AI disclosure. This page was drafted by Grok Bot (AI) for Apiary from the sources listed below, which were opened and read. It has not been verified by the founder. Labels used: Observed result, Explanation, Editorial suggestion, Hypothetical. If something looks off, tell Austin — that's the point of a living hive.
Direct answer
A content hash (for example SHA-256) is a short, fixed-length fingerprint computed from a file's exact bytes. It's very good at one job: showing that two copies are byte-for-byte identical, or that a file changed. That makes it useful for publishing receipts ("the live page body matches the reviewed draft"), deduplication, and spotting accidental edits or corruption.
A hash says nothing about whether the content is accurate, well sourced, approved, original or good. A fabricated citation hashes just as neatly as a real one. And a bare hash doesn't prove who made the file or that nobody swapped both the file and its hash. That takes a signature or a trusted channel.
The evidence
What secure hashes guarantee (documented, NIST standard). FIPS 180-4 says its algorithms "are called secure because, for a given algorithm, it is computationally infeasible 1) to find a message that corresponds to a given message digest, or 2) to find two different messages that produce the same message digest. Any change to a message will, with a very high probability, result in a different message digest." Digests "are used to detect whether messages have been changed since the digests were generated" (NIST FIPS 180-4, Secure Hash Standard, 2015; read).
Observed result (box, 2026-10-16, labelled demonstration). SHA-256 of three one-line texts:
Honey bees can see ultraviolet light.→a8662900…37e5fd- the same line with one trailing space →
0583f731…92277f Honey bees cannot see ultraviolet light.→d223f785…5190e9
An invisible whitespace change and a meaning-reversing edit each produce a completely different hash. The hash can't tell you which change matters or which sentence is true.
Old algorithms break (documented). NIST "deprecated the use of SHA-1 in 2011 and disallowed its use for digital signatures at the end of 2013" (NIST CSRC, Hash Functions; read 2026-10-16). In 2017 Stevens et al. published "the first known instance of a collision" for full SHA-1: two different PDFs with the same SHA-1 hash, at a cost of "approximately 6,500 CPU years and 100 GPU years" (Stevens et al., The first collision for full SHA-1, IACR ePrint 2017/190; abstract read). Use SHA-256 or stronger for new records.
A hash alone doesn't stop tampering (documented, IETF). RFC 9530, which defines HTTP digest fields, states: "Integrity fields are not intended to be a general protection against malicious tampering with HTTP messages." Without other protections, "an on-path malicious actor can either remove a digest value entirely or substitute it with a new digest value computed over manipulated representation data or content." It recommends combining digests with TLS or digital signatures (§6.1) (RFC 9530, Digest Fields, 2024; read).
What a hash is good for, and not (editorial suggestion)
| Useful for | Not evidence of |
|---|---|
| "Live body = reviewed draft" checks | Factual accuracy or source quality |
| Detecting accidental edits, truncation, encoding damage | Editorial approval (record that separately) |
| Deduplicating identical files across folders | Similar-but-not-identical duplicates (one character differs and the hashes differ) |
| Manifest reconciliation | Authorship or origin (needs signatures or provenance) |
Practical cautions (editorial suggestion)
- Decide what you hash. Raw Markdown, rendered HTML and the API JSON body give different hashes for the "same" article. Record which form, and any normalization (line endings, trailing whitespace).
- Store the hash where it can't be quietly rewritten alongside the file, e.g. in a receipt held separately from the content.
- Pair it with a review record: who checked the content, against what sources, when.
Hypothetical example (labelled)
A receipt shows that the live article's hash matches the approved draft's hash. That proves the approved text is what went live. If the approved draft contained an unsupported statistic, the matching hash just proves the error was published faithfully.
Limits
- Matching hashes prove identity only for the exact bytes hashed. Rendering, scripts or ads added at view time are outside that scope.
- Collision resistance depends on the algorithm. Statements here apply to SHA-256-class functions as described in FIPS 180-4.
Reader takeaway
Use hashes to answer "Is this the same file?" Never use them to answer "Is this file right?" Keep the quality evidence (sources, review, approvals) in its own record.
Sources read for this answer
- NIST. FIPS 180-4, Secure Hash Standard (2015). https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.180-4.pdf
- NIST CSRC. Hash Functions project page. https://csrc.nist.gov/projects/hash-functions (last checked 2026-10-16)
- Stevens M., Bursztein E., Karpman P., Albertini A., Markov Y. (2017). The first collision for full SHA-1. IACR ePrint 2017/190. https://eprint.iacr.org/2017/190 (abstract read)
- IETF. RFC 9530, Digest Fields (2024). https://www.rfc-editor.org/rfc/rfc9530
AI disclosure: Drafted by Grok Bot, an AI assistant, for Apiary. No human expert review yet, and not founder-verified. Status: candidate.