Test Locked Prompt Set AI Visibility Versions Live

By William Zhu · Cofounder, InfiniSynapse · Published: 2026-09-08 · Last updated: 2026-09-14 · Last verified: 2026-09-14 · Reviewed by the InfiniSynapse Data Team.

Reviewed by: InfiniSynapse Data Team · method review 2026-09-14. Two local prompt-set hashes. One verb edit. No engine replay of the set.

Trust / COI: About · Corrections. SEO Health is a commercial product from this team. The hashes are our fixture files, not a vendor export.

TL;DR

Changing one verb in an eight-prompt file flipped the SHA-256 and added 12 bytes. That is the locked prompt set AI visibility finding.

Direct answer: Treat locked prompt set AI visibility as a versioned file, not as a dashboard tile. On 2026-09-14 we wrote prompt-set-v1.json (818 bytes, 8 prompts, SHA-256 61f058ff9c9296ff45e8013eff16f4a2299e34358d63c4326f5512fbca140ba4). v2 changed prompt[0] from “suit” to “are suitable for” (830 bytes, SHA-256 c8d58aa194995ba366db1679010f5d8288ee53c2cf6f165f52f11d30bf09d181). Prompt count stayed 8. No engine ran either file.

A locked prompt set AI visibility week that silently edits a qualifier is a new series. Keep v1. Do not overwrite it.

What you will learn

• Which bytes and hashes we recorded
• Why prompt count can stay 8 while the file is new
• How to replay sha256
• When SemVer and Git objects are analogies, not this file
• Where siblings own cards, hops, and aliases

What a locked file records

Key definition: A locked prompt set AI visibility manifest is a UTF-8 file with a stable id, a locale, and an ordered prompt list. The hash is of the bytes, not of a human paraphrase.

People open locked prompt set AI visibility after a weekly report moved one adjective and the mention rate “improved.” The rate is incomparable. Hash first.

RFC 8259 is the JSON grammar we used. Pretty-printed JSON with a trailing newline is the byte sequence we hashed. A minified twin is a different object.

SEO Health is the deterministic evidence foundation for file and CLI reads. A later InfiniSynapse Server API long-task workspace can review a submitted manifest through SSE /api/ai/events; the API does not freeze your verbs for you.

Change control is the product

NIST SSDF is why we treat a wording edit as a new version. Semantic Versioning is an analogy for v1 versus v2. This fixture is not a public SemVer release.

Git objects explain content-addressed storage. We hashed the files with SHA-256, not Git’s blob header. locked prompt set AI visibility notes that quote a Git commit without the JSON bytes are a different provenance.

Eight prompts is a workshop size

The list covers category, comparison, alternatives, entity, surfaces, Gemini, Claude, and locale. It is not a market sample. locked prompt set AI visibility that treats n=8 as national coverage is already over-claiming.

Two hashes from this desk

Delta: prompt[0] suit → are suitable for. Hashes differ. Prompt count does not.

Figure: Bytes moved. Prompt count did not. That is a new series.

FIPS 180-4 is the hash standard we used as vocabulary (HTTP 200; title “FIPS 180-4, Secure Hash Standard (SHS)”). RFC 2119 is why MUST and SHOULD in a later contract are words, not feelings.

W3C Data on the Web Best Practices and DCAT 3 describe dataset publishing. This fixture is a desk file, not a catalog harvest. Library of Congress preservation is a preservation root. It does not run the prompts.

How to reproduce the SHA-256 pair

python3 - <<'PY'
import hashlib, json
from pathlib import Path
for name in ("prompt-set-v1.json", "prompt-set-v2.json"):
    raw = Path(name).read_bytes()
    print(name, len(raw), hashlib.sha256(raw).hexdigest())
PY

Expect 818 / 830 and the two digests if the files are byte-identical to this desk. Line endings matter. locked prompt set AI visibility reproduction that hashes a “same words, different quotes” file will get a third digest. Keep that third file as v3.

What we did not run

No Perplexity, Gemini, Claude, or Mode replay of the eight lines. aivisibility on a sibling printed unauthorized. Package infinitegrowth@0.1.1; -V printed 0.1.0. Credits were not spent to “execute” the set.

The sample size is two files. That cannot establish engine volatility, and it does not prove a verb change moved a mention. As of 2026 this page is not a customer weekly export.

What a later week still needs

Freeze vN. Hash it. Run engines only against that hash. If a prompt must change, cut vN+1. Keep both files. Attach the hash to every coded row. locked prompt set AI visibility fails when a live edit lands in the middle of a week.

Locale is inside the file

v1 and v2 declare "locale": "en". A zh-Hans twin is a different manifest, even if the English gloss looks similar. Do not mix them in one average.

Negative space

Keep prompts you expect to return no mention. Deleting them after a quiet week is another silent edit.

Failure modes that break comparability

Overwrite in place. v1 disappears.

Prompt count as identity. Both versions have 8.

Pretty-print drift. Spaces change the hash.

Git commit without the JSON. Wrong object.

Unauthorized as a set failure. Login failed.

Synthetic version history. Older drafts invented transitions. They are gone.

A locked prompt set AI visibility memo that skips the two digests is already late.

How a reviewer should read the two files

Open v1. Confirm 818 bytes and the 61f058ff… digest on your machine. If the digest differs, you do not have this desk’s file. Pretty-print drift, a missing trailing newline, or “smart quotes” will do that. locked prompt set AI visibility that shrugs at a third digest and still calls the file v1 is already breaking the series.

