[Jepson2k]

Two changes that together let a downstream host move its waldoctl pin to
0.13.

ShapeWire was one field short. waldoctl 0.13 adds physics to
shapes, so ShapeBase.to_wire() returns seven elements. ShapeWire
declared six, and set_shapes builds one by positional unpack —
ShapeWire(*s.to_wire()) — so every call raised TypeError: Extra positional arguments provided the moment the pin moved. The two read
paths dropped physics for the same reason.

physics is now a seventh field with a default, which keeps the struct
readable both ways across that boundary: a peer still sending six
decodes with physics absent, one sending seven decodes fully.

Verified against waldoctl 0.13.1: 7-element encode, physics round-trips
intact through shape_from_wire, a six-element peer still decodes, and
msgspec agrees in both directions.

tcp_offset() faked an answer. It returned [0, 0, 0] when nothing
replied. Zero is a legitimate offset — a tool deliberately cleared — so a
caller handed it as a not-answered sentinel cannot tell "the offset is
zero" from "there is no controller", and a host that adopts the readback
quietly erases the offset the user just set. waldoctl's RobotClient
docstring forbids exactly this. It raises ConnectionError now;
_request already retries, so reaching that line means unreachable
rather than one lost datagram.

The waldoctl pin moves to v0.13.1 with them, since the shape fix is only
meaningful there.

Tests: 280 passed, 8 skipped. The one failure,
test_msg_mounts_on_the_flange_not_inside_the_wrist, fails identically
with these changes stashed — pre-existing, not from this.

🤖 Generated with Claude Code

https://claude.ai/code/session_012tyAH93sLjc6W8Kp9ygFS8

[Jepson2k]

Closing as superseded. Both halves of this change are already on main:

• Seventh shape wire element: c25b06a ("Carry the seventh shape wire element and refuse physics explicitly") adds physics to ShapeWire with a default, and passes it through both read paths.
• No faked TCP offset: 460e4c7 ("Apply full TCP transforms through planning and readback") made tcp_offset() raise instead of returning [0, 0, 0] when the controller doesn't answer. It raises TimeoutError rather than the ConnectionError proposed here, and Guard the TCP readback contract with a test #50 adds a test for that.

The branch is also 130 commits behind main and conflicts with it.

Written by Claude (Claude Code) on behalf of @Jepson2k.