Specification sheet

MEC-16 Intervention specification

Replace the human-in-the-loop checkbox with a specification of who intervenes, what they can see, what they can prevent or alter, and what happens when they disagree.

Where it fits

Filters and handoffs

Tagged for Governance, Friction with validator handshakes. Includes 3 steps, 3 reusable assets, and 4 snippet blocks.

Scholarly metadata

Authorship

Contact: research@ethotechnics.org

Publication details

  • Published: Dec 3, 2025
  • Last updated: Jan 9, 2026
  • Version: v1.1.0
  • DOI: Pending Zenodo deposit

License: CC BY-SA 4.0

Credit Ethotechnics Institute, include the page title + version, and link to the canonical permalink.

Archive snapshot: Wayback capture

Changelog

  • v1.1.0 · 2026-01-09 — Added citation metadata, mechanisms-level authorship details, and structured usage guidance.
  • v1.0.0 · 2025-12-03 — Initial public mechanisms release.

Copy citation (APA/BibTeX)

Cite this page Formats: APA, MLA, Chicago, BibTeX, RIS

Version

v1.1.0

Last updated

Jan 9, 2026

DOI

Pending Zenodo deposit

APA

Ethotechnics Institute Research Team. (2026). MEC-16 Intervention specification. Ethotechnics Institute. https://ethotechnics.org/mechanisms/patterns/intervention-specification

MLA

Ethotechnics Institute Research Team. "MEC-16 Intervention specification." Ethotechnics Institute, 2026, https://ethotechnics.org/mechanisms/patterns/intervention-specification.

Chicago

Ethotechnics Institute Research Team. "MEC-16 Intervention specification." Ethotechnics Institute. Jan 9, 2026. https://ethotechnics.org/mechanisms/patterns/intervention-specification.

BibTeX

@misc{mechanism_intervention-specification,
  title={MEC-16 Intervention specification},
  author={Ethotechnics Institute Research Team},
  year={2026},
  howpublished={Ethotechnics Institute},
  url={https://ethotechnics.org/mechanisms/patterns/intervention-specification},
  version={v1.1.0}
}

RIS

TY  - WEB
TI  - MEC-16 Intervention specification
AU  - Ethotechnics Institute Research Team
PY  - 2026
UR  - https://ethotechnics.org/mechanisms/patterns/intervention-specification
ER  -

Specification overview

At-a-glance details for planning and handoffs.

Capture the scope, assets, and validator handshakes before diving into the steps.

Spec essentials

What ships with this mechanism

  • 3 steps with checklist-ready owners.
  • 3 reusable assets for briefs and runbooks.
  • 4 reuse-ready snippet blocks.
  • 3 glossary anchors for shared terms.

Filters & validators

Where this spec sits in the system

Tagged in 2 filters with 2 linked validators.

Glossary anchors

Link back to the definitions.

Jump to the glossary terms that frame this mechanism.

Steps

Put the mechanism in motion.

Start with the field cues, then use the assets to keep the work legible.

3 steps with checklist-ready owners, plus linked assets for handoffs.

Assets

Keep outputs reusable.

Link or copy these assets into design docs, runbooks, and briefs so the mechanism travels with the work.

Intervention specification template

Owner, information, preventable actions, alterable states, disagreement path, incentives, cost, and reach time.

Reach-time measurement

Observed time between the trigger and the intervener being in a position to act.

Disagreement log

Cases where the intervener disagreed, what route it took, and what changed.

Reuse-ready snippets

Copy policy, audit, and incident language.

Use these snippets in requirements, audits, and postmortems with the mechanism permalink.

Policy requirement

Policy requirement (MEC-16 Intervention specification)
- Any control described as human oversight is documented as an intervention specification with owner, information available, actions preventable, states alterable, and on-disagreement path.
- The specification records the incentives, the cost of intervening, and the time to reach the decision.
Reference: https://ethotechnics.org/mechanisms/patterns/intervention-specification

Product requirement

Product requirement (MEC-16)
- The interface presents the intervener with the information the specification names, before the action commits.
- Disagreement is a recorded outcome with a route, not an unlogged override.

Audit evidence checklist

Audit evidence checklist (MEC-16)
[ ] Every claimed oversight control has a current intervention specification.
[ ] Reach time and cost of intervening are measured, not asserted.
[ ] Disagreement outcomes are logged with what changed as a result.

Postmortem trigger

Postmortem trigger (MEC-16)
Trigger review when a decision described as human-reviewed proceeded without the reviewer being able to alter it.

Example usage

Specifying review on an automated suspension

A concrete scenario to help teams see how the pieces fit together.

How it plays out

A trust and safety team documents that the on-call reviewer sees the evidence and the account history, can prevent suspension for 30 minutes, can reverse it for 14 days, and that disagreement routes to a policy owner within one business day.

Anti-patterns

Common failure cases and counterfactuals

Use these to avoid superficial compliance and clarify what success requires.

Rubber-stamp review

A human approves in bulk without the information or the time to alter the outcome, and the approval is counted as oversight.

Counterfactual: The specification states what the reviewer sees, what they can change, and how long they have.

False-positive warning: High approval rates are not proof of rubber-stamping; the test is whether refusal was possible and reachable.

Oversight named but not scoped

Policy says a human is in the loop without saying which decisions, which states, or which disagreement path.

Counterfactual: Each oversight claim resolves to a specification covering named decisions.

False-positive warning: One specification can cover a class of decisions if the class is stated.

Validators

Pair validators with this mechanism.

Use these tools to size risk and keep the stewardship path visible.

Validator

Maintenance Simulator

Tabletop simulation that plays through outages, maintenance windows, and handoffs to stress-test coverage.

Validator

Burden Modeler

Quantifies task load, cognitive friction, and risk exposure so you can reroute toil before it burns people out.