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

History-free snapshot of the private development repository at cbe45c5
(2026-09-09). Code, docs and public catalogue only; no plant or job
knowledge layers, no client or job names (tools/scrub.py clean). The
intro deck is left out of this snapshot because its slide images could
not be checked for job drawings.

knowledge/README.md

@@ -0,0 +1,66 @@
+# `knowledge/` in this checkout
+
+The plant layer and the job layer are not here — never were, never will be
+in a colleague distribution. `catalogue.yaml` **is** here, but it is a
+**carve-out**, not the owner's real file: `tools/make_dist.py`'s
+`build_catalogue_carveout()` reads the real one and re-emits only the
+branches that are trade-standard or that are this tool's own built-in
+configuration, never a manufacturer's product rating. Read the header at
+the top of `catalogue.yaml` itself for the exact list of what's in and
+what's out, and why. `catalogue.yaml.template` is a separate thing: a
+blank-value authoring template for building a real knowledge layer from
+scratch, not what this carve-out is built from.
+
+## Why
+
+WillTools reads its facts in layers, each overriding the one before it:
+
+| layer | where | what it is |
+| --- | --- | --- |
+| `knowledge.DEFAULTS` | in the code | what the tool assumes when nobody has said |
+| `knowledge/catalogue.yaml` | here, carved out | published hardware ratings — manufacturer data |
+| plant layer | `%APPDATA%\WillTools\knowledge.yaml` | one office's naming conventions, casting-bed rails, parameter names |
+| job layer | `<model folder>/_ticketqc/knowledge.yaml` | what this job does differently |
+| project config | wherever the engineer put it | the pinned, merged answer for one job |
+
+The plant and job layers, and `catalogue-rationale.md`, are either a
+specific office's private template or narrate real measured incidents —
+neither belongs in a colleague trial. The catalogue's two real capacity
+tables (`catalogue.inserts`, `connectors.products`) ship **empty** for the
+same reason a manufacturer's rating is not this tool's to redistribute.
+Ask whoever gave you this checkout for the real layers, and the real
+catalogue, once you are ready to point the tools at a real job.
+
+## What runs without the missing pieces
+
+Everything that does not need a lookup: unit conversion, the ribbon itself,
+Self Check, and — outside Revit entirely — `panellift`'s handling
+calculations, which read a frozen job file and never touch `knowledge/` at
+all (see `examples/example-panel.job.json`).
+
+Everything the shipped carve-out covers also runs for real: face
+designations, the up-end orientation rule, ASTM rebar grades, and the whole
+shop-ticket QC checklist (row ids, grades, which check answers which row) —
+none of that needs the real, private file. What stands down cleanly is only
+what actually needs a manufacturer's number: a lifting insert's rated
+capacity, a sandwich-wall tie's rated capacity, and anything downstream of
+either.  `python tools/setup_machine.py --check` reports this state as a
+pass, not a failure — a distribution with the carve-out and no product
+data is an expected, first-class state, not a broken one. Self Check
+reports the same thing inside Revit, row by row.
+
+### Running the test suites here
+
+`CONTRIBUTING.md`'s "What does not run in a trial install" has the full
+accounting. Short version: both suites' `tests/conftest.py` carries an
+appended, auto-detecting block (written by `tools/make_dist.py`'s
+`append_test_exclusions()`) that skips, by name and with a reason, the
+handful of tests that assert the owner's real published numbers rather
+than this tool's logic. Drop in the real `knowledge/catalogue.yaml` and
+those tests un-skip on their own — nothing to edit.
+
+## Getting real knowledge layers
+
+Ask the owner. They come from a second, private repository that is not part
+of this one on purpose — see the top of `.gitignore` in the repo root for
+why the split exists at all.