# Part Fit

> Check a custom lab part against the labware it mates with - plate carriers and adapters, cuvette holders, tube
> racks, breadboard and cage mounts, microfluidic chips, behavior-rig parts - then write the design brief and the
> parametric build123d model. Every interface dimension comes from a bundled standard or is named as unchecked; the
> browser evaluates it at maximum material condition, budgets the process undersize, and holds walls, threads,
> corners, the table grid, the kerf and the material to the process and the service. Built on the SkillSafe skill
> @k-dense-ai/lab-hardware-cad, from k-dense-ai/scientific-agent-skills (MIT), by K-Dense Inc.
> A planning aid, not engineering sign-off: the page does not build the model, so check every dimension against
> the catalogue drawing and a test coupon before fabrication.

https://part-fit.skillsafe.ai/

## What is free
- Bundled standards: ANSI/SLAS 1-4 microplate footprint (127.76 x 85.48 mm +/-0.50), height, flange and well positions (96, 384, 1536); a 10 mm cuvette format (marked unverified); metric 25 mm and imperial 25.4 mm breadboard grids; the 30 mm cage system; SM1 threads.
- The skill's two intents: envelope (a pocket, bore or slot must be at least maximum material condition plus the clearance) and match (inside the nominal band, widened by an allowance - never shifted); a negative clearance is refused. The receiving feature's process undersize is budgeted against the mating part's maximum size.
- First-article clearances per side by fit and process (free-sliding 0.40 mm FDM, 0.20 SLA, 0.10 CNC); minimum walls and features per process; printed holes undersize; printed fine threads refused; pocket fillets held under the plate's 1.58 mm minimum corner radius; metric and imperial grids never mixed; a laser kerf convention required.
- Materials held to service: autoclave (printed PLA, PETG, ABS, resin fail), 70% ethanol on acrylic, solvents on polycarbonate, uncured resin and cells, autofluorescence in an optical path. Every stated requirement becomes a clear, material or bounding-box gauge. Plus a to-scale top view, a parameter block, the fit sheet (.csv), the checks (.csv) and everything (.json).

## The metered lanes
Two tasks on gpt-terra, metered per run, routed by `task`. Input string fields: `task` (brief or model), `part`, `interfaces`, `carry` (model only: the brief decisions) and `facts`.
- `brief` replies with `summary`, `family`, `process` (process, material, why), `interfaces` (feature, source, intent, value_mm, why), `unchecked`, `gauges` (requirement, gauge, how), `coupon` and `caveats`.
- `model` replies with `summary`, `filename`, `code` (the build123d source), `parameters` (name, value_mm, why) and `caveats`.
The page holds every reply to its own facts: every brief value re-evaluated against its standard at maximum material condition and through the process undersize, a failed material replaced, unchecked interfaces named, one gauge per requirement sized from its own number; the model checked for its docstring, INTERFACE and DESIGN blocks, every standard id and nominal, interfaces(), checks() and build(), no bare numbers in build(), and no file, process or network calls.

## Links
- API tutorial: https://part-fit.skillsafe.ai/api.html
- Source skill: https://skillsafe.ai/skill/@k-dense-ai/lab-hardware-cad