AnasSarkiz/usb-c-pd-brushed-motor

docs/TOOLING-ISSUES.md

# A9 coordinate serialization investigation

## Official packages

The board keeps official tscircuit 0.0.2742, core 0.0.2056, props 0.0.677 and circuit-json 0.0.510. CLI is updated to 0.1.2237. No dependency override or supplier-model patch is installed.

An independent project also tests official core 0.0.2069 with official circuit-json 0.0.512 and React 19.2.0. The unchanged C23162 supplier probe specifies pcbX="2mm", pcbY="-2mm". Both official combinations still emit numeric display_offset_x/y, failing the unmodified schema. The latest official package was restored and the failure reproduced again after the experimental test.

Evidence: coordinate-serialization-A9.json; official-core-coordinate-A9.json; official-core-restored-coordinate-A9.log. No coercion, schema downgrade or warning suppression is used.

## Isolated source fix

Source: tscircuit/core at 24d72602641a1bccb3516d6ab9fad7b95ec523bd, version 0.0.2069. The patch is evidence/core-coordinate-fix-A9.patch. Numeric resolved PCB offsets are serialized as millimeter strings at the database boundary. Existing group expressions remain text. An absent group anchor is omitted rather than serialized as invalid null. No coordinates, transforms, geometry or imported models are rewritten.

A regression fixture validates the entire emitted document and checks numeric/unit/one-axis/calc/edge/nested-group placements against actual geometry. It fails on unmodified source and passes with the patch. The related positioning suite passes: 26 tests across 22 files, five existing snapshots, zero failures. Both ESM and declaration builds pass.

A task-local yalc store was used to test the built package against the actual unchanged supplier probe: strict full-document schema validation passes. The local package signature was 0.0.2069+b45d5d11. The link was removed and the official package reinstalled. Main board dependencies were never linked to this package. This is an experimental fix, not an upstream publication or released toolchain.

## Outstanding source-tooling checks

The upstream canonical dependency installation fails resolving its pinned pcb-trace-linter GitHub dependency. A separate diagnostic environment supplies official runtime and declared test dependencies; it is recorded in tooling/core-recheck-A9/package.json and bun.lock. It is not described as a successful canonical installation.

Whole-source TypeScript checking remains failed with eight pre-existing diagnostics: schematic-match-adapt return-type mismatch and six TI fixture symbol-type incompatibilities in this nested environment. The unchanged source and patched source produce exactly the same eight diagnostics; see core-typecheck-comparison-A9.json and both complete logs. None are suppressed or excluded from the source-project check. Board formatting, TypeScript and board tests are checked separately. No all-core-tests pass is claimed.

Product placement remains blocked until a validated board toolchain and the remaining schematic/power gates pass. A local regression result does not authorize routing. No core PR or upstream publication was created.


## A10 CLI cache mode issue

CLI 0.1.2237's getOrGenerateCircuitJson cache is keyed by source file hashes, without the requested build mode. A placement check run after a schematic-only build therefore reused an artifact with zero PCB components and falsely returned success. That result is withdrawn in main-native-checks-schematic-cache-A10.json and main-placement-schematic-cache-A10.log. It is not accepted as placement evidence.

A fresh full unrouted build was generated and archived as evidence/full-build-output-A10.circuit.json, with 128 purchased source components, 140 generated PCB-component records and zero PCB traces. Checks were then run sequentially against that full build. Netlist/pin_specification/source/schematic-placement pass; placement fails on the default overlaps. An additional official placement check directly reads the archived circuit.json-compatible filename and also fails. The unsupported arbitrary .json filename attempt is retained separately. Neither CLI source nor validation thresholds were modified.

When reviewing schematic-only output, never accept subsequent cached placement results. Check that the artifact contains all expected PCB components, and directly check the archived full circuit.json. Coordinate-schema and power/firmware prerequisites remain unresolved; product placement remains unstarted.

## A12 supported manual-placement API investigation

Official props types accept manualEdits.pcb_placements entries with selectors and numeric or unit-string centers. A12 independently exercises this public API with two unchanged C23162 supplier imports on a native A4 sheet, one rotated 90 degrees. Main dependencies remain official and unchanged. No supplier model, emitted JSON, schema, check or diagnostic is modified.

Numeric millimeter centers correctly place R1 at(-4,3) and R2 at(4,-3). A warning-free network-enabled build passes, strict full-document schema passes, both electrical pads/ports remain coincident and rotate correctly, and netlist/pin_specification/source/schematic-placement/placement pass with zero errors/warnings. The PCB rendering was inspected. Placement directly checks the archived full evidence/native-numeric-placement-probe-A12.circuit.json, containing both PCB components and no PCB traces. This establishes a supported API option for these tested inputs, not all board/group configurations or permission to begin product placement.

Unit-string centers remain broken: the core manual transform reads raw this.props.manualEdits instead of parsed numeric centers, producing NaN geometry that JSON serializes as null. The renderer and strict schema fail; 24 top-level schema issues are retained. The original pcbX/pcbY probe also still fails strict schema on numeric display offsets (one top-level issue), even though its native build exits zero. The combined audit exits one because it retains both failures. These regressions are not patched, ignored or represented as fixed.

Initial sandboxed builds emitted supplier-network warnings; successful numeric and failed unit-string/explicit probes were rerun with network access. The original logs remain retained as superseded attempts. An attempted netlist check of .circuit.json fails because that subcommand expects circuit source; the corrected source check passes. The other four checks consume the full archived artifact. All attempted commands/results are recorded; a command-input error is not accepted as a completed check.

Evidence: native-placement-audit-A12.json/log; three archived probe .circuit.json files; native-*-build-network-A12.log; native-numeric-*-A12.log and native-numeric-netlist-source-A12.log; native-numeric-placement-A12.png. Product placement remains blocked by unfinished power, assembly and firmware qualification. Main product coordinates, mounting holes and routing are unchanged/unstarted.

Complete unfiltered union-schema diagnostics are losslessly archived in native-placement-audit-A12-full.json.gz. The review JSON retains every probe's status/count, geometry and source/output hashes and the hash of the uncompressed complete report. Compression changes evidence storage only; all failures remain visible and the combined audit still exits one.