ASAP7 tech LEF vs. the design rule manual: what should the router be held to?

[jeffhsu3]

Checking ORFS ASAP7 designs against the DRM (asap7_drm_201207a.pdf) with gdscheck found that asap7_tech_1x_201209.lef diverges from the manual in about 20 rules. The full per-rule list, with the LEF statement for each, is here: gdscheck: Against the tech LEF. They fall into four kinds:

1. Rounded values that disable a rule: asap7: DATABASE MICRONS 4000, so the 1x tech LEFs' quarter-nanometre rules hold #4663 rounds them to whole nanometres; a review of asap7: DATABASE MICRONS 4000, so the tech LEF's quarter-nanometre rules hold siliconcompiler/lambdapdk#246 suggests DATABASE MICRONS 4000 instead, which holds them exactly.
2. The LEF's own vias break the DRM: VIA89 has no enclosure, and VIA78's cut isn't flush with M8 (asap7: fix M8 enclosure in VIA78 to be minwidth #142 widened its M8 pad to the minimum width but not the cut). A fix is ready; it changes routing for every design using M8/M9.
3. Rules the LEF states more weakly: M1 corner spacing, V6 cut spacing, the M8/M9 spacing table.
4. Rules the LEF can't or doesn't state: M1 tip spacing by edge length, V0/V1 M1 enclosure, M8/M9 width by length, grid and track rules for non-router shapes.

Maybe an update to the DRM itself might be better and go the other way and make the asap7_tech_1x_201209.lef the source of truth?

Related: #4541 (switching ASAP7 to the #4532 LEF; these divergences are in both files), The-OpenROAD-Project/asap7#22 (V3/V5 LEF58_ENCLOSURE in the 4x LEF) and The-OpenROAD-Project/asap7#3 (an M6 SPACING that contradicts its own table).

Flow-side fixes in progress: an on-grid ASAP7 PDN (asap7-pdn-ongrid; ORFS gcd/aes go to zero DRM violations on router-held rules, with lower IR drop), fin-grid macro snapping, and The-OpenROAD-Project/OpenROAD#11700 for syn's DEF units.

Questions:

1. Should the tech LEF follow the DRM where LEF can express it (kinds 2–3), accepting QoR changes on existing designs?
2. For kind 4, is a documented list in platforms/asap7/README.md the right home?
3. Should ORFS's ASAP7 DRC (asap7.lydrc) check the grid and track rules? It currently has them switched off (OFFGRID = false).