The House, measured.
W-000 · Touchstone wave · Wave Zero Preregistered
Rebel.Rebel run through its own instrument, under the rules that govern any wave. The embarrassing number is included.
Touchstone™ · the public market read
Touchstone is Rebel.Rebel’s independent research program on machine-mediated representation.
Private Assays tell a company where it stands. Touchstone studies whether the same phenomena travel across a deliberately constructed public cohort.
Published, not sold.
Why it exists
A method cannot earn trust by being right only in private.
Touchstone is designed to make the discipline inspectable: the cohort, decision space and hypotheses are declared before the wave; findings, disagreement and misses are reported after.
Two headline scores
Retrieval
Does the subject survive into the relevant answer?
Fidelity
Is the rationale specific and attributable?
Two supporting reads
Presence
Is the subject named at all? Reported, never ranked, and never confused with Retrieval.
Provenance
Whose record appears to supply the evidence?
Independence
A place in the table cannot be bought. Neither can an absence from it.
The cohort is constructed for what it can teach the method, not for who is paying. Brands in a Touchstone set may be clients. Where one is, it is named on the cover.
No brand pays for inclusion, placement, exclusion or review, and no brand sees its result before publication. Editorial exclusivity within a category is purchasable. A place in the table, or an absence from it, is not.
What the House learned from an engagement is the House’s. What the House measured in an engagement is the client’s.
Method discipline
Hypotheses before. Findings and misses after.
A published miss carries the same weight as a published finding.
The waves
Whose words does the machine use to sell a car?
W-001 · Touchstone wave · US automotive Preregistered
Ten brands, none a current client. Scored against a frozen control set and read by three independent readers.
W-000 is reserved permanently for Wave Zero. References are never reused. A withdrawn wave keeps its number and states its withdrawal.
W-001 · the cohort
Subject
US automotive.
Cohort
Chevrolet, Ford, Honda, Hyundai, Jeep, Mazda, Subaru, Toyota, Volkswagen, Volvo. None is a current client.
Cadence
Quarterly. The battery is identical across issues.
Reading
Three independent readers. Reference lexicons are presented unlabelled. Disagreement rates publish with the wave.
Status
Preregistered. No findings are claimed before the wave publishes.
What publishes, what stays sealed
Legible, not reproducible.
Published
The cohort and decision space. The measures. The machines and their versions. The preregistered hypotheses. Reader disagreement rates. Findings, limits and misses.
Sealed
Question files and their construction. Answer logs. Reference lexicons. Scoring arithmetic and weighting. Anything that would let the instrument be manufactured or aimed.
The ruler does not move because the answer disappoints us.
Limits, stated with the findings
Aggregation
Aggregates mask variation between individual machines.
No control arm
A score change cannot be attributed to a single cause.
Authorship
Fidelity measures whether published language was used. It does not verify who authored that language first.
Follow the program