every experiment published in the Raising Agents Lab registry." />
Raising Agents Subscribe →

Home · Workbench · experiment-card-v0

Schema · stable URL · part of the Raising Agents system contract layer. Markdown source on GitHub: system/schema/_src/experiment-card-v0.md.

Experiment Card — schema v0

Status: v0 · 2026-05-25 Purpose: Canonical metadata schema for every experiment published in the Raising Agents Lab registry.

Why this schema exists

Raising Agents is a AI and Innovation lab publishing preregistered behavioral-regression experiments at industrial cadence. The unit of work is the experiment, not the paper. Each experiment is a citable, reproducible artifact with a stable URL, a frozen protocol, a public corpus reference, and an honest verdict — including null and overturned results.

Papers remain as synthesis layers over experiment clusters; experiments are the atomic units underneath.

Card metadata block

Every experiment page opens with the same metadata block. Keys are stable; values are experiment-specific.

ID            EXP-NNN
TYPE          experiment · pilot · confirmatory · null · overturned · heuristic
STATUS        preregistered · running · confirmed · overturned · null · withdrawn
DATE          YYYY-MM-DD (freeze date)
PAPER         Paper N · Study M (if part of a paper)
DOMAIN        refund · claim · memory · code · cross-domain
EVIDENCE      N runs · K domains · J models
PROTOCOL      EXP-NNN_PROTOCOL_FREEZE.json (or PAPER_PROTOCOL_FREEZE.json#study-M)
CORPUS        OPBR-Bench / synthetic / redacted-production
RESULT        headline number with CI or scope
REPLICATION   abw run --experiment EXP-NNN
LICENSE       CC BY 4.0
AUTHOR        Adrian Sanchez de la Sierra, Genesis Rojas, and Radu Simonescu

Status taxonomy

Status Meaning
preregistered Protocol frozen, runs not yet executed
running Execution in flight
confirmed Hypothesis supported by data, claim boundary clearly stated
overturned Hypothesis contradicted by data, prior claims retracted publicly
null No significant effect; published anyway because nulls are evidence
withdrawn Protocol violation or methodological flaw discovered post-hoc; full disclosure published
heuristic Non-empirical analysis (e.g. completeness analyzer) that supports an experiment cluster

Type taxonomy

Type Meaning
experiment Primary, preregistered, hypothesis-driven
pilot Small-N exploratory; informs preregistration of larger experiment
confirmatory Replicates a prior experiment under tightened conditions
null Reports absence of effect
overturned Reports refutation of prior published claim
heuristic Analytical artifact attached to an experiment (e.g. specification audit)

Required sections on the experiment page

  1. Metadata block (the table above)
  2. One-sentence headline — what changed and what we found
  3. Preregistered hypothesis — verbatim from protocol freeze
  4. Method — what was varied, what was held constant, exclusions
  5. Result — primary metric, baseline comparison, CI, McNemar / Wilson / SPRT where relevant
  6. Claim boundary — what this experiment does not establish
  7. Replication — one CLI command + corpus reference
  8. Related — paper section, prior experiments, follow-ups

Citation form

Sanchez de la Sierra, A. (YYYY). EXP-NNN — Short Title. Raising Agents Lab.
https://raisingagents.is/lab/exp/exp-NNN/

Ledger

The registry index at /lab/ lists every experiment in reverse-chron order with status badge, headline number, evidence count, and link. Counter strip aggregates: total experiments, total runs, domains covered.

Nulls and overturned results are not hidden from the index. They are listed with full badges. This is the brand.