Replicating a physical part

You have the part in your hand: a broken clip, a wall plate, a knob, a bracket from an appliance. Usually it was injection molded. You want a printed one that does the same job. That is not the same as copying its shape: a molded part is designed around the mold (draft, thin even walls, coring, ribs, a living hinge in polypropylene), and a print is better without most of that. What must stay the same is what it fits: the holes, their spacing, the faces that touch something else.

So replicating is three jobs:

1. Measure the right things, well: every size that fits something, from one datum, three times; the rest only as well as the look needs.
2. Decide, for each dimension, whether it must match (critical) or may change (free), and what to do with each molded feature.
3. Keep the record (the replication spec) and check the model against it.

The MCP prompt replicate_part walks an assistant through all of it, with two stops for your confirmation: after the measurement plan, and after the list of changes. The record is the part's MEASURE.md (the measurement sheet with four more columns), and the replication_spec tool checks the script's Dims against it.

If you also have a 3D scan of the part, it helps with the shape; it is used the way a remix uses a downloaded mesh, but calipers win for every size that fits something.

How a session goes

1. Identify. You describe the part, what it mates with and how (screws, a snap, a press fit, a seal), and attach photos or a drawing. The assistant names how it was made (molded parts show draft, a parting line, round ejector-pin marks, a gate) and asks what it cannot see: the mating parts, the material (look for a recycling mark such as >PP< or >ABS< molded inside), heat, sunlight, how often it comes apart.
2. Measurement plan (stop). The assistant writes MEASURE.md with one row per dimension, the values empty: the feature, its class, the tool and method, the accuracy you need, the Dims field it will become, and the photos to take. You measure and fill it in.
3. Classify and decide (stop). The assistant reads your readings back, points out readings that disagree (a tapered wall, an oval hole), and proposes what to do with each molded feature, each with its reason in the Why changed column; plus the material, the print orientation and the fits. You confirm.
4. Model. Every dimension becomes a named Dims field; critical ones as measured, or with the reason they differ (a clearance, a standard size).
5. Verify. The usual checks, then replication_spec until it has no warnings, measure on the critical features of the built part, and a fit-test template of each critical interface before the full print.

The measurement plan

The measuring guide has the methods: calipers (outside, inside, depth, step), gauges, tracing, offsets, photos with a scale. For a part you are copying, the plan adds:

• One datum for everything (the centre of the main hole pattern, a flat back, a mating face), and hole spacings measured edge to edge (outside of both and inside of both, averaged) rather than chained.
• Three readings of every critical size, at slightly different places, all three in the Value cell: 4.21, 4.19, 4.22. The spec averages them and warns when they spread wider than the accuracy you wrote.
• Where on a molded wall you measured. Molded walls taper (draft), so a wall or a boss reads differently at its root and at its tip. Measure where it mates, or at mid-height, and say so.
• Threads with a thread pitch gauge and calipers on the outside diameter; metric and inch sizes look alike (M3 and 4-40, M6 and 1/4-20).
• Standard sizes (a screw clearance hole, NEMA box screw spacing, a bearing seat) get the source standard with the standard's name, and are still checked against the part.
• Photos: an overview, each critical feature square on with a scale in its plane, the back (coring, ribs, bosses) and any marking.

Critical or free

Each row of the spec has a Class:

Critical: must match	Free: may change to print better
mating interfaces: faces that sit against something	outer styling, the look of the outside
hole sizes and hole spacings; slots that take a screw	ribs, gussets, coring
fits: press, sliding, free (with the clearance from your settings)	wall thickness away from the interfaces
sealing and bearing faces	fillets and chamfers that touch nothing
snap and latch geometry: the hook, the catch, its travel	text, logos, texture
threads (size, pitch, length of engagement)	features for other variants of the product (extra bosses, knock-outs)
the envelope where it fits into or against something	the overall size where nothing constrains it

When in doubt, the assistant asks; "it fits into this recess" makes the outline critical.

Molded features and what to do with them

A molded part carries features that exist for the mold, not for the job. The prompt proposes what to do with each, and the reason goes in the spec's Why changed column. The rules (C1–C12) and their sources are in BEST_PRACTICES §19.

Molded feature	Why the mold has it	In the printed copy	Rule
Draft (faces taper along the mold's opening; molded 0.5–2°, more under texture)	so the part releases	remove it; keep a taper only where it does a job (a taper fit, parts that nest or stack) or is part of the look. Measure a drafted wall where it mates, or at mid-height, and say which	C1, C2
Thin even walls, coring (molded walls ~2–4 mm, thick areas hollowed)	to fill and cool evenly without sink	solid, or thicker walls in whole line widths; keep a cored pocket only where it does a job (weight, clearance, a mating feature)	C3
Ribs and gussets (molded 40–60 % of the wall)	stiffness without sink	keep where they run (the stiffening intent) at full wall thickness, or drop them where solid replaces them	C4
Sink marks, parting line and flash, ejector-pin marks, gate vestige, knit lines	marks of the molding	never model them or measure on them; a sink shows a thick section behind it; a crack on a knit line is a molding weakness	C5
Living hinge (PP or PE, 0.2–0.35 mm, a million cycles)	PP flexes for ever	cannot be copied: 0.4–0.6 mm in nylon, TPU or PP with the hinge width along Z (tens of cycles), or a pin or print-in-place hinge	C6
Snap fit (sized for POM, PP, nylon)	molded resins take high strain	recompute the arm for the print material's strain (the snap_arm recipe), print it flat, ~60 % of the strain for snaps opened often	C7
Screw bosses, self-tapping holes, molded threads, inserts	sized for sink and molding	the position and the screw, not the molded proportions: thicker wall and gussets; a heat-set insert or nut for screws taken out often; threads modeled only at M5–M6 and up; declare the screw as hardware	C8
Material (PP, POM, PA, ABS, PC)	chosen for strain, heat, creep, chemicals or sunlight	pick the print material by that property: strain or a hinge → PETG, nylon, TPU or a redesign; heat → ASA, PC, PA (PLA softens near 55–60 °C, molded PP ~80–85, POM ~110); outdoors → ASA; sliding → nylon. The copy is weaker across its layers: put the load along them and say so	C9
Mold shrinkage	the mold was cut oversize	none: the part is already at nominal size, so model what you measure; FDM shrinkage is the slicer's	C12
Extra bosses, knock-outs, features for other models of the product	one mold serves several products	drop what this copy does not use; say so in Why changed	C10

Measuring a part to copy: one datum, not chained; the design intent (what fits and works), not the as-built shape; threads by major diameter and pitch against a chart; a worn or broken part measured on its unworn areas and the mating part [C10]. Provisional (one source): round a reading to the value the designer meant (43.97 → 44 mm, inch vs mm, standard sizes) and say so [C11].

Heat, the usual surprise: heat deflection temperatures at 0.45 MPa are about 57–60 °C for PLA, 69–78 °C for PETG, 87–100 °C for ABS, 100–103 °C for ASA and 111–117 °C for PC (Bambu Lab and Polymaker data), against about 80–85 °C for molded polypropylene and 110 °C for acetal. A clip from a car or a dishwasher should not become PLA.

Strength across the layers: Bambu Lab's own impact data put the across-layer value at 12–52 % of the in-plane one, and CNC Kitchen's hooks printed standing were at least 50 % weaker. Turn the copy so its load runs along the layers (S4, S5).

The replication spec

The spec is the part's MEASURE.md: the header of the measurement sheet and one table with these columns (the first five are the sheet's own):

Column	What goes in it
Feature	what was measured: "box screw hole Ø", "wall at the clip root"
Value	the reading in mm; several readings separated by , or ; are averaged; in or " means inches; text in parentheses is a note: 83.3 (3-9/32 in, NEMA WD-6)
Method	the tool and how
Accuracy	what you can trust the reading to: ±0.05; without it the spec assumes ±0.1
Status	CONFIRMED or estimated (a ~ in front of the value means estimated too)
Class	critical or free
Source	measured, standard, derived, changed (for printing) or estimate
Field	the Dims field that models it, or a simple expression of fields: PLATE_W - 2 * RIM_WALL
Why changed	why the copy differs from the original: "draft removed, modeled at the mating height"

replication_spec(name) (CLI b123d-print spec <part>) joins it with the script's real Dims values (the variant's overrides, derived @property values and hole_comp() included) and prints one table plus flags:

| Feature | Field | Class | Source | Measured | Tol | Modeled | Δ | Why changed |
| Box screw hole Ø | `CORNER_HOLE_D` | critical | measured | 3.7 | ±0.05 | 3.7 | +0.00 |  |
| Wall around each corner counterbore | `BOSS_WALL` | free | changed |  |  | 2 |  | the molded boss is thin; ... |

- WARN Clip root thickness (MEASURE.md line 24): readings spread 0.30 mm, more than ±0.05:
  tapered (draft), oval or warped? Say where it was measured ...

Flag	When
WARN critical and not measured yet	a critical row with no reading
WARN critical but an estimate	read from a photo, ~, or estimated
WARN (critical) / note (free) readings spread	the readings differ by more than the accuracy
WARN (critical) / note (free) modeled differs	the Dims value is off the average by more than the accuracy and Why changed is empty
WARN unknown field	the Field names nothing in Dims
WARN critical with no Field	the spec cannot check it
note	rows without a class; a plain measurement sheet without the replication columns

A variant that changes a size on purpose (the bigger 140 wall plate) is not flagged: its differences are labeled (variant 140). Nothing here fails check; the spec is the record of the copy, and --strict exits 1 on a warning for scripts. The spec export format writes the same table to out/docs/<name>_spec.md, next to the print files.

Checking the copy against the original

• measure on the built part: hole diameters, spacings, the opening, against the spec's measured values.
• A fit-test template of each critical interface: the mounting face with its holes, the snap, the opening. Minutes to print; lay it on the original or the mating part.
• An outline: a 1:1 paper print of a slice, laid on the original; or trace a flatbed scan of the original and compare.
• With a scan: define original(d) returning import_model("ref/<scan>") placed like the copy, and call before_after: the free changes show in the deviation view, the critical faces should not move. compare gives the score.

Example

The plate_2gang_to_1gang example is a copy of an injection-molded 2-gang to 1-gang adapter plate, started from the original's drawing, not a mesh:

Its MEASURE.md keeps the hole spacings (NEMA standard), the hole and counterbore sizes and the opening as critical and modeled as measured. The outline and depths are free; the 140 variant makes them bigger. The molded back (a 2.4 mm face, a 6.5 mm rim, a cored inner frame with extra screw bosses for other boxes) became a solid frame around the opening with straight ribs, and the wall around each counterbore became 2.0 mm after the first print broke through a 0.05 mm skin there. Each of those rows says why.

b123d-print new plate --from plate_2gang_to_1gang
b123d-print spec plate