percentThreshold is accepted and validated but discarded before comparison

The public comparison options accept percentThreshold, and prepareOptions rejects negative/non-finite values with a threshold-specific error. However, a valid value is then discarded: PreparedOptions has no threshold field, the returned object omits it, and compare always uses differentPixels === 0 for equal.

For two fully opaque 10x1 RGBA images differing at exactly one pixel, omitted/0/0.1/1/10/50/100 all return { equal: false, differentPixels: 1, totalPixels: 10 }. This occurs in both strict and non-strict mode with caret/antialiasing ignoring disabled. Negative, NaN and Infinity thresholds throw, so this is an actively accepted/validated option that does not affect comparison, rather than an arbitrary unknown JS property.

The option was introduced in #3 alongside pixel counts. Its tests calculate a percentage on the caller side and do not exercise percentThreshold. I found no documented unit or boundary convention, so this report does not assume a particular 0..1 versus 0..100 interpretation. Please either wire the option into a documented threshold policy or remove/deprecate the inert option and direct callers to compare the returned counts themselves. Existing raw differentPixels/totalPixels should remain available.

Reproduced on main 81a79bc3665a4fedab02d3a8e273b9ac5d1af906 (v0.0.10), Bun 1.4.2. Save at the repository root as reproduce-percent-threshold.ts and run bun reproduce-percent-threshold.ts:

import { encode } from "fast-png"
import looksSame from "./lib/looks-same"

const referenceData = new Uint8Array(10 * 4)
for (let i = 3; i < referenceData.length; i += 4) referenceData[i] = 255
const currentData = referenceData.slice()
currentData[0] = 255
const png = (data: Uint8Array) => encode({ width: 10, height: 1, channels: 4, depth: 8, data })
const reference = png(referenceData)
const current = png(currentData)
for (const strict of [false, true]) for (const percentThreshold of [undefined, 0, 0.1, 1, 10, 50, 100]) {
  const result = await looksSame(reference, current, {
    percentThreshold, strict, ignoreCaret: false, ignoreAntialiasing: false,
  })
  console.log(JSON.stringify({ strict, percentThreshold: percentThreshold ?? "omitted", ...result }))
}
for (const percentThreshold of [-1, Number.NaN, Number.POSITIVE_INFINITY]) {
  try {
    await looksSame(reference, current, { percentThreshold })
    throw new Error("invalid threshold unexpectedly accepted")
  } catch (error) {
    if (!(error instanceof TypeError)) throw error
    console.log(JSON.stringify({ invalid: String(percentThreshold), error: error.message }))
  }
}
console.log(JSON.stringify({ identicalControl: await looksSame(reference, reference, { percentThreshold: 0 }) }))

Validation: all 14 changed-image configurations returned the same one-pixel difference; all three invalid-value controls threw the documented TypeError, and the identical-image control returned equal=true. TypeScript check passed. This is independent of #39: all source pixels here have alpha 255.

Prepared with Codex (Astra) assistance and verified locally.

Start in lib/looks-same and trace prepareOptions, PreparedOptions, and compare to see where percentThreshold is validated and lost. Run reproduce-percent-threshold.ts first, then inspect the existing comparison tests; done means the option has a documented, tested policy or is removed/deprecated while differentPixels and totalPixels remain available.