconditional — one extra `build_drawing` per generated script, measured
+  at 0.19 s for a plain box, 0.34 s for a pocket and 1.17 s for a part that actually
+  replans — and an **undrawable** source can never be the reason a script is not
+  written, the standard the inspection sidecar beside it already meets. A deliberate
+  refusal (an unknown page size, `ScaleIncompatibilityError`, `ViewPlanIncomplete`)
+  still reaches the caller.
+
+  A script that keeps `auto_dimensions()` is not handed a pinned view set: `Sheet`
+  refuses `authored_views()` beside it, so the two together wrote a script that raised
+  when run. The settled set is reported in a comment instead.
+
+- **A bolt circle is no longer stated unless something other than the fit supports it
+  (#1596).** `EQ SP ON ø… BC` tells a machinist to work from a centre; #1595 found a real
+  part where that centre was in mid-air. Six ⌀2.4 holes in a 2×3 rectangular grid came out
+  as `4× ⌀2.4 THRU EQ SP ON ø34.4 BC` plus an unrelated `2× ⌀2.4 THRU`, the circle centred
+  on (0, −20.11) — nothing is concentric with it.
+
+  The fit was not a near miss. **Any three non-collinear points are concyclic, and so are the
+  four corners of any rectangle** — for those arrangements the residual is zero by
+  construction, so no tolerance on it can help. Nor does equal spacing rescue it: that
+  rectangle is near-square (24.30 × 24.38), so its corners sit 90.19° / 89.81° apart and
+  would pass an `EQ SP` check too.
+
+  What gives it away is the leftover callout beside it. Corroboration must now come from
+  somewhere the fit cannot reach — any one of:
+
+  1. **five or more members** — a circle through five holes could have failed to fit;
+  2. **a concentric physical feature** — a boss, spigot, central bore, or a round body whose
+     own axis is the circle's, something a machinist can indicate off;
+  3. **it accounts for every hole of its kind** — no identical hole left off the circle.
+
+  (3) is the discriminator here: six identical holes and the circle claimed four. A drawing
+  does not put two of six bolts on a different plan. It is also a question about the part
+  rather than about its angle to the axes, unlike the row/column test it replaced.
+
+  Refused patterns keep every hole: the members fall through to ordinary grouping, so the
+  #1595 part now reads `6× ⌀2.4 THRU` — which also repairs the fragmentation into 4 + 2.
+
+  **What refusing costs, measured.** `hole_requirement_outcomes` builds its pattern-member
+  set from the recogniser's patterns rather than the ones the IR adopted, so after a refusal
+  it keeps those holes out of the loose groups, finds no `PatternFeature` to join, and warns
+  `hole_requirement_unverifiable` about holes the sheet states correctly — two on this part,
+  26 issues against 23. The pre-existing oblique-pattern refusal (#971) has the same hole and
+  was simply never reachable; #1596 makes it so. Tracked as **#1607** with three attempted
+  fixes recorded, and guarded by an `xfail(strict)` here.
+
+  Refused at the recognition→IR adapter, not in the recogniser: a circle through those holes
+  is genuinely findable, and ADR 3 leaves the recogniser to report the geometry it finds.
+  Whether it may be *stated as a drafting datum* is drafting policy, which is draftwright's.
+
+### Changed
+
+- **The title block carries every ISO 7200:2004 mandatory data field.** Three had
+  no representation at all before — segment/sheet number (5.1.6), approval person
+  (5.3.4) and document type (5.3.6) — and are now `sheet=`, `approved_by=` and
+  `document_type=` on `build_drawing`, `make_drawing`, `Sheet`, the CLI and the
+  `--script` emitter. Cell widths come from the standard's own recommended
+  character counts (via `build123d-drafting-helpers>=0.15.4`), so they follow ISO
+  7200 and the font rather than a chosen proportion.
+
+  Eleven fields need four rows, so **the block is 32 mm tall rather than 16**, and
+  the band reserved for it goes 35 → 40 mm. `material` and `general_tolerance`
+  stay although neither is an ISO 7200 title-block field; they are on every real
+  drawing.
+
+- **The scale is stated outside the title block**, beside the projection symbol,
+  per ISO 7200 §4 ("presented outside the title block only when used, e.g. scale,
