Methodology
This page explains the search scope, the limits, and the review workflow behind 221B. The product is designed to help users discover publicly available candidate pages and then verify those pages manually.
Public-web candidate retrieval
Manual source review required
Last reviewed: August 6, 2026
Search scope
Out of scope
Review workflow
Use only an image you own or are permitted to use for this purpose. Prefer one visible, front-facing face with stable lighting and minimal obstruction. Do not upload intimate images or images of children.
Use exact-image search for copies, crops, edits, or the same photo. Use facial-similarity search only when different public photos are necessary for the legitimate question.
A 0–100 match-quality score helps prioritise which pages to inspect first. It is not a probability and does not prove that a person has been identified correctly.
Check the URL, publisher, date, image role, names, usernames, captions, linked pages, and surrounding context. A person pictured in an article may not be the named subject.
Distinguish independent sources from reposts, seek an authoritative second source where appropriate, and record plausible alternatives. Ten copies of one image remain one source chain.
Conclude at the source level as supported, not supported, or unresolved. Stop when the narrow purpose is answered. A no-result is inconclusive and does not prove authenticity, identity, privacy, or absence from the web.
Evidence labels
The same or materially edited image appears on the reviewed source. This establishes reuse, not who controls an account.
The page contains a visually similar face and needs further source review.
The image is similar, but the available context does not support the material claim.
A public source conflicts with a material statement, but the conflict still needs proportionate review.
The available evidence does not support a responsible conclusion.
Avoid labels such as identified, verified person, criminal, suspect, or safe unless an authoritative source independently establishes that exact claim and the use is lawful and appropriate. A 98/100 rank is still not a 98% probability.
Coverage and no-result
221B does not maintain a complete or permanent copy of the web. Retrieval can change between searches without either result being dishonest or definitive.
Pages can appear, change, move behind a login, block crawling, remove an image, or disappear. A public page is not guaranteed to remain searchable.
A newly published or rarely linked page may not yet appear in an index. Coverage is not a live copy of every public page on the web.
Resolution, crop, pose, lighting, obstruction, compression, filters, age, and changed appearance can alter which candidates are retrieved.
A result can be an outdated page, lookalike, namesake, repost, or page where the face is incidental. Original-source context determines relevance.
Image and result data
Provider terms and product behaviour can change. Read the current Privacy Policy before uploading sensitive material and request deletion when needed.
Prohibited and high-impact uses
For missing-person, exploitation, emergency, legal, or other high-stakes situations, use competent authorities and specialist organisations rather than amateur facial identification or public crowdsourcing.
Why this matters
The homepage, blog content, and methodology should all say the same thing clearly: 221B is built to help users inspect public-web matches responsibly.
That means being explicit about scope, input quality, changing coverage, provider processing, source context, uncertainty, and prohibited uses. A trustworthy workflow preserves the reader's ability to review, disagree, correct, and stop.
Editorial principles
