Classification design

The capstone, in two parts. Part 1 at week 7 is the system the rest of the semester tries to break. Part 2 at the end is the transfer test — run it on material this course never gave you, and declare what does not survive.

Due: 28 May 2027 · Weight: 30%

The brief#

Part 1, week 7: design the classification system this course will spend its second half trying to break, and say in advance what it will not decide.

Part 2, the end of semester: run that system on material this course never gave you, and declare which of its characters the new material does not supply.

Weeks 1 to 6 have supplied the parts. The formula gives three slots. Morphology gives a set of body parts that recur. Geography gives a habitat field that behaves differently in the 山經 (Shānjīng, the Mountains chapters) than in the 海經 (Hǎijīng, the Seas chapters). Omens and uses give two more fields, and both turn out to belong to the formula rather than to the animal. A system is what you get when those parts are fixed in an order and a rule is written for each.

The proposal is judged on its rules, not on its coverage. A system with five fields, each defined tightly enough that two readers filling in 鹿蜀 (lùshǔ) from 有獸焉，其狀如馬而白首，其文如虎而赤尾，其音如謠，其名曰鹿蜀，佩之宜子孫。 would produce the same five values, is stronger than a system with twelve fields that each invite judgement. Tight fields fail visibly. Loose fields absorb anything and report nothing.

The hardest requirement is the refusal. Every taxonomy decides some questions and declines others, and the ones that decline in writing are the ones that can be argued with. 陆吾 (lùwú) and 开明兽 (kāimíngshòu) stand on the same mountain with nearly the same body, and the text files one as a 神 and the other as a 獸. A system that quietly picks one answer has hidden its most interesting decision. A system that states it will not rank 神 against 獸, and says what it loses by not ranking them, has made the same decision in public.

What you submit#

Part 1 at the end of week 7: two thousand words or the equivalent as a specified schema plus commentary. A table of fields alone is not a design; a rule has to accompany each field.

Part 2 at the end of semester: the new material’s records filed, plus the list of characters it did not supply. The filing can be a table. The list cannot — each entry needs the sentence saying what you did about it.

Work the system through at least three creatures from weeks 1 to 6. One of them must be a creature the system handles badly. Name the field that breaks, and say whether the fix is a new field, a looser rule, or a refusal.

Nothing in Part 1 is provisional in the wrong direction. The second half of the semester is spent putting this system under load, and the reclassification essay and the bestiary entry both report against it. A design that hedges every rule gives those later pieces nothing to push on.

Part 2, and why it is last#

Part 1 can be passed by a scheme for one book. Part 2 is what separates a design from a transcription: take the system you proposed in week 7, amended by whatever weeks 8 to 12 did to it, and run it on records this course never supplied.

Choose the material yourself. A field guide, a fault log, a parts catalogue, a tasting note, a set of case reports — anything with repeated records in it and enough of them to count. Cite it the way you cite the 山海经 (Shānhǎijīng, the Classic of Mountains and Seas): a locatable reference per record, and nothing quoted that you cannot point at.

Then write the list this whole course has been for. Which of your characters does the new material not supply? For each one, say what you did — dropped it, renamed it, filled it by inference and declared the inference, or refused the record. Week 11 is the rehearsal for this and week 12’s congress is where the objections come from, so it is last for a reason: it is the only piece you cannot write before the system has been broken.

A Part 2 reporting that everything transferred has chosen material too close to the Classic of Mountains and Seas, or has not looked.

How it is marked#

The criteria weight definition and refusal above breadth. A four-field system that survives 旋龟 (xuánguī) scores above a fifteen-field system that has never been tried on anything.

Part 2 carries 35 of the 100 and is marked on the declarations rather than on how well the system did. A design that transfers badly and says exactly where scores above one that reports a clean fit without evidence.