4d-3. Beam compensation column + micro-segment cleanup

kerfmaster pin list · Big capabilities

rulingA decision by Aristide or Jordan. True because it was decided; it can be superseded, but it cannot be stale.

Content last changed 2026-08-14 — computed from the item itself, not typed.

Contract — Post stage

This contract states a decision, not an implementation.

takes material-table

makes gcodenc-audit ?

fails if A compensated block shorter than the material's beam comp value in anything we post; a micro-segment surviving into G-code; a material row posting without a beam_comp value.

Contract last changed 2026-08-14 — computed, not typed. Dated separately from the text above, so neither date can speak for the other.

The defect, measured on the laser (Jordan, 2026-08-14): “we got comp interference alarms on all the files we've run so far. -81 was run first and had a few. This is caused when points are closer together than the beam comp value.” Confirmed in every delivered program: 15 of 19 carried compensated moves shorter than .007 — ~700 total, one literally zero-length. They ride in from the customer's dirty CAD: near-coincident points that survive healing because they sit far below the healing tolerance, then post verbatim. The audit checked comp SIDE and geometry faithfulness but never block length — that gap let all of them ship.

Ruling (Aristide, 2026-08-14): “Add a column to the table for beam compensation and we are going to give you those values per material and thickness. For 1/8-inch steel it is going to be 7 thou. That's a micro segment cleanup. and it should never end up with points smaller than that value… make a good note that we give you the values later. for now we just keep testing everything on 1/8” steel.”

Wired (same commit as this pin):

materials.json — beam_comp column; MSO7,0.125 = 0.007 (decided, Aristide). NO other row carries one — VALUES ARRIVE FROM ARISTIDE AND JORDAN per material × thickness; a row without it cannot post (MaterialIncomplete), which is deliberate: an alarming program must never ship silently again. Numerically equal to kerf on 1/8 TODAY, but its own column by ruling — if a future row's beam_comp and kerf differ, ask which feeds the D register.
emit.merge_micro_segments — any prim whose PATH length (line span / arc r·sweep — never chord, so full-circle holes are untouched) is under beam_comp is absorbed into a neighbour; a line extends over it, between two arcs the following arc is rebuilt at its own radius. Each absorption moves the path by at most the removed length — sanctioned sub-tolerance smoothing (Q27). All-debris contours drop, counted. Stats: micro_merged / micro_dropped / micro_worst / beam_comp.
nc_audit — new check: every compensated block's path length against the material's beam_comp (from the header if not passed); shorts are an anomaly naming bnd/position. Skipped-and-said-so for materials with no value (third-party NC). Geometry gate for cleaned programs widens to max(0.005, beam_comp) — the sanctioned cleanup bound.
build_board generic post now MSO7,0.125 (testing-on-1/8); check_corpus passes beam_comp=0.0 (byte-reproduction control, never a deliverable).

Verified: O64's nest (41 micro prims of 150) re-posted as 1/8: 41 absorbed, 0 short blocks in output, audit clean, path moved 0.0002 from source. The O64 post used to carry 40 alarm blocks.

Every program delivered before 2026-08-14 predates the rule and alarms on the machine. Ruling (Aristide, 2026-08-14): “no need to redeliver. moving forward always apply this.” The delivered programs stay as they are; no re-posts. The cleanup + short-block audit run on every post from here on. The 14-ga and 0.187 rows still await their beam_comp values — until then those materials cannot post at all (MaterialIncomplete), and all testing stays on 1/8.