+  projection symbol"). It is drawn unconditionally, and the new **`scale_not_stated`
+  error** checks the rendered sheet rather than trusting the code that draws it.
+  The `SCALE` cell is gone from the block.
+
+  **Two consequences worth knowing before you upgrade:**
+
+  - A dense sheet may escalate a page size. In the golden corpus one fixture of
+    fourteen (`flange_dense`) moves A4 → A3; the other thirteen keep their page
+    and scale, and a sweep of ten ordinary shapes showed no change at all.
+  - **Dimension placement now avoids the title block (#1593)**, which it never
+    did: the block is drawn after placement, so only GD&T frames ever avoided it
+    (#481). A two-row block kept that out of reach; a four-row one does not.
+    `build_drawing(Box(40, 30, 12))` used to run its `30` dimension 2.1 mm into
+    the block with `lint()` reporting a clean sheet; a stepped
+    `Box(80, 60, 30)` put the whole `60` label inside it, reported only as an
+    `annotation_overlap` between the labels `60` and `DRAWING`.
+
+    Most displaced dimensions move to the view's opposite strip and stay on the
+    sheet. Where both strips are full the dimension is **withheld as an error
+    naming the block**, rather than drawn through it: in the golden corpus
+    `pocketed` and `prismatic_ladder` lose their overall depth at the automatic
+    A4/1:1, and place it on A3 or at 1:2. Likewise a **pinned** A4 for a
+    five-step turned shaft no longer reaches 5:1, because at that scale the front
+    view runs down past the block and the overall height cannot be placed; the
+    2:1 sheet is returned with `axial_length_missing` stated. Leaving the page
+    automatic still carries every station, on A3 at 5:1.
+
+    Choosing a page and scale without knowing which dimensions will still need
+    room is #1590.
+
+### Changed
+
+- **Exports are reproducible by default.** `reproducible` now defaults to `True`
+  on `build_drawing`, `make_drawing`, `Drawing` and `Drawing.export`, so two
+  exports of one drawing are byte-identical without asking. Pass
+  `reproducible=False` to opt out.
+
+  It was opt-in on a cost argument, and the cost turns out to be small at the
+  scale most drawings are made: on the NIST CTC-01 AP242 fixture, +0.27 s on a
+  13.4 s job (+3.2% of export, +2.0% overall). The +32% of export measured when
+  the flag was introduced was a 358-part assembly; the ordering work is one
+  bounding box and one edge walk per part, so it scales with part count and a
+  part-heavy sheet is still the case to opt out of.
+
+  Two things pushed the default over. A file that changes between runs cannot be
+  diffed, checksummed or cached — the reason to want it is not speed-shaped. And
+  neither the CLI nor `Sheet` exposes the flag at all, so every drawing made
+  through those surfaces was non-reproducible with no way to change it.
+
+  One behaviour follows from the default rather than from this change: a DXF
+  export that cannot reach the ezdxf document to pin its metadata now raises
+  rather than writing live data, because every export now claims
+  reproducibility. Opting out restores the old fallback.
+### Fixed
+
+- **A supplied title-block `date` is stated on the sheet** (#1585). `TitleBlock`
+  could show only one of `revision`/`date` in its top-right cell and revision
+  won, while draftwright defaults `revision` to `"A"` — so every drawing that
+  set a date lost it, silently, in every version that has had the parameter.
+  Adopting `build123d-drafting-helpers>=0.15.4` gives the date a cell of its
+  own whenever a revision is also set, and `_make_title_block` now names that
+  cell, so the date reaches the PDF text layer and the cell-overflow lint
+  rather than only the rendered geometry.
+
+  `date`, `revision` and `company` are stripped before use, matching what the
+  block itself does: a whitespace-only `company` previously raised `KeyError`
+  from `cell_bbox`, and padded values were measured at a width the sheet never
+  drew. (The DRAWN BY narrowing this entry originally described, and the shared
+  top-right cell it named, are both gone with the ISO 7200 layout above — the
+  date has its own cell in its own row, so it costs DRAWN BY nothing.)
+
+
 ### Added
 
