Mathine — Evidence In, Trust Out

TTOkay for Coding Assistants in Regulated Pipelines

Mathine: Time-to-Okay Operating License Machine
Link: https://doi.org/10.5281/zenodo.18688855 [1]

Coding assistants are increasingly used inside regulated software delivery pipelines, yet deployment decisions are still often justified by static benchmark wins or aggregate offline scores. This paper proposes TTOkay: a policy and evidence schema that ties “okay-to-operate” to continuously auditable conditions, turning evaluation from a one-time score into an operating license enforced by verifiable receipts.

TTOkay binds permission to three live constraints: receipt freshness relative to the active dependency graph, worst-slice bounds (not averages) as the primary risk gate, and drift corridors that define admissible distributional change plus automatic de-licensing triggers. The key move is that “safe enough” becomes a checkable claim under an explicit regime, not a narrative inference from a leaderboard.

The paper makes this operational by tiering permissions by action class—e.g., suggest-only → patch-only → PR-open → auto-merge—and requiring that each tier be backed by a minimal, replayable receipt set suitable for audit. In other words: the more irreversible the action, the stronger and fresher the receipts that must accompany it.

Two constructs do the heavy lifting. First, a worst-slice upper-bound mechanism for risk gating under finite budgets (so “tail safety” is bounded, not assumed). Second, drift corridors that connect change control to measurable signals, so capability drift, dependency drift, and data drift are handled as explicit license conditions rather than surprise incidents.

The falsifiable predictions are direct: worst-slice gating should outperform pooled means for predicting real pipeline failures; receipt freshness should correlate with reduction in false-closure events after dependency changes; and drift-corridor enforcement should reduce recurrence by making de-licensing automatic and defensible under scrutiny. [1]–[5]

References
[1] R. Figurelli, “TTOkay for Coding Assistants in Regulated Pipelines”. Zenodo, Feb. 19, 2026. https://doi.org/10.5281/zenodo.18688855
[2] R. Figurelli, “Benchmarks-as-Contracts: A ReceiptBench Spec Template for Regimes and Closure”. Zenodo, Feb. 17, 2026. https://doi.org/10.5281/zenodo.18675035
[3] R. Figurelli, “From Scores to Receipts: Introducing ReceiptBench, a Typed-Receipt Protocol for Governance-Ready Evaluation”. Zenodo, Feb. 16, 2026. https://doi.org/10.5281/zenodo.18661829
[4] R. Figurelli, “Field Definition Language (FDL): A Proposal to Evolve APIs into Governed Fields”. Zenodo, Oct. 18, 2025. https://doi.org/10.5281/zenodo.17382665
[5] R. Figurelli, “Large Language Fields (LLFs): The Invisible Layer Above LLMs”. Zenodo, Oct. 03, 2025. https://doi.org/10.5281/zenodo.17254137