Skip to main content

RawIndexData

Payload describing a user-defined model index. Maps directly to the ModelIndex schema's flat key/value vectors with optional slicing.

Mode is inferred from which fields are populated:

  • keysOnly: only keys.
  • oneToOne: keys + values, no start/end.
  • oneToNLinear: keys + values + end.
  • oneToNNonLinear: keys + values + start + end.

Keys must be homogeneous (number[] or string[]). Values, if present, must also be homogeneous.

Properties​

end?​

optional end: number[]

Per-key end offset into values (exclusive). Empty for keys-only and 1:1.

keys​

keys: string[] | number[]

Key vector. Drives the storage type (number_keys or string_keys).

Performance: prefer number keys for large or hot indexes. Number keys are read straight from a Uint32Array, while string keys cost a UTF-8 decode plus a long-string hash on every lookup-map build. If your keys are long strings (e.g. hashes), hash them down to a uint32 at build time and keep a small side table if you need the original string back; both the build and every lookup get dramatically cheaper, and it also keeps the worker-to-main payload a plain typed-array copy.

name​

name: string

Index identifier, unique within the model.

start?​

optional start: number[]

Per-key start offset into values. Empty unless slicing is non-linear.

values?​

optional values: string[] | number[]

Value vector. Drives number_values or string_values. Omit for keys-only.