Discriminator classifier

classify:discriminator reads a configured JSON Pointer from the record, grabs whatever string is there, and proposes it as the class name. That is the whole recipe. A _type: "Feat" becomes a Feat proposal at confidence 1.0. A new _type value you have never seen before classifies automatically — the discriminator does not need to know your class list up front.

Priority defaults to 50; set it to 80 to beat every enumerative fallback. One field, any domain, no enumeration to maintain.

Configuration

Wire the discriminator inside register(dispatcher) in your plugin's index.ts:

import { DiscriminatorClassifierNode } from 'squashage/classifiers';

dispatcher.registerNode(new DiscriminatorClassifierNode({
  from:     '/_type',       // JSON Pointer (RFC 6901) into the record
  sanitize: 'pascalCase',   // optional — 'verbatim' | 'pascalCase' | 'kebabToPascal'
  priority: 80,             // optional — defaults to 50
  fallback: '/category',    // optional — second pointer tried when `from` is absent
}));

Field reference

Field	Type	Required	Default	Notes
from	string	yes		JSON Pointer (RFC 6901) into the record. Any non-empty string at this path becomes the proposed class name.
sanitize	string	no	'verbatim'	'verbatim' — use the value as-is. 'pascalCase' — split on -, _, or whitespace, capitalise each segment. 'kebabToPascal' — same transform as pascalCase.
priority	number	no	50	Written onto the emitted ClassificationProposal. Higher wins conflict resolution.
fallback	string	no		Pointer evaluated when from resolves to undefined or a non-string.

When the resolved value is an empty string, or when both from and fallback resolve to nothing, the node outputs no-match and no proposal is written.

Emitted proposal

{
  "source":     "classify:discriminator",
  "className":  "Feat",
  "priority":   80,
  "confidence": 1.0,
  "reasons": [
    "discriminator at \"/_type\" resolved to \"feat\""
  ]
}

Example — plugin registration

// plugins/aonprd/index.ts
import { DiscriminatorClassifierNode } from 'squashage/classifiers';
import type { SquashageDagonizer } from 'squashage';

export function register(dispatcher: SquashageDagonizer): void {
  dispatcher.registerNode(new DiscriminatorClassifierNode({
    from:     '/_type',
    sanitize: 'pascalCase',
    priority: 80,
  }));
}

With this config, a record carrying "_type": "monster-family" is classified as MonsterFamily. Add a new _type value to the dump, get a new class — no code changes.

See also

• Classifier cascade — full classifier set and conflict resolution.
• Structural classifier — predicate-based fallback for records without a _type field.
• Configuration — full config reference.