 - Report schema v1 now projects one recognition-owned physical requirement ledger shared with
@@ -154,6 +407,21 @@
 
 ### Changed
 
+- Updated the exact Quiddity runtime and declared-inspection contract to **0.2.8**, which stops
+  a modelled thread being read as a stack of turned shoulders and coalesces contiguous
+  equal-diameter bands into the one span they physically are. Measured on the repo corpus,
+  NIST CTC-05 is again the only fixture that moves: its ⌀304.8 region was three adjacent
+  rungs and is now one, in both STEP encodings (#1135, quiddity#587).
+
+- Updated the exact Quiddity runtime and declared-inspection contract to **0.2.7**. Unlike the
+  0.2.5 adoption below, this one changes drawings: the provider now requires circumferential
+  cylinder support to establish a turned-profile axis line, so rounded plate corners and parallel
+  offset cylinders no longer form a false stepped shaft. Measured on the repo corpus, NIST CTC-05
+  is the only fixture whose recognised turned profile moves — its two STEP encodings previously
+  disagreed (3 steps and 9 steps for the same solid) and now agree on 5 — and a plate part that
+  previously raised out of `generate_sheet_script` now emits and builds. No family, record schema
+  or adapter changed (#1555, quiddity#586).
+
 - Updated the exact Quiddity runtime and declared-inspection contract to 0.2.5 for the
   provider's recognition performance improvements.
 

{draftwright-0.4.27 → draftwright-0.4.29}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: draftwright
-Version: 0.4.27
+Version: 0.4.29
 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: build123d-drafting-helpers>=0.15.1
+Requires-Dist: build123d-drafting-helpers>=0.15.4
 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'
 Requires-Dist: cadquery-ocp-novtk!=7.9.3.1.1; python_full_version >= '3.13'
@@ -692,7 +692,7 @@ Requires-Dist: numpy>=1.24
 Requires-Dist: ocp-gordon<0.3
 Requires-Dist: pillow>=10
 Requires-Dist: pypdfium2>=4
-Requires-Dist: quiddity==0.2.6
+Requires-Dist: quiddity==0.2.9
 Requires-Dist: reportlab>=4.0
 Requires-Dist: rich>=10.11
 Requires-Dist: steputils==0.1
@@ -758,8 +758,18 @@ 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
+draftwright /parts/my_part.step --out-dir .   # explicitly write into the working directory
+draftwright my_part.step --out-dir drawings   # create/use a separate output directory
 ```
 
+CLI outputs default to the directory containing the STEP input, including an absolute or
+nested input path. `--script` writes its Python file and inspection sidecar there too; the
+script retains that output destination when run from another directory. Use `--out-dir DIR`
+to choose a directory, or `--out PREFIX` to choose a basename as well (these options are
+mutually exclusive). Relative overrides use the current working directory. Live object
+scripts (`module:part`) default to `drawing.py` in the working directory.
+
+
 `draftwright --help` lists every flag; `--version` prints the version.
 
 ### Python

{draftwright-0.4.27 → draftwright-0.4.29}/README.md

@@ -55,8 +55,18 @@ 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
+draftwright /parts/my_part.step --out-dir .   # explicitly write into the working directory
+draftwright my_part.step --out-dir drawings   # create/use a separate output directory
 ```
 
+CLI outputs default to the directory containing the STEP input, including an absolute or
+nested input path. `--script` writes its Python file and inspection sidecar there too; the
+script retains that output destination when run from another directory. Use `--out-dir DIR`
+to choose a directory, or `--out PREFIX` to choose a basename as well (these options are
+mutually exclusive). Relative overrides use the current working directory. Live object
+scripts (`module:part`) default to `drawing.py` in the working directory.
+
+
 `draftwright --help` lists every flag; `--version` prints the version.
 
 ### Python

{draftwright-0.4.27 → draftwright-0.4.29}/docs/reference/recogniser-capabilities.md

@@ -70,11 +70,34 @@ issue. The contract test derives package record outputs, converter registries, l
 and emitter branches independently, so copying a new name into the declaration alone cannot make CI
 pass.
 
