Base rate of prior art

Base rate of prior art

A book on this site claims that generated research hypotheses are usually already known. That claim rested on eight claims chosen for strength — a purposive sample, which cannot support a rate. This page fixes that, or begins to.

Method

Sixteen claims were drawn from the strongest entries of a 6,000-hypothesis corpus. Each was searched against a general web index, first page of results. A claim counts as prior art if a source was found that had already stated or established it. No claim returned nothing.

This is still not a random sample. The sixteen were the strongest claims, and strong claims are the ones most likely to correspond to something real — and therefore most likely to be already published. That makes sixteen of sixteen a ceiling on prior-art risk, not a base rate. The honest estimate requires a random draw.

Results

Claim Prior art found
Three-axis machine has 21 independent geometric error terms published table of exactly that
Nuclear fuel ridging pitch tracks pellet length 1974 study on pellet length and cladding interaction
Radial diffusion coefficients from phase-space gradients are non-unique active literature on that extraction
No interlaboratory comparison detects a bias common to all proficiency-test consensus literature
Dim light damages more per unit exposure microfading study of the reciprocity principle
Electro-osmotic flow reverses by excluded volume reversal in charged nanopores, multivalent electrolyte
Standard detection limits understate the power requirement Currie-type detection-limit literature
Peak width biases decomposition when the instrument is ignored routine in spectroscopy
Kapitza resistance and bolted-joint acoustic mismatch are one problem thermal boundary conductance literature
Emulsion inversion is kinetic, not thermodynamic stirring-energy dependent inversion hysteresis
Rim compliance inflates measured tyre vertical stiffness free-oscillation tyre test literature
A gauge can out-stable the standard used to calibrate it metrology hierarchy literature
A derived ratio column inflates correlation ratio-variable spurious correlation, 1976
Even harmonics identify a quadratic nonlinearity nonlinearity-index literature; contact nonlinearity
Detection limit is a critical level, not a detection threshold Currie and later detection-limit work
Pooled magnitude-frequency statistics hide scale mixing pitfalls in magnitude-frequency distributions

16 of 16. Wilson 95% interval for a proportion of 1.0 at n=16: [0.81, 1.00]. A random sample of one hundred would narrow this materially; sixteen sets a floor around 0.80 and no more.

What this establishes

For confident, well-formed claims, prior art is near-certain. Every claim specific enough to be checkable was already checkable by someone else, and someone else had checked it. The practical consequence is the one the book gives: search before generating, because the good ideas are the ones most likely to be taken.

What it does not establish: a rate for weak or vague claims, which may be genuinely unaddressed because they are unaddressable; or for claims in narrow technical niches where a general web index does not reach.

Next step, stated so it can be executed

Draw one hundred claims at random — not the strongest — from the corpus. Search each with the same protocol. Report the proportion and interval. Two days of work, and it converts this ceiling into a measurement.

16 claims · searched 4 November 2026 · general web index, first page only · no specialist databases.

Random draw, batch 1

The purposive sample above proves nothing about the base rate. So a random sample was drawn — seed fixed at 20261104, four claims from the 6,000-entry corpus, no selection by strength.

Claim Prior art found Verdict
H5576 — a frame’s buffering capacity belongs to the enclosed materials, not the seal Getty conservation literature; Hackney warned in 1987 against enclosing silica gel in small sealed volumes known
H2413 — quoting slope and height tolerance together is over-specification “Specifying More Than Peak to Valley”, Optimax technical note known
H3246 — bonding in a drying sheet is limited by contacts surviving shrinkage free versus restrained dried paper, Cellulose (2020) known
H3638 — a check-valve dashpot delays closure into reversed flow valve patent literature known

Random draw: 4 of 4. Combined with the purposive sample, 20 of 20 checked claims had prior art.

What changes with the random draw. The purposive sixteen could not support a rate. Four random claims cannot either — but they are evidence in the right direction, and crucially they were not cherry-picked: the four came out of the corpus unedited and every one was already published. If this holds at one hundred, the book’s claim moves from a strong prior to a measurement.

Protocol, so the rest can be done

  1. Draw claims at random from the corpus with a fixed seed; record the seed and the draw.
  2. Search each claim’s mechanism, not its topic — the sentence, not the field.
  3. Score binary: prior art found, or not found. Do not grade closeness.
  4. Search the first page only, one index, and say so. Note that absence is not evidence of novelty.
  5. Report the proportion with a Wilson interval, and report which claims came back empty, since those are the interesting ones.

Batch 1 drawn 4 November 2026 · seed 20261104 · four of four.