Skip to content

Engineering Stage

ES-107 — Evidence

Define the evidence required to show that AI-assisted implementation was controlled, traceable, verified, and accountable.

Evidence
Implementation Proof
Next: Outputs

ES-107 Evidence

Purpose

This page defines the evidence expected from ES-107.

Implementation evidence demonstrates that code was produced in a controlled, traceable, verified, and accountable way.

Required evidence

Artifact Evidence Purpose
implementation_log.md Records implementation activity by work item.
ai_assistance_log.md Records material AI assistance and verification.
work_item_completion_records.md Records completion status and evidence per work item.
implementation_evidence.md Summarizes implementation evidence and guardrail preservation.
verification_notes.md Records tests, checks, failures, fixes, and gaps.
implementation_decision_records.md Records implementation decisions.
implementation_risk_updates.md Records risks discovered or changed during implementation.
implementation_summary.md Summarizes readiness for ES-108.

Evidence quality expectations

Implementation evidence should be tied to work items, tied to commits or pull requests where possible, traceable to design and requirements, clear about AI use, clear about verification, explicit about guardrails, honest about known gaps, and useful to code review and integration.

AI-use evidence standard

AI use should be recorded when it materially affects implementation.

Record the work item affected, AI assistance purpose, generated or suggested output, what was accepted, what was rejected, verification performed, and responsible engineer.

Do not store sensitive prompts or data if policy or project rules prohibit it.

Verification evidence

Verification evidence may include unit test results, integration test results, manual review notes, static analysis results, build results, access-control checks, evidence-flow checks, AI-output review, and failure-path checks.

Verification does not need to be perfect at this stage, but it must be visible.

Common pitfall

Do not write “tested” without saying what was tested and how.

Engineering insight

Implementation evidence is what lets reviewers distinguish disciplined engineering from lucky coding.

Continue to Outputs

Review the outputs ES-107 produces for review and integration.

Continue to Outputs →