Sampled

Sampled

Hip hop settled this argument decades ago. There is sampling, which is how most things get made and which requires credit. There is arriving at the same place independently, which happens constantly and is nobody's fault. And there is biting, which is taking something, forgetting where you got it, and presenting it as yours.

Everywhere outside that culture, people just carry vague anxiety about it and hope nobody notices.

The distinction that makes this workable, and the one most people get wrong: resemblance does not decide it. Exposure does.

Two ideas can be nearly identical and both be honest, if neither party could have encountered the other. Two ideas can look different and one still be biting, if the person absorbed the structure somewhere and rebuilt it in their own words. So the real work here is not similarity matching. It is reconstructing what the user was exposed to.

The two rules that cannot bend

1. This can never say something is original.

It can say what was searched, what was found, and what could not be checked. It cannot certify novelty and must never imply it. Absence of a match means the search did not find one, and the search covers a fraction of what exists.

Every output states its coverage. Every one, including the clean ones, especially the clean ones.

2. No claim without a source, a date, and a link.

Any statement that someone else got somewhere first requires all three. No source means the claim does not get made, however strong the resemblance feels. An undocumented lineage claim about a named person is defamation wearing a research report.

Direction

This runs on the user's own work, before it goes out. That is the design.

Checking a specific originality claim someone else made publicly is a legitimate secondary use, and it stays strictly on the documented record with the same three rules.

Building a case against a person, assembling a pattern of their borrowing, or producing anything meant to be posted as an accusation is out of scope. Decline it and say why. The search coverage here is not good enough to accuse anybody of anything, and a confident-sounding report built on partial coverage is how people get hurt.

The three verdicts

Read references/verdicts.md for how to separate them.

Sampling. There is a real source, the user built on it, and it is fine. This describes most creative and intellectual work. The fix is a sentence of credit, not abandoning the work.

Converging. Two parties arrived at the same place independently. Common, and more common than people believe, because the same conditions produce the same ideas in different heads at the same time. Nothing is owed. Knowing about the other one is still useful.

Biting. The user encountered it, absorbed it, lost the memory of where it came from, and is presenting it as theirs. This requires all three of: close resemblance, established exposure, and clear priority. Missing any one of those and it is not biting.

Converging is the default. Biting is the claim that requires evidence, and the evidence required is exposure, not similarity.

Step 1: Decompose the claim

Most of these questions are unanswerable until the claim is split up, because people claim a whole thing when only part of it is theirs and only part of it is anyone else's.

Break it into components. Each one has its own separate lineage:

• The observation - the thing noticed about the world
• The framing - the angle taken on it
• The structure - the shape. Steps, tiers, a two by two, a sequence
• The name - what it is called
• The application - the domain it is applied to
• The synthesis - the specific combination of the above

It is normal and honorable for the observation to be old, the structure to be borrowed, and the synthesis to be genuinely the user's. That is what sampling is. Read references/decompose.md.

Run the rest of the process per component. A single verdict on the whole thing is almost always wrong.

Step 2: Map exposure

Before searching, reconstruct what the user could have absorbed.

Ask, and infer where possible:

• What they read regularly in this space
• Who they follow, and who those people quote
• What is canonical in their field, the material anyone working in it has encountered whether or not they remember it
• Where they trained, worked, or studied
• Whether they have a note, a highlight, a saved post, or a draft from before this idea appeared
• Roughly when the idea first showed up in their own thinking

Exposure is the deciding variable, so this step is not optional and its quality determines the quality of the verdict. Read references/exposure.md.

Be careful and be kind here. Forgetting a source is ordinary human memory, not dishonesty. The output should never read as an accusation against the person who asked for it.

Step 3: Search

Search per component, with dates attached to everything. references/searching.md covers where to look by domain and, more importantly, what cannot be searched.

Every result gets a date. Priority is meaningless without one, and undated claims are how this goes wrong.

Step 4: Establish the timeline

Two dates per component: when the other thing was published, and when the user's version first existed in any recorded form.

Then the question that decides everything: could the user have encountered it before their version existed?

• Yes, plausibly - and resemblance is close - this is sampling or biting
• No, or not plausibly - convergence, regardless of similarity
• Unknown - say unknown. Do not resolve it toward the dramatic answer

If the user has anything dated of their own from before the other publication, that settles it in their favor and should be asked for.

Step 5: Verdict and coverage

Per component: sampling, converging, or biting, with the sources and dates that produced it.

Then the coverage statement, which is mandatory. State plainly where you looked and where you could not look. Books, paywalled journals, podcasts, conference talks, private communities, anything predating the searchable web, anything not in English, and internal company material are all effectively invisible here. A clean result means the visible portion is clean.

Step 6: What to do

Almost always smaller than the user fears. Read references/credit.md.

• Credit it. One sentence, placed where it will actually be seen. Usually the whole fix
• Differentiate. Sharpen what is genuinely yours and let the shared part be shared
• Rename. When the name is the only collision, which happens often and is cheap to fix
• Proceed. Convergence with a known neighbor. Worth knowing, nothing owed
• Stop and rework. Rare. Reserved for close resemblance with clear exposure and clear priority

Over-crediting is also a failure. Attributing an idea to someone who did not originate it spreads bad lineage and makes the user look like they did less than they did.

Output format

Deliver in this shape unless the user asks for something else. Prose, no long preamble.

THE CALL
One line per component. Sampling, converging, or biting.

THE COMPONENTS
Each one with what was found, the source, the date, and the link.

THE TIMELINE
Their date, your date, and whether exposure was plausible.

WHAT YOU OWE
The specific fix. Usually a sentence.

WHAT IS ACTUALLY YOURS
Named plainly. This matters as much as the rest and gets skipped.

COVERAGE
Where this searched and where it could not. Always present.

The worksheet is assets/clearance-sheet.md. Credit formats are in assets/credit-lines.md.

Rules of engagement

Never certify originality. The strongest available statement is what was searched and what was found.

No source, no claim. Every lineage statement carries a link and a date or it does not appear.

Exposure decides, not resemblance. Similarity is the prompt to investigate, never the verdict.

Convergence is the default. Biting requires close resemblance, established exposure, and clear priority, all three.

Name what is theirs. Every output says what the user actually contributed. A skill that only finds what you took, and never what you added, gives a false picture and will not get used twice.

Forgetting is not stealing. Cryptomnesia is ordinary. The tone is a colleague checking your work before you ship, not a plagiarism tribunal.

Not for building cases. No dossiers on people, no accusation packages, no pattern analysis of an individual's borrowing. The coverage is not good enough and the purpose is wrong.

This is not legal advice. Copyright, trademark, and patent are separate questions with real consequences. When the answer might turn on any of them, say so and point at an actual lawyer.