Skip to content
Foundable
Foundable Research

Research you can check, published before the result exists

Four protocols, each registered before it runs. We publish the method, the frozen inputs and the empty result table first, then the numbers on a stated date — including the ones that do not flatter us. Where another company's figures appear, they are attributed and dated as prior work.

Published Updated Published by The Foundable team

Overview

Everything here follows the same four rules. Prompts and queries are published before the first request. Fixtures are frozen and one request is made per cell, with no retries and no selective reruns. Scoring is done twice and adjudicated, with reviewer affiliation stated plainly. Counts never appear without their denominator, and corrections are versioned in public rather than edited away.

Browse

Registered protocol · results 6 Oct 2026What AI money answers leave outThirteen prompts, two models, 26 cells, seven pass-or-miss criteria. The prompt list is published; the result table is empty until the window closes.Open Registered protocol · capture 14 Sep 2026What search surfaces for AI-builder queriesTwenty-seven predeclared queries in three intent groups, one capture window, no weighting. Our own visibility is reported at the same rigour as anyone else's.Open Registered protocol · vectors 22 Sep 2026A permission should name one exact actionThree grant states, four execution attempts, six published decision vectors. Synthetic and local: nothing sends, saves or spends.Open Registered protocol · run window 29 Sep 2026Measuring the path from idea to signalTwenty-five synthetic scenarios, ten unscored dimensions, eight failure criteria written before the run and a dated window we can be held to.Open

Quick answers

Why publish a protocol before you have any results?

Because a method chosen after seeing the data is not a method. Publishing prompts, denominators and failure criteria first means the result cannot be a table we drew to fit an outcome. It also means a missed date is visible, which is the cost of doing it this way.

Are the numbers on these pages yours?

Not yet. Every figure currently on a research page belongs to earlier published work by Foundable, attributed with its publication date, and is shown as prior work we are replicating. Our own result cells read pending until the stated publication dates in October 2026.

What makes a count readable?

Its denominator, printed beside it. Nineteen of twenty-two is a finding; nineteen is a number. Denominators shift within a single study — response states, eligible answers, completed overviews, retained attachments — and mixing them is the most common way an honest study becomes a misleading one.

How do corrections work?

Write to us through the contact page. Typographical corrections are recorded with the date and the before-and-after text on the page itself. A change of meaning creates a new protocol version rather than an edit, and the earlier version stays available.

Does any of this predict what a business will earn?

No. These are diagnostic instruments about answers, search surfaces, agent authority and controlled runs. None of them tests whether a method produces income, and nothing here should be read as a forecast of earnings.

How to read this hub

  • Three of the four protocols on this hub have produced no data. Their result tables are published empty on purpose.
  • We compete in the categories we measure, so treat the framing as interested and check the denominators rather than the summaries.
  • Prior work by other companies is quoted with its publication date and never restated as our measurement.
  • Raw provider outputs stay private. What we publish is counts, denominators, frozen inputs and the record of scoring disagreements.
  • None of these studies has been reviewed by an unaffiliated party. When one is, the findings get published unedited.

Frequently asked questions

Yes, and that is the point of publishing the query lists and prompt sets in full. If you rerun one and get a different result, send it through the contact page and we will publish the discrepancy rather than resolve it quietly.

Start free with Ted.

Bring the messy version of the idea. Ted turns it into a first move, something real, and a way to see what people actually do — free to start, no card.

Start freeSee pricing

AI output needs your review. No income is promised or guaranteed.