Methodology v0.1

A preregistration scaffold for repeatable research observations and the calculation contract behind the browser-local Lab. The provider-collection protocol is not frozen and no provider collection has started. A dated synthetic example makes the current local calculation contract reproducible.

  1. 01

    Observation unit

    One user-supplied public answer, its visible cited or linked sources, the public surface and configuration, locale, and timestamp.

  2. 02

    Sampling frame

    Question classes and a stable control set must be fixed before provider collection starts.

  3. 03

    Measures

    Visible source-domain recurrence, mean pairwise token Jaccard overlap, literal phrase coverage, and missingness—with explicit denominators and limitations.

  4. 04

    Comparability breaks

    Prompt, route, model, interface, policy, locale, or sampling changes must create a new series or a visible annotation.

  5. 05

    Redistribution

    Only material permitted by provider terms and applicable rights may be retained or published.

  6. 06

    Corrections

    Every correction records a version, affected records, reason, reviewer status, and effect on comparability.

Observation schema

labels only

{
  observation_id:
  observed_at:
  question:
  surface:
  surface_version:
  locale:
  question_id:
  answer_text:
  response_hash:
  source_urls:
  coverage_criteria:
  missing_reason:
  protocol_version:
}
rev. 0.1
07

Launch gate tracker

  1. 01Fixed sampling protocolopen
  2. 02Named research ownermet
  3. 03Current tool cost ceilingmet
  4. 04Redistribution reviewopen
  5. 05Comparable provider batches0 / 3
  6. 06Inspectable provider evidence packageopen

The public tools and reference pages are indexable. Open gates apply to future provider-output research and benchmark claims—not to the browser-local website.

Inspect research-publication gates