Role Exposure Doctrine
- Identifier
- RE-DOC-1.0
- This edition
- 24 August 2026
- Contains
- Laws 01–06
- Supersedes
- none (first public locator)
A change is a new edition, dated, and applies only forward. A prior edition remains readable.
Law 01. Empty is not zero.
An empty cell is untested. It is not zero, not safe, and not exposed.
Law 02. No measurement, no number.
A number is published only from a measured task. Exposure is the share of tasks a test touched. It is not a layoff number. A role-level score is not published until a coverage threshold has been fixed in advance. That threshold is not fixed in this edition. Until then there is no role evaluation and no refusal of a role score as insufficient evidence. A task cell is published on its own evidence and does not wait for that threshold.
Law 03. Independence.
The outcome cannot be bought. A run that fails is published under the same rules as a run that passes. If measurement is ever sold, the customer pays for the process, never for the score. PUBLIC, HOLDOUT and CUSTOM are never combined into one number. Whether paid measurement is offered in a given period is an operational decision, dated separately. It is not this law.
Independence covers three different things and they are handled in three different places. Whether the measurer builds the models it measures is testable from outside and is the first column of the evidence table. Whether an outcome can be bought is a rule of conduct, and it is this law. Whether a reviewer is independent of the result is a property of a run, disclosed as field 11 of the run manifest.
Law 04. No invented model scores.
A named model reaches Models only through a versioned Role Exposure pack run, or a recorded attempt at one. A refusal is a line on Models. It is never a cell on the map and never a score. A refusal is a fact about the candidate, not about the role.
Law 05. Freeze before the run.
Tasks, rubric, fatal conditions, and eligibility are frozen before a run. They are not silently recalculated afterwards.
Law 06. Instruments are not candidates.
An instrument is a grader that produces a verdict. A candidate is the system under test. The two are not the same thing. Role Exposure may use an instrument of its own. An instrument is never drawn from the candidate's model family. An instrument is not sold as a product and is not embedded in an institution's decision-making. Providing an instrument for measurement under a Role Exposure protocol is not a sale.
Surfaces
PUBLIC, HOLDOUT and CUSTOM are three shelves. They are never blended after the fact. A CUSTOM run never enters the map, Models, or a public score. P01-PUBLIC Harbor SAMPLE remains published. It is not recalculated.
From the first CUSTOM run onward, the existence of that run and its date are public even when nothing else about it is: the commissioning party, the data and the result stay closed. Without that rule a reader has no way to tell a project with no customer runs from one that does not mention them.
Results from different surfaces are labelled and never averaged. Where one task carries both a PUBLIC and a HOLDOUT result, both are shown with their surfaces named, and no weighting is applied between them.
How a result is read sits at /method. That reading convention was recorded on 23 August 2026 and is not yet part of this doctrine. It enters only through the amendment rule above, as a new dated edition.