fix(scoring): use normalised key weights in _searchObjectList

[JSap0914]

Problem

_searchObjectList built its key list from this._myIndex.keys, which stores the raw user-specified weights — before the normalisation that KeyStore performs (weights sum to 1). _searchObjectList never read from KeyStore, so it applied the original values directly.

This caused two observable bugs:

1. Score underflow for extreme weight values

A single key always normalises to weight = 1.0 (only key = 100 % share). With a raw weight of 100, the exact-match formula becomes:

Math.pow(Number.EPSILON, 100 * fieldNorm)  // underflows to exactly 0

whereas the normalised exponent (1.0) gives ~2.2e-16.

const f1   = new Fuse([{ text: 'hello' }], { keys: [{ name: 'text', weight: 1   }], includeScore: true })
const f100 = new Fuse([{ text: 'hello' }], { keys: [{ name: 'text', weight: 100 }], includeScore: true })

f1.search('hello')[0].score   // ~2.2e-16  ✓
f100.search('hello')[0].score // 0         ✗  (underflow)

2. Score inconsistency between query forms

• Plain string queries → _searchObjectList → raw weights
• Logical Expression queries → _searchLogical → this._keyStore.get(keyId) → normalised weights

Identical searches against identical data produced different absolute scores depending solely on how the query was expressed.

Fix

Replace

const { keys, records } = this._myIndex   // raw weights

with

const { records } = this._myIndex
const keys = this._keyStore.keys()        // normalised weights

KeyStore and FuseIndex both iterate this.options.keys in the same order, so the positional indexing item[keyIndex] is unaffected.

Tests

Three new tests in test/key-weight-normalization.test.js (import from source):

Test	RED before	GREEN after
weight:1 and weight:2 (single key) produce the same score	✗	✓
weight:100 does not underflow to 0	✗	✓
Scaling all weights by the same factor does not change ordering or scores	✗	✓

All 374 existing tests continue to pass.

Test Files  21 passed (21)
     Tests  374 passed (374)

Closes #833

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[krisk]

Thanks, the diagnosis is spot on. _searchObjectList scored off the raw index weights while the keyed logical path already went through the KeyStore, so the two disagreed and big weights underflowed.

I landed a slightly wider fix on main instead of merging this directly, for two reasons:

1. The keyless logical form had the same bug. search({ $or: ['hello'] }) also read this._myIndex.keys, so weight: 100 still underflowed there even with _searchObjectList fixed. Wanted all three forms consistent.

2. Swapping in _keyStore.keys() positionally misaligns if a prebuilt index's key order differs from options.keys. I used a small _normalizedKeys() helper that looks each key up by id, so the weight stays pinned to the right value.

Your tests carried over almost as-is. Closing in favor of the version on main. Credit's yours.