A money-mindsets section, and Tsumugu's reading layer

Two threads moved today: the public site grew a money section, and Tsumugu picked up the reading layer that makes it feel like the tools it’s meant to replace.

The site — a money-mindsets section

The top-of-mind WNAC question — what are good teachable mindsets for budgeting and investing? — needed a home before any of the applied AI work. So the base now has a money MOC: a reading list of six books, each distilled to its mindsets and first principles, with a full note started for the one already read — the Almanack. The rest get promoted to standalone notes as they’re read.

The same through-lines kept surfacing across all six — spend less than you earn, time beats timing, define enough, behaviour beats knowledge, money buys freedom not status, price things in life energy — which is a good sign the mindsets are real and not author quirks. The one tension worth holding: frugality-to-independence (Simple Path, Your Money or Your Life) against spending-down-on-experiences (Die With Zero), reconciled by enough and time-bucketing.

This is the mindset layer. The applied layer — where AI actually changes the budgeting and investment loop — is still WNAC‘s to work out, and the base will grow budgeting and investment nodes alongside it.

Tsumugu — the reading layer

Yesterday was the engine; today was making it read like Migaku. The reading layer landed across the reader and YouTube:

• Zhuyin ruby and coloured underlines. Each word can render 注音 above the character, aligned to its pre-baked reading, with a per-status underline ramp instead of a fill — the Migaku look, scoped so the default fill model stays untouched.
• A real Taiwan guard. OpenCC now runs cn→twp (Taiwan-idiom), so 軟件 becomes 軟體 and 信息 becomes 資訊 — not just character conversion, but the vocabulary a Taiwanese reader would actually use.
• Two-way sync without data loss. The word store now carries provenance on every word — when, where, and from which source its status last changed — which fixed a confirmed bug where re-importing could demote a word already learned. Re-imports are now idempotent.
• YouTube, the honest way. A synced reader highlights the playing line in Tsumugu’s own text; no code runs on youtube.com, with a local scrubber fallback when offline.
• A richer Migaku importer. Reading Migaku’s SQLite directly — skipping soft-deleted rows, keeping each word’s origin — seeds roughly 2,100 known Mandarin words with provenance, and re-applying blocks zero demotions.

Green throughout: 395 public plus 195 private tests. Still open — wiring real Netflix and YouTube caption ingest into a reading session, and the optional, default-off write-back to Migaku.