โ† back to pin list

Zero-click sheet intake -- ruled, chunk 1 (auto-extract at upload) BUILT

kerfmaster pin list ยท UI / workflow decisions

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 True because it was decided; it can be superseded, but it cannot be stale.

Content last changed 2026-10-18 โ€” computed from the item itself, not typed.

The five ruled expectations (Jordan 2026-10-05, each answered Yes by number):

Jordan's four design answers (2026-10-05, summarized -- recorded from the sitting, not verbatim): the dropped views/border stay visible as a ghost layer; the picked view is click-to-repick when the proof got it wrong; the extraction review is on demand, never an interruption; the Intake drawer states what was done in plain words.

CHUNK 1 BUILT 2026-10-05 (record): engine sheet_intake.auto_plan(text) + auto_extract(text) -- scale trusted only when the sheet states exactly ONE distinct scale note (two scales or none = SheetInvalid, operator asked); flat view picked by dimension proof (codes 13/14 defpoints assigned to the SMALLEST containing cluster so the border never wins; most votes, at least 2, strictly ahead of the runner-up = proven), else best-read = largest non-container cluster with proven:false; both ride the recorded extract() road so the output is byte-for-byte the confirm click's. NEW honest refusal in extract() when the annotation split leaves no cut geometry. Server: upload seam auto-extracts any detected sheet (writes source.dxf, extracted.auto=true, a refusal recorded as auto_refused -- the upload never fails on it); run_pipeline intake words carry the zero-click statement ('drawing sheet handled automatically... stated scale 1:4 (dimension-proven: N of M dimensions sit on this view)'), a best-read adds a yellow warning, a stood-down sheet keeps the detected warning with the reason appended. Client unchanged by construction. Proofs: 29/29 shipped-flat entity identity, 389-source sweep (auto fires only on the stated-scale class), test_sheet_intake 54/54, verify_dwgtray 20/20, identity gate green, F1 no cost, live browser end-to-end. Honest limit: the Inventor IDW class (79602) states its scale as an unrendered field expression, no text in the DXF -- stands down honestly. Rollback: tag pre-zeroclick-intake (bfbbe578).

Build chunks (2-3 intent): (1) auto-extract at the upload seam -- scale from the sheet's own scale note, flat cluster picked by dimension-proof / best-read with the pick stated and flagged when unproven, annotation + view ink dropped by the recorded brick-33 rules, original preserved as drawing.dxf; (2) the extracted flat lands on the one road as a part-list row, tab present; (3) a red 'unextracted drawing sheet' finding on every plan/post surface for any job that still cannot extract. Nothing built at pinning -- the chunks land one go at a time.

DIMENSION-DERIVED SCALE 2026-10-06 (Jordan 'Go' + Aristide 'Go', Jordan's answer 1 -- 'I thought we could calculate this from dimensions also?'): the honest limit above CLOSED. Measured finding: both CAD classes declare the drawing scale on their dimensions as DIMLFAC (group 144) -- SolidWorks/Lathrup in the DIMSTYLE table (4.0 on 1:4 sheets), Inventor/79602 as a per-dimension ACAD DSTYLE xdata override (0.25, measurement 17 on a 68 drawn span); displayed = drawn span x DIMLFAC, so DIMLFAC IS the scale. sheet_intake.dim_scale reads it per linear dimension (override > style > none; angular/ordinate, typed text overrides, undeclared or non-positive factors and a stored measurement matching neither reading give NO evidence -- never assumed 1); auto_plan picks the view first then _decide_scale: one note + agreeing or silent dims = the note governs ('confirmed by k of K dimensions'); note vs dims disagree = refuse naming both; several notes = dims pick one; no note + 2+ flat-view dims all declaring one factor = DERIVED ('4:1 (drawn at 4x full size) derived from 3 of 3 dimensions on the flat view'); anything else = honest refusal; the scale is never a guess. report.auto carries scale_how + dim_scale; the Intake note states it. PROVEN: 79602 now ZERO-CLICK at 0.25 -> 17x17 with all four holes, identical to the manual extract; 29/29 Lathrup sheets every dimension agrees with its stated note, zero conflicts, same scales; corpus sweep 2548 DXFs / 131 detected sheets -- 55 identical, 74 refusal-wording only, exactly 2 changed (79602 intended; the 3A76 customer DWG, drawn full size, now derives 1:1 and extracts 2.7179 x 3.8753, consistent with the 3A76 DWG precedent); zero crashes; test_sheet_intake 66 -> 81; verify_dwgtray 20/20 (check 9 updated to the intended 3A76 auto-extract; one check-15 miss on a reused scratch instance, green on re-run); identity gate GREEN (2.4s, byte-identical); F1 8.35/7.64 -> 8.32/7.62 + 8.31/7.61s, no cost. Ordinate dims excluded (conservative). Zero dev O-numbers; progseq untouched.