-### Quiddity 0.2.6 runtime contract
+### Quiddity 0.2.9 runtime contract
 
-The runtime pins the published `quiddity==0.2.6` package, following the 0.2.4 adoption in
+The runtime pins the published `quiddity==0.2.9` package, following the 0.2.4 adoption in
 [#1486](https://github.com/pzfreo/draftwright/pull/1486) and the migration in
-[#1471](https://github.com/pzfreo/draftwright/issues/1471). Its public `quiddity`, `quiddity.evidence` and
+[#1471](https://github.com/pzfreo/draftwright/issues/1471). 0.2.7 is a patch adoption: it requires
+circumferential cylinder support to establish a turned-profile axis line, so rounded plate corners
+and parallel offset cylinders no longer form a false stepped shaft
+([quiddity#586](https://github.com/pzfreo/quiddity/issues/586), [#1555](https://github.com/pzfreo/draftwright/issues/1555)).
+No family, record schema or adapter changed.
+
+Quiddity 0.2.9 exposes native cylindrical seats as open-arc `SectionRecess` channels and
+keeps coaxial bores separated across exterior air gaps. The pinned frame now retains six
+Ø1.1 × 3.0 mm hinge bores and three Ø14.3 seats. The seat grammar remains explicitly
+unsupported in Draftwright pending [#1613](https://github.com/pzfreo/draftwright/issues/1613).
+Adapter refusals of derived hole patterns are recorded with same-run identity so their
+physical members remain in the ordinary-hole requirement ledger (#1607/#1619). These changes
+use the existing public schemas. Identical patterns on different opening planes retain the
+provider's documented grouping limitation ([quiddity#595](https://github.com/pzfreo/quiddity/issues/595));
+separate physical bores must not be merged to recover a pattern.
+
+0.2.7 also partitions bands by physical axis line rather than by axis direction, requiring two
+distinct diameters *per line*. One body can therefore publish **several** coaxial
+`TurnedProfile`s where 0.2.6 published at most one. `Analysis.profiles` already holds every
+body-local profile and is unaffected; `Analysis.prof` remains the compatible zero/one view and
+is `None` for a plural result, which sends `holes.py`'s `a.prof is not None` branches down their
+non-rotational path. **No fixture in this repository reaches plural cardinality on either
+version**, so the green tiers are not evidence that path behaves well — it is newly reachable
+and untested. Its public `quiddity`, `quiddity.evidence` and
 `quiddity.inspection` surfaces supply recognition, exact occurrence evidence and declared geometry
 reads. There is one recognition aggregate per build; consumers project that result rather than
 calling the document builder to obtain another run.

{draftwright-0.4.27 → draftwright-0.4.29}/docs/reference/sheet.md

@@ -28,6 +28,47 @@ retain the requested convention. Annotation footprints, scale/page selection and
 repacking use the same convention. An authored relation contradicting the principal layout
 fails before projection; a whole-view pin must preserve the convention's relationships.
 
+## Sheet margins and title-block width
+
+Set each physical sheet edge independently with `margin_left`, `margin_right`,
+`margin_top`, and `margin_bottom` (millimetres). Omitted edges remain 10 mm. `frame=True`
+draws the border at those margins and reserves a further 6 mm inside it for content.
+`zones=True` adds the zone ruler and implies a frame; each margin must fit its labels.
+
+Sergio's folded-sheet arrangement is an explicit option on A4, A3, or A2:
+
+```python
+from draftwright import build_drawing
+
+drawing = build_drawing(
+    "part.step", page="A3", frame=True,
+    margin_left=25, margin_right=10, margin_top=10, margin_bottom=10,
+    title_block_width=175,
+)
+drawing.export("sergio", formats=("pdf", "svg"))
+```
+
+The same keywords work on `Sheet(...)` and `make_drawing(...)`. An explicit
+`title_block_width` measures between border centrelines and aligns the block to the right
+and bottom sheet margins. The 175 mm block therefore fits the 210 mm folded face with
+25 mm left and 10 mm right clearance. Its visible border extends half its 0.15 mm stroke
+beyond each centreline. Without an explicit width, the existing widths remain 120 mm on
+A4 and 150 mm on larger ISO sheets, with the existing extra 1 mm right/bottom clearance.
+Revision defaults also remain unchanged.
+
+CLI parameters carry the same options into rendered drawings and generated scripts:
+
+```sh
+draftwright part.step --page A3 --frame \
+  --margin-left 25 --margin-right 10 --margin-top 10 --margin-bottom 10 \
+  --title-block-width 175 --out sergio --format pdf,svg
+```
+
+Add `--script` to write a replayable `Sheet` script. Margins must be finite and nonnegative;
+the width must be finite and positive. An explicit sheet too small for its margins and
+block is refused. `drawing.drawable_bounds` exposes the physical drawable rectangle as
+`(left, bottom, right, top)` page coordinates in millimetres.
+
 ## Dimension text style
 
