Before Saying Done: Checking the Outcome Behind a Receipt
An external service can acknowledge a request without establishing that the intended outcome occurred. For an agent asked to complete a task, the difference matters: a response describes an interaction with a service; completion describes a state of the world.
Acknowledgment and outcome
An external service can acknowledge a request without establishing that the intended outcome occurred. For an agent asked to complete a task, the difference matters: a response describes an interaction with a service; completion describes a state of the world.
The prepared method
ZEISMO’s Before Saying Done preparation examines this distinction through synthetic task records and a proposed evaluation schedule. The current package contains 160 synthetic benchmark records and a planned 720-trial evaluation. It includes parsing, scoring and analysis tools, together with blank review materials. These quantities describe the prepared evaluation, not responses collected from a model.
The analysis uses paired comparisons within task families. Its default uncertainty calculation uses 10,000 cluster-bootstrap draws with a fixed seed. Reproducing that analysis from a fresh archive checks that the software and packaged inputs produce consistent outputs. It does not establish that an agent will perform well on unfamiliar tasks.
One retained counterexample is particularly useful: a misleading authenticated receipt can still defeat the proposed gate. Authentication can establish where a message came from while leaving its account of the outcome insufficient. A completion check therefore needs to identify which state would independently establish success, when that state was observed and what uncertainty remains.
A retained counterexample
The contribution is a reproducible evaluation method and an explicit synthetic failure case. A study applying this method would specify model settings, reviewed task labels and a process that retains failures and costs. Its report would include missing trials and ambiguous outcomes alongside completed tasks.
Sources and evidence scope
- Related ZEISMO research and methods
Background reading. Quantities in this article refer to the dated internal case described; the linked work provides context rather than an independent reproduction of that case.
Authorship and preparation
Prepared from ZEISMO research materials and documented project records. Proposed methods and illustrative cases are identified in the article.
Published 9 Nov 2026. Internal editorial checks covered source attribution, scope, consistency with existing research and obvious sensitive information. These checks do not establish the outcomes of the proposed experiments.
Suggested citation
Mohanad Osama Zahran (2026). Before Saying Done: Checking the Outcome Behind a Receipt. ZEISMO Research & Lab, version 1.0.