muse/book-reading-clip-lamp

cloud-build-investigation.txt

Schematic

RESOLVED IN0.2.5: LED7=C965805 and LED8=C965800 now have explicit imported
footprints and correct pin1=cathode / pin2=anode mappings. The supplier-enabled
CI build took32.73 seconds with0 error records; all-layer shorts pass.
The failure findings below are historical for0.2.3/0.2.4.

CLOUD BUILD INVESTIGATION - 0.2.4

Failed release: https://tscircuit.com/muse/book-reading-clip-lamp/releases/ae67c03d-15a3-4a45-a0c9-b76c1b46cc87
Cloud command: bunx tscircuit build --ci --concurrency 4
The cloud log confirms installation of tscircuit0.0.2646 and capacity-autorouter0.0.919.
It discovered TWO boards: index.circuit.tsx and reroute.circuit.tsx.
index.circuit.tsx completed successfully at21:27:37 after starting at21:26:05
(about92 seconds, including previews, with another routing worker competing).
The fresh reroute job timed out at21:46:04 after1200000 ms. Total job1267 s.
Increasing the timeout was not the appropriate fix for this discovery mistake.

Fix: includeBoardFiles=["index.circuit.tsx"] in tscircuit.config.json.
mainEntrypoint alone does not limit normal automatic *.circuit.tsx discovery.
The20-minute worker timeout is preserved. bun run reroute remains opt-in.

Measured local builds on this Mac (single runs, not timing guarantees)
1. Explicit index build with parts engine disabled, PCB/schematic PNGs,
   and180-second routing-phase limit: 22.91 seconds, exit0.
2. Fixed CI build with fresh dependency install, supplier checks, concurrency4,
   site, preview images and transpilation: 66.09 seconds, exit0.
   It printed "Building 1 file(s)" and "Circuits 1 passed".
Versions: tscircuit0.0.2646; CLI0.1.2237; core0.0.1971;
capacity-autorouter0.0.919; local Bun1.3.14 (cloud Bun1.3.12).

Routing distinction
The normal board validates and replays the reviewed saved copper. It has291
traces and366 vias,0.30 mm hole/0.45 mm pad. The full CI build's copper is
identical to the copper already verified by Gerber shorts on all four layers.
This is not evidence that fresh automatic routing converges. Fresh pipeline9
was active in the cloud, reaching portPointPathingSolver, but did not finish
within20 minutes. Its outputs must not replace reviewed copper without checks.

Separate supplier-check finding
The full parts-engine run produced TWO source_component_misconfigured_error
records even though this CLI returned exit0: LED7 and LED8 were automatically
assigned C965799, whose pin1 is cathode, while the generic0603 symbols use
pin1 as anode. Both LEDs remain explicitly PART SELECTION PENDING in the BOM;
no fixed C965799 is present in the hardware source. This automatic assignment
must not be treated as an approved procurement/assembly choice. Select and
verify the actual green/orange LED part and orientation before manufacturing.
Other supplier footprint-match warnings also remain for review. The full CI
run must NOT be described as having zero error records. The earlier zero-error
build explicitly disabled the parts engine; core copper checks were enabled.

This revision fixes cloud board discovery only. It does not change copper,
component placement, net connectivity, or approve unselected supplier parts.