Mechanism MEC-29

Reconstructable delegation

Require a decision record that can answer what would have changed the outcome, or lower the system's autonomy until one exists.

  1. 1 Evidence
  2. 2 Authority
  3. 3 Decision acts here
  4. 4 Consequence
  5. 5 Challenge
  6. 6 Reconsideration
  7. 7 Correction acts here

Where it acts

At Decision and Correction, two of the seven stages every consequential decision passes through. It tests whether a decision can be reconstructed before the institution claims it can correct the rule.

Steps

Put the mechanism in motion

4 steps, in order. Check each one off as it is done; the records show the work was done.

Records this mechanism keeps

What a team running this mechanism keeps, and what each record is for.

Reconstruction record

The fields that could have changed the outcome and the counterfactual answer.

Autonomy decision log

For each failed test: lower autonomy, narrow scope, or stop, with the record behind it.

Counterfactual test sample

Cases reviewed for what would have produced a different outcome.

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-29 Reconstructable delegation)
- Each consequential decision class has a reconstruction record that answers what would have changed the outcome.
- A failed reconstruction test triggers a recorded autonomy, scope, or stop decision.
Reference: https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation

Product requirement

Product requirement (MEC-29)
- The decision record carries the fields needed for a counterfactual review, not only the output and timestamp.
- A failed reconstruction test is visible as a deployment state, not a support ticket.

Audit evidence checklist

Audit evidence checklist (MEC-29)
[ ] Each decision class has a reconstruction record.
[ ] A sample was tested for what would have changed the outcome.
[ ] A failed test links to an autonomy, scope, or stop decision.
[ ] The record is available to the affected party or their advocate.

Postmortem trigger

Postmortem trigger (MEC-29)
Trigger review when a challenge cannot be reconstructed from the record, or when the response to a failed test is another guardrail without an autonomy decision.

Example usage

A multi-agent claims chain that cannot explain a denial

One scenario with each part of the mechanism in place.

How it plays out

A claims chain uses three agents, dynamic tool calls, and a stochastic planner. The record shows the calls and model versions but cannot answer what would have changed the denial. The institution lowers the chain to recommendation-only until the record can answer the counterfactual, rather than adding another guardrail and calling it correction.

Anti-patterns

Common failure cases and counterfactuals

Each case passes on paper and fails in operation. The counterfactual shows what would have worked.

Explanation as reconstruction

A post-hoc narrative is treated as the rule, so the institution can describe the decision without being able to say what would have changed it.

Counterfactual: A reviewer tests the record against a counterfactual and records the answer.

Exception: A fluent explanation is not evidence that the decision is reconstructable.

Guardrail stacking

Each failed reconstruction adds another filter or prompt instruction, while the underlying delegation keeps the same autonomy.

Counterfactual: A failed test triggers a recorded autonomy, scope, or stop decision.

Exception: A new guardrail is not a correction if the rule remains unreconstructable.

Diagnostics for this mechanism

Tools that size the risk this mechanism addresses.

Diagnostic

Delegation audit

Walks a team through one workflow against six questions and returns an exposure score, the permissions nobody can justify, and whether its decisions can be undone. Start here when the system keeps no decision records yet; use the record conformance checker once it does.

Cite this page · v1.2.1 · updated Oct 6, 2026 · CC BY-SA 4.0

APA

Ethotechnics Institute. (2026). Reconstructable delegation (MEC-29). Ethotechnics Institute. https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation

MLA

Ethotechnics Institute. "Reconstructable delegation (MEC-29)." Ethotechnics Institute, 2026, https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation.

Chicago

Ethotechnics Institute. "Reconstructable delegation (MEC-29)." Ethotechnics Institute. Oct 6, 2026. https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation.

BibTeX

@misc{mechanism_reconstructable-delegation,
  title={Reconstructable delegation (MEC-29)},
  author={Ethotechnics Institute},
  year={2026},
  howpublished={Ethotechnics Institute},
  url={https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation},
  version={v1.2.1}
}

RIS

TY  - WEB
TI  - Reconstructable delegation (MEC-29)
AU  - Ethotechnics Institute
PY  - 2026
UR  - https://ethotechnics.org/mechanisms/patterns/reconstructable-delegation
ER  -

Authorship

Contact: hello@ethotechnics.org

Publication details

  • Published: Dec 3, 2025
  • Last updated: Oct 6, 2026
  • Version: v1.2.1

License: CC BY-SA 4.0

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

Archive: Wayback Machine history

Latest change

v1.2.1 · Oct 6, 2026 — Rewrote the copy of MEC-01 to MEC-12 in plain sentences. A halt is described as containment while the rule is fixed and people are restored, upheld appeals are counted against the rule in MEC-06 and MEC-08, and MEC-04 is renamed Hard clock.

Earlier versions (3)
  • v1.2.0 · Oct 1, 2026 — Filed every mechanism under the six safeguards, led the starters and role paths with challenges that reach the rule, and added that obligation to MEC-06.
  • v1.1.0 · Jan 9, 2026 — Added citation metadata, mechanisms-level authorship details, and structured usage guidance.
  • v1.0.0 · Dec 3, 2025 — Initial public mechanisms release.