Twelve drawers, one chip, and the word nobody sees
How Thaw classifies a document on three axes at once, and why nothing is ever "Uncategorized".
Ask someone what a document is and they answer in three parts without noticing. “It’s the auto insurance policy for the Civic.” Insurance is an area of life. A policy is a kind of paper. The Civic is what it is about.
Most tools squash the first two into a single tag, and then the tag list grows until nobody wants to pick from it. Thaw keeps the three apart, because each one does different work.
Category: the drawer label
There are twelve, and the list is fixed: Taxes, Money, Insurance, Health, Home, Vehicles, Work, Business, Legal, Identity, Education, Purchases. Category is the color on every document and the top-level filter in the Library. It never grows, so it never needs pruning.
Each category carries an adjective. That adjective matters in a minute.
Type: the chip
Type is the name a person would say: W-2, Lease, Passport, Auto policy, EOB. It is the chip on every document. Types live in a registry that starts with about a hundred and twenty curated entries and grows from what the reader sees.
Each type is data, not code: which category and form it belongs to, the fields it usually carries and what role each plays (an identifier, an amount, a renewal date), how long it should be kept, and what to do about the dates in it. An auto policy inherits its carrier, policy number, term and premium fields from the insurance policy above it, so a new kind of policy the reader has never met still gets the right facts and still produces a renewal.
When the reader meets a document no type fits, it proposes one in plain words and files the document under it immediately. Proposed types are reviewed and either promoted for everyone, merged into an existing type, or set aside. The document is never blocked.
Form: the word nobody sees
Form is the kind of paper: policy, statement, bill, receipt, invoice, form, certificate, card, contract, letter, record, report, license, manual, note. It is never a label in the interface.
It does two jobs anyway. It powers search language (“show me all the statements”) and expectations (“a vehicle should have a current policy”). And it builds the fallback name: when the reader knows the category and the form but no specific type, the document is named adjective plus form. Insurance policy. Medical letter. Financial statement.
That name is real and filterable, which is why nothing in Thaw is ever “Other” or “Uncategorized”. There is always a drawer and always a kind of paper.
Subject is not a classification
Pets are not a drawer. A vet record is Health, about Biscuit. Subjects are the people, pets, vehicles, properties, accounts and businesses a document is linked to, and they get their own pages. Keeping subject out of the classification is what lets the twelve categories stay at twelve.