Update class mechanics, stat rulesets, and CoA reference catalogs

data/coa-reference/README.md

@@ -0,0 +1,29 @@
+# CoA offline reference
+
+This directory holds the local-first CoA reference dataset and the evidence for
+explicit active-catalog promotion. Historical reference builds and active
+promotion are separate commands.
+
+Run `npm run coa:reference:build` from the project root. The command reads the
+existing Ascension client, extracted base-client DBCs, and HealerMan's July talent
+snapshot. It never downloads anything or writes to those sources.
+
+`current.json` points to a verified immutable directory under `snapshots/` with
+`reference.json`, `audit.json`, `audit.md`, and a checksum manifest. Downloaded
+archive supplements live under `cache/`. These generated files are local and
+ignored by Git. The lossless July baseline under `baselines/`, the active
+`promotion-report.json`, pipeline, tests, and usage documentation are tracked.
+
+Use `npm run coa:reference:fetch` explicitly to cache the pinned August talent
+archive, then rebuild to include its comparison. The supplement remains separate
+from the primary July/local-client reference dataset.
+
+`npm run coa:reference:fetch:spells` selectively caches missing global BASE spell
+records and their linked payloads. `npm run coa:reference:promote` validates the
+cached August Vol'Jin trees and local/global spell evidence, then updates the
+active CoA catalog offline. Follow it with `npm run coa:catalogs:refresh` to
+regenerate CoA runtime/UI shards and missing icons. Classic catalog records and
+progression identities are protected; CoA readiness remains Needs Testing.
+
+See [the pipeline guide](../../docs/COA_REFERENCE_PIPELINE.md) for source roots,
+lookup examples, validation, failure handling, and version limitations.