Research question

Post-hoc explanation and decision consistency

Abstract: When a model explains a decision after making it, does the explanation match the evidence that was available to justify the decision? This entry defines the question without any claim about hidden internal reasoning. It reports no results.

Research question. A problem worth investigating. No result is implied. About publication states

Research question

Does an explanation given after a decision accurately reflect the evidence that was available when the decision was made, and does it stay consistent when the same decision is reproduced under the same conditions?

Scope

This work compares two observable things: the evidence present when a decision was made, and the explanation offered afterwards. It makes no claim about a model’s internal cognition. An explanation is treated as an output to be checked, not as a window into the process that produced the decision.

What would count as evidence

The original outputs preserved verbatim, with the candidate-selection conditions recorded.
Explanations elicited under controlled prompts, separately from the decision itself.
Repeated decisions under identical conditions, so that inconsistency can be distinguished from noise.

Planned next step

Specify how decisions, conditions and explanations are captured and compared, and publish that as a protocol.

Hypothesis

Unavailable No formal hypothesis is registered yet.

Method

Unavailable The protocol has not been published.

Results

Unavailable No experiment has been run. No result is implied.

Revision history

2026-10-15Entry created as a research question.

Cite this entry

Sai Bharath Siliveri (2026). Post-hoc explanation and decision consistency. ALGEDI SI Research Labs, ALG-003, research question, revision of 2026-10-15. https://algedi.si/research/post-hoc-explanation-consistency/