EVODA designs, tests and delivers; you execute. The execution prompt ships with the environment assumptions, the validated DDL, the verification steps and the rollback, ready for a person or for the agent on your side.
The problem
Between the approved document and the applied change there is a gap: someone translates the PDF into commands, under pressure, without the tests the author had in mind. That is where the approved architecture and the deployed one start to diverge.
The translation comes done and fenced: explicit environment assumptions, application order, checks after applying and a rollback if any of them fails. Whoever executes follows the script; what is not in the script is not authorized.
How it works
The prompt declares the environment it expects to find and instructs stopping outside it. The post-application checks are executable, and the stop instruction is explicit: one fails, apply the rollback and stop.
No embedded credentials, no SUPERUSER, inside a transaction. The rules are written in the prompt itself, because it was made to be handed to an agent.
Project / Analytics layer / Deliverables
Execution prompt: create the analytics schema
Version 3 · for the client agent or the DBA
Environment assumptions
PostgreSQL 16 · user without SUPERUSER
run inside a transaction · no embedded credentials
if any check fails, apply the rollback and stop
Checks after applying
A data model becomes a dbt project or migrations; a document becomes a PDF or a revocable public link. The seal travels along: what was verified goes with the document.
And the executor’s report comes back to the project, closing the loop: the engagement knows whether the change happened, and the follow-up starts from there.
Transactional model for e-commerce
Version 2 · audited · 6 entities, 31 columns
| DDL executed on a real database | done |
| Business questions answered by query | done |
| Adversarial review | done |
| Proof under load | not measured |
Second opinion, from a different model
Disagrees with normalizing stock_movements: at high volume, a balance computed by summation becomes the bottleneck. Suggests materializing it. The disagreement was recorded, not resolved.
Under the hood
What the execution prompt always carries.
What comes next
The Claude Code on your side connects over MCP, runs read-only investigations and returns reports and environment snapshots on its own.
See it from the inside →ProofMachine verificationScripts executed on a disposable PostgreSQL, business questions answered by query, and the seal showing what was and was not verified.
See it from the inside →LGPDLGPD per columnEvery column with a classification and an anonymization technique, views for non-production environments, and the impact report when the design demands it.
See it from the inside →Paste the structure, get the most serious findings in about two minutes, and decide later whether an engagement is worth opening. No account, no card.