 ### Title fields and notes at paper size
@@ -278,7 +319,7 @@ produces `angular_support_unverifiable`; contradictory supports produce
 `angular_support_mismatch`.
 
 Automatic drawings derive outside-profile angular requirements from ordered face
-supports in Quiddity 0.2.6. Angles already defined by a recognised chamfer or
+supports, available since Quiddity 0.2.6. Angles already defined by a recognised chamfer or
 regular-polygon callout do not add duplicate automatic requirements. This requires
 the same run's exact face or shared-edge evidence, not matching numerical angles;
 explicit angle declarations remain available on those corners.
@@ -642,8 +683,15 @@ and reports capacity/crossing failures. Trailing zeroes are intentional manufact
 sheet.authored_dimensions()
 sheet.dimension(envelope, "width.length").format(decimals=2)  # 13.5 prints as 13.50
 sheet.dimension(tapped_hole, "bore.diameter", view="plan", side="left")
+sheet.dimension(tapped_hole, "location").format(decimals=2)  # all selected directions
 ```
 
+A location intent's precision applies to all its selected directional values, including
+side-drilled holes and slots. The policy survives generated-script replay. When several
+location intents share one coincident mark, give them matching display precision; incompatible
+printed values raise an actionable error instead of silently choosing one intent's label.
+
+
 ::: draftwright.sheet.DimensionIntent
     options:
       filters: public
@@ -665,6 +713,16 @@ sheet.dimension(tapped_hole, "bore.diameter", view="plan", side="left")
 Circular-blind, paired-ramp and through steps use a multi-parameter handle because their
 requirements remain separately addressable:
 
+Circular seats open at both run ends use `sheet.circular_channel(...)`. The
+`centreline` gives ordered endpoints on the cylinder axis; `section` gives three
+physical arc points (start, angular midpoint, end) in ascending transverse world axes.
+The geometry supports minor, semicircular, and major arcs. The engine anchors its
+leader on the curved wall and locates the axis from the stock bounding-box minimum.
+In an authored set, request `seat_diameter.diameter`, `seat_run.length`,
+`seat_sweep.angle`, and `location` independently. A location intent covers X, Y, and Z;
+its `format(decimals=...)` applies to all three. Generated scripts retain the full
+geometry and these intents.
+
 ```python
 ramp = sheet.paired_ramp_step(
     axis="y",

{draftwright-0.4.27 → draftwright-0.4.29}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "draftwright"
-version = "0.4.27"
+version = "0.4.29"
 description = "Automated technical-drawing generation for build123d"
 readme = "README.md"
 license = { file = "LICENSE" }
@@ -22,8 +22,8 @@ dependencies = [
     # This novtk release lacks OCP.GccEnt on macOS. The exclusion must reach
     # published wheels, not only uv's development resolver constraints.
     "cadquery-ocp-novtk!=7.9.3.1.1 ; python_full_version >= '3.13'",
-    "build123d-drafting-helpers>=0.15.1",
-    "quiddity==0.2.6",
+    "build123d-drafting-helpers>=0.15.4",
+    "quiddity==0.2.9",
     "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