Three days of raw AI answers to one question, each file with its SHA-256 hash and the commit that added it. A ten-line script recomputes the hashes on your machine.
One question, one engine, three days in September 2026. Each file is the raw answer file from our public record, namebeam-visibility-index. The commit column is the first (and only) commit that touched that file, with the date GitHub shows for it.
| Date | Engine | Question | File | SHA-256 | First commit |
|---|---|---|---|---|---|
| 2026-09-28 | Perplexity (API) | Who is the best roofing company in Dallas for a roof replacement? | NB-CZ-API_2026-09-28_segment_roofing_dallas_perplexity.json | 298f6fdb8f2c349776ffa73032119cef5d53abf684c90c0b49e853ea56a15a75 | 7d97fe4019 2026-09-28 |
| 2026-09-29 | Perplexity (API) | Who is the best roofing company in Dallas for a roof replacement? | NB-CZ-API_2026-09-29_segment_roofing_dallas_perplexity.json | 41274d0127f1291a6854f6b9c199619992b65ec1e7a7f9baf67fe2ec6b9df875 | b612984a05 2026-09-29 |
| 2026-09-30 | Perplexity (API) | Who is the best roofing company in Dallas for a roof replacement? | NB-CZ-API_2026-09-30_segment_roofing_dallas_perplexity.json | 04ae95798cc36aa92072376c2292ad5951e2dbdbd82594f497522134ccc76b0b | f0b6d92c22 2026-09-30 |
The hashes in this table are first published on this page, on 2026-10-06. The commit dates come from the git history of the public repository, which does not depend on this page.
Python 3, standard library only. It downloads each file from the public repository, hashes the bytes, and prints the result next to the hash in the table above.
import hashlib, urllib.request
B = "https://raw.githubusercontent.com/mrcrtr1979-droid/namebeam-visibility-index/main/corpus/e1/"
F = {
"NB-CZ-API_2026-09-28_segment_roofing_dallas_perplexity.json": "298f6fdb8f2c349776ffa73032119cef5d53abf684c90c0b49e853ea56a15a75",
"NB-CZ-API_2026-09-29_segment_roofing_dallas_perplexity.json": "41274d0127f1291a6854f6b9c199619992b65ec1e7a7f9baf67fe2ec6b9df875",
"NB-CZ-API_2026-09-30_segment_roofing_dallas_perplexity.json": "04ae95798cc36aa92072376c2292ad5951e2dbdbd82594f497522134ccc76b0b",
}
for name, published in F.items():
got = hashlib.sha256(urllib.request.urlopen(B + name).read()).hexdigest()
print(got, "MATCH" if got == published else "DIFFERENT", name)
Expected output: each line starts with the same hash as the table and says MATCH. If any line says DIFFERENT, the file has changed since the table was written, and we want to hear about it.
Each answer file in the public repository is stored in git, so each one has a git blob hash (a SHA-1 of its contents) and the commit that added it, with a date. Anyone can read both after cloning the repository. To check a file, run the three lines below, replacing the path with any file in corpus/e1/: the commit line shows when the file was added, and the blob hash printed by git hash-object should equal the one in the git ls-tree line.
git clone https://github.com/mrcrtr1979-droid/namebeam-visibility-index && cd namebeam-visibility-index
git log --format="%H %ad" -- corpus/e1/NB-CZ-API_2026-09-28_segment_roofing_dallas_perplexity.json
git hash-object corpus/e1/NB-CZ-API_2026-09-28_segment_roofing_dallas_perplexity.json && git ls-tree HEAD corpus/e1/NB-CZ-API_2026-09-28_segment_roofing_dallas_perplexity.json
Example, run on the 2026-09-28 file above: the commit is 7d97fe4019944612511db3230c89f8a8834146d9, dated Mon Sep 28 20:16:42 2026 +0000, and both blob hashes are b4f4e3c0018c15ac794f74924bf4ad12422d9d10.
It shows that these three files are the ones committed on those dates and that they are unchanged since: the bytes you download today hash to the values above, and the repository history shows a single commit for each file. It does not prove that any AI engine will name a business, and it does not prove that our method changes outcomes. It also cannot show that the text came from the engine, because we ran the query ourselves; it shows that our record of the run was not edited afterward.
Published 2026-10-06.
Related: the agency audit pack and why one check a month sees one day.