WillTools 2026-09-09 -- clean snapshot of the development repo for review

History-free snapshot of the private development repository at 77a82eb.
Code, docs and the public catalogue only: no plant or job knowledge
layers, no client, job or producer names (tools/scrub.py clean against
the widened term list; a second word-level grep clean). The intro deck is
left out because its slide images cannot be checked for job drawings.

WillTools.extension/lib/vibe/pa_prefill.py

@@ -0,0 +1,64 @@
+# -*- coding: utf-8 -*-
+"""Panel Analysis: the pure merge rule for the model-measured pre-fill.
+
+`script.py`'s `_reads_for` reads three separate things off one panel --
+`pa_supports.read_supports`, `pa_lifters.read_lifters`, `pa_reinf.read_
+reinforcement_prefill` -- each returning its own small dict of form-field
+values (or {}, defensively, when it could not measure anything). This module
+holds the one rule for folding all three into the button's running `reads`
+dict, kept apart from `script.py` because that file imports the Revit API at
+module scope and cannot be imported under `pytest` on host CPython (see
+`tests/conftest.py`'s docstring) -- so the part that is pure logic, and
+therefore worth a fast host test, lives here instead.
+
+THE RULE, in one sentence: everywhere else in `_reads_for` a value already
+read off the piece wins over a model guess (`setdefault` -- fill only what is
+still blank); `OVERRIDE_KEYS` are the exception, because for those keys the
+NEW read *is* the measurement and anything already sitting in `reads` is a
+weaker guess from an earlier, cruder pass -- `mild_as` is the clearest case:
+`_assembly_reinforcement`'s old marks-list (a bare comma list of piece marks)
+is worth less than `pa_reinf.rebar_prefill`'s `(bar size, type name)` summary
+with a real total area, so the new value has to win even though `reads`
+already carries something under that key.
+
+`None` never overwrites anything, for any key, override or not -- a module
+that could not measure something says so by leaving the key out (or handing
+back `None`), never by asserting a blank.
+"""
+
+# The new measurement wins over whatever is already in `reads`, blank or not.
+# Every other key follows plain `setdefault` -- only a still-empty slot gets
+# filled. See the module docstring for why `mild_as` is here: the new
+# `(bar size, type name)` summary from `pa_reinf.rebar_prefill` outranks the
+# old marks-list `_assembly_reinforcement` may already have written under the
+# same key.
+OVERRIDE_KEYS = ('supports_ft', 'face_lifters', 'edge_lifters',
+                 'ps_rows', 'mild_as')
+
+
+def merge_model_reads(reads, parts):
+    """Fold `parts` into `reads` IN PLACE, per `OVERRIDE_KEYS`, and return it.
+
+    `parts` is either one dict or a list/tuple of dicts (`pa_supports.
+    read_supports`'s, `pa_lifters.read_lifters`'s, `pa_reinf.read_
+    reinforcement_prefill`'s, in whatever order the caller has them handy --
+    order only matters between two dicts that both carry the same key, which
+    none of the three do today). A `None` value in any part is skipped
+    outright; it never counts as "the model answered" and never overwrites
+    an existing value OR fills a blank.
+
+    Returns `reads` (the same object, mutated) so a caller can chain it, but
+    the mutation is the real effect -- this is not written to be used as a
+    pure expression.
+    """
+    if isinstance(parts, dict):
+        parts = [parts]
+    for part in (parts or []):
+        if not part:
+            continue
+        for key, value in part.items():
+            if value is None:
+                continue
+            if key in OVERRIDE_KEYS or reads.get(key) is None:
+                reads[key] = value
+    return reads