Skip to main content
Technology & EngineeringEnso Method81 lines

Enso Method Evidence-Based Review and Testing

Review deliverables against intent, audit delivery and tests by acceptance criterion, and verify repaired evidence at the boundary where the outcome is observed.

Quick Summary3 lines
This SkillDB entry is a self-contained adaptation of the Enso Method by EnsoDynamics.
skilldb get enso-method-skills/evidence-based-review-and-testingFull skill: 81 lines
Paste into your CLAUDE.md or agent config

Enso Method Evidence-Based Review and Testing

This SkillDB entry is a self-contained adaptation of the Enso Method by EnsoDynamics.

Use this workflow after a phase produces a deliverable, when a complete requirements document is ready for close-out, or when a passing suite provides uncertain evidence. Keep separate records for observed delivery and the ability of tests to detect a regression. Pin both to requirements.md, the artifact revision, and the available verification environment.

Review the deliverable

Write a short statement of the user intent, the in-scope acceptance IDs, and the proposed operating boundary. Inspect the actual result with supporting requirements, design, decisions and assumptions. Review a rendered deliverable in its target form when visual behavior matters. For code, follow the normal entry point and check the observable outcome with a suitable fixture. Record methods and limitations if the environment prevents execution. Use independent judgment against the artifacts and ask for an independent reader when available and permitted.

Record findings with ID, criterion, location, expected and observed behavior, impact, and the smallest useful correction. Reproduce a material finding where feasible before changing behavior. Resolve contradictions in the owning documents, then implement the accepted correction. Compare the repaired artifact against the original intent and rerun affected checks. A review with unresolved findings needs an explicit disposition, owner and evidence boundary.

Audit whole-requirements delivery

After the scoped phases or completion work land, do a fresh pass over the entire in-scope requirements document before the test audit. Check the baseline revision in the acceptance matrix. Revisit each required variant even when the first variant already exposes a gap. Confirm neighboring cases and shared boundaries before writing a fix. Evidence from a previous revision can be reused only where the input, boundary and outcome are demonstrably unchanged.

Delivery rowRequired record
Identity and outcomeCriterion and variant IDs, requirements revision, approved expected behavior
InspectionArtifact and entry point, test data, environment, reproducible observations
DispositionVerified, partial, missing, or blocked, with observed facts and limits
Completion workGap, risk, dependencies, exit evidence, accountable owner

Use verified where trustworthy evidence meets the entire criterion. Use partial where one boundary or variant is demonstrated and another is missing or unresolved. Use missing where inspection establishes absent required behavior. Use blocked where authorization, environment or a decision prevents valid assessment. For a partially verified row, preserve both the supported evidence and the unobserved part. Group related gaps into bounded completion phases. Carry forward the original criteria and record the missing condition instead of rewriting success around the implementation. Repeat the whole-requirements audit after completion work to catch regressions and newly affected siblings.

Audit and harden the tests

For every required variant, inspect test setup, the boundary crossed, exact assertions and cleanup. Grade missing when none of a criterion's behavior is exercised. Grade weak when tests exist but a required part could break unnoticed. Grade adequate when each required part has an applicable assertion at its boundary, with red verification or implied edge cases still open. Grade strong when relevant edge cases are covered and controlled failures have demonstrated detection of each required break. Use the least-proven required part to grade the criterion and record an unavailable environment as a separate blocker. A skipped test, or one excluded from the project's normal run, supplies no passing evidence.

Use a local controlled failure or equivalent red check where feasible, starting with failures that could silently affect money, data, access or external state. Confirm the expected assertion fails for the intended reason, restore clean state, then rerun the original check. If red verification is unavailable, describe inspection evidence and retain the corresponding limitation.

Choose the lowest-cost check that actually crosses the important boundary. A deterministic policy test suits a clock cutoff. Use ordinary application entry points and persistent state for authentication, concurrency, or recovery behavior. A synthetic provider can simulate lost responses, count modeled calls, and return controlled status. It can substantiate behavior against that contract. Verifying the real provider contract calls for an approved sandbox or equivalent evidence. Record differences in claims, avoid production data or credentials in fixtures, and link sanitized output. Repair under-asserted behavior before adding more routine cases. Run focused regressions and the appropriate full suite, with exact filters, exit status, environment and relevant failures recorded.

Project Returns review

Project Returns approves full refunds to the original payment. R-01 covers authorization and correct completion. R-02 covers duplicate requests. R-03 requires an appropriate safe result for denied or ineligible requests, with no payment change, provider effect or committed refund. R-04 preserves an uncertain outcome until authoritative reconciliation. An old packet used the proposed A-01 cutoff of 30 elapsed 24-hour days from the recorded payment timestamp through the recorded request timestamp in UTC, inclusive. The owner later confirmed 45 days and supplied an effective date. Use the revised requirements and include a day-40 case under the new policy. A prior day-20 success is compatible with both policies and cannot establish the new cutoff.

VariantPositive and negative evidence
R-01 authorized eligible orderAt day 40, authorized approval produces the exact full refund for the original payment and a matching receipt. With an unauthorized actor the same request creates neither refund state nor provider call.
R-02 repeated operation IDEquivalent retries including concurrent requests return one operation and produce one modeled provider effect. A changed payload for the same ID returns a conflict with no mutation.
R-03 ineligible orderA request beyond the cutoff leaves payment state unchanged with zero modeled calls. Compