lril3d-dfm

name	lril3d-dfm
description	Use when a part needs checking for printability rather than correctness — before committing to a print, after a design change, or when the user asks whether something will print, needs supports, or is too thin.

lril3d-dfm — will it print?

Thin skill, thick library. No thresholds belong in this file. Every number lives in profiles/dfm-rules.json next to the source that justifies it. Read PRD.md §7 for where this sits in the loop.

lril3d-inspect answers "is this the part I asked for?". This answers "will that part come off the plate?". They are independent: a dimensionally perfect part can be unprintable, and a printable part can be the wrong part.

Run it

from threedp import dfm
report = dfm.evaluate(feats.mesh, "PLA_generic", part="bearing-holder")
print(report)

Materials are the keys of profiles/dfm-rules.json — the same names as profiles/calibration.json. An unknown one raises and names the valid list; do not guess one.

Read the report

Every line carries the measured value, the threshold it was compared against, and the source string. Report those, not an impression.

BLOCKER max_overhang_deg  measured 75.099 deg  threshold 45.000 deg  above the 45.000 deg
        maximum - past this angle from vertical the underside droops without support
        [conventional FDM support threshold, measured from vertical]

• A BLOCKER is a blocker. Do not downgrade one in prose. If you believe the threshold is wrong rather than the part, say that explicitly and ask — a rule quietly narrated away is a rule that is not there.
• A WARNING is not a gate, and saying so is not the same as ignoring it. Report it with its number and let the user decide.
• skipped lines are not clean bills of health. "No bore was small" and "no bore was measurable" are different facts and the report distinguishes them. Pass the distinction on.
• Overhang angles are measured from vertical: 0 is a vertical wall and perfectly fine, 90 is a horizontal ceiling and the worst case.
• Wall, feature and bridge numbers are sampled or derived from face geometry, and every one of them says ESTIMATE. Never present one as a dimension.

Changing a threshold

Edit profiles/dfm-rules.json, and change the source in the same edit. A threshold with no traceable source will not load — dfm.load_rules refuses it — and that refusal is deliberate: an uncited number cannot be argued with by anyone who was not in the room.

Per-material overrides sit under the material key and inherit _defaults for everything they do not mention. ABS and PETG are stricter on overhangs and bridges than PLA, and the reason is in the note on each rule.

Making DFM gate a verdict

By default it does not, and that is correct: DFM is advice about a process, intent.json is a claim about a part. To make it gate for a specific part, assert on it:

{"dfm_blockers": [0, 0], "source": "user-confirmed",
 "measure": {"kind": "dfm_violation_count", "severity": "BLOCKER", "material": "PLA_generic"}}

Now a DFM regression fails an ordinary assertion, in the ordinary report, with a measured count.

Non-negotiable

Banned from a critique: "looks printable", "seems fine", "should be OK", "appears thin". Every claim carries a measured number and the rule it was measured against, or an explicit ESTIMATE label. State the verdict plainly at the top — blockers or none — then the numbers, then what to change and by how much.