Cresva Research

Research

Papers on what marketing agents can be shown to do, written so that somebody who does not work here can check the result.

CRV-2026-001Aug 202636 min read

Proof of Write: An Evidentiary Standard for Agent Action Capability

Shubham Raghav

Agentic products describe their ability to act on external systems in the language of breadth: platforms connected, verbs shipped, integrations closed. Breadth answers a question nobody asked. Running one hostile, evidence-gated audit against one system on one day produced four different coverage figures from the same 130-capability matrix, differing only in what the auditor accepted as evidence: a breadth narrative for which no value was ever computed, 56.6% for code that exists and looks contract-correct, 5.4% once a requested permission scope had to be enforced before the call rather than merely requested at consent, and 0.0% for writes proven to have reached a vendor. We propose an evidentiary standard that resolves the ambiguity: seven tiers, each with a criterion a third party can apply without access to the code, the artifact that satisfies it, and who is able to verify it. We then apply the standard to the system it was derived from. The top-tier count is zero.

Agent capabilityEvidenceEvaluationWrite access

What we publish, and why

Cresva builds agents that take actions inside advertising and commerce accounts. Almost everything written about that category, including by us, is a description of intent. A paper is where we are obliged to write down the method instead, and to write it down in enough detail that a reader can disagree with it on the specifics rather than on the vibe.

Three things go here. Standards, meaning a definition of what would count as evidence for a claim in this field, ours included. Measurements, meaning results from instruments we run in public, such as the brand index. And methods, meaning the mechanics of something we built, described closely enough to be reimplemented by somebody who does not have our code.

Every paper carries an identifier, a version and a date. The version moves when the argument changes, not when the wording does, and a superseded version stays reachable, because a citation that quietly changes underneath the person who made it is worse than no citation. Where a paper reports a number, the number comes with the method that produced it. Where we do not know something, the paper says so.

For the applied version of the same material, the guides cover what to do about it, and methodology covers how the product itself decides.

New papers go out with the newsletter.