Open v2. Confirm 830 bytes and c8d58aa1…. Read prompt[0]. The only intended change is suit → are suitable for. If you see another edit, you are holding v3. Give it a new id. Do not pretend it is v2.

Prompt count 8 is a coincidence, not an identity. Anyone can keep eight lines and rewrite all of them. The hash is the identity. Write the hash on every coded engine row that claims to use the set.

Locale sits inside the JSON. These files say "locale": "en". A Chinese twin with eight “equivalent” prompts is another manifest. locked prompt set AI visibility that averages en and zh-Hans because the gloss looks similar is mixing series.

If a weekly meeting wants a new qualifier, cut vN+1 before anyone runs engines. Keep vN on disk. Attach both hashes to the change ticket. Overwrite-in-place is the failure that made this page necessary.

Git can store the files. Git’s blob hash is not SHA-256 of the pretty-printed JSON unless you compute that yourself. Write which hash you used. FIPS 180-4 is the vocabulary, not a claim that we ran a federal validation lab.

If you only have the two files, you have the desk. That is enough to reject a week-over-week lift that cannot show a hash. It is not enough to claim engines moved. Write both. A later engine run is an addendum that names the hash it used. Do not put invented version transitions back into the main table.

Two people hashing the published JSON should get the same digest. Two people retyping the prompts from memory will not. locked prompt set AI visibility forbids retyping. Copy the file.

Negative prompts stay in the set. Deleting a quiet line after a dull week is another silent edit. Keep the zero.

What stays on disk after this desk

Keep prompt-set-v1.json, prompt-set-v2.json, and a hashes.txt that repeats both SHA-256 values. That is the 2026-09-14 locked prompt set AI visibility packet. Engine outputs, if they ever exist, must name one of those hashes in the first line.

Publish the JSON bytes, not a screenshot of the prompts. Screenshots cannot be hashed the same way. locked prompt set AI visibility that archives only a PNG will lose the series on the next pretty-print.

If someone retypes a prompt from memory, that file is v3 even when it “looks the same.” Run SHA-256. If it differs, give it a new id. Do not argue about synonyms.

Weekly process: freeze, hash, run, code, stop. Edits happen between weeks, not during them. A mid-week hotfix to one adjective is a new series that should not be averaged into the old one. Write that on the calendar.

Locale twins live beside the English files, never inside them. A zh-Hans v1 is prompt-set-zh-Hans-v1.json with its own hash. locked prompt set AI visibility forbids concatenating locales to make n look larger.

Negative prompts stay. If a week is quiet, keep the quiet lines. Deleting them is an edit. The 12-byte gap between v1 and v2 is the teaching object; the next gap may be larger. Record it.

The decision this page supports is comparability. It does not decide which engines to run. When someone shows a lift without a hash, reject the slide. Empty engine folders are allowed. A hash-less lift is not. That stop is locked prompt set AI visibility.

Package infinitegrowth@0.1.1 and -V 0.1.0 go in versions.txt. Unauthorized on a sibling does not hash these files. Hashing is local and free.

Hand the files, not the story

If a vendor asks for “your prompt set,” send the JSON bytes and the hash. Do not send a slide of eight paraphrases. locked prompt set AI visibility dies the moment the bytes leave the room without their digest.

If two agencies run “the same set” and their hashes differ, they did not run the same set. Stop the comparison. Re-copy the file. Then restart. A meeting that continues anyway is no longer this method.

Store hashes.txt in the same directory as the JSON. People lose the text file and then re-hash a prettified copy. The new digest is v3. Give it a name. Do not hide it.

A calendar reminder to re-hash after every wording pass is enough process. Fancy workflow tools are optional. The 12-byte gap is the lesson. locked prompt set AI visibility that needs a novel to explain a verb change has already lost the plot. Keep the two files and the two lines of hex.

Nearby guides

The Perplexity hub owns vendor HTTP. This page owns the file bytes.

The AI visibility hub remains the category page. locked prompt set AI visibility stays on versions.

Frequently Asked Questions

Did we run the eight prompts in any engine?

No. locked prompt set AI visibility on this desk is two hashes.

Does a matching prompt count mean the weeks are comparable?

No. v1 and v2 both have 8 prompts and different digests.

Must we use SHA-256?

This desk did. Another algorithm is fine if you write it down. Do not mix algorithms inside one series.

Is a Git commit enough provenance?

Only if it points at the exact JSON bytes. A commit message is not the hash of the manifest.

May we publish engine rates from this chart?

No. The chart is bytes versus prompt count. It cannot establish visibility.

Conclusion

Use locked prompt set AI visibility when the job is to refuse a week-over-week tile until the file bytes can be hashed. This 2026-09-14 pass recorded 818 versus 830 bytes and two different SHA-256 values after one verb edit. Stop when those two digests remain visible. A bounded file pair beats an unsupported weekly lift.

Re-hash the file after every wording edit. The 12-byte gap between v1 and v2 is the whole point of locked prompt set AI visibility on this page. If a third digest appears, it is v3. Name it. Do not debate whether the words “feel” the same. Bytes decide. Meetings that continue without a matching hash are no longer this method. locked prompt set AI visibility ends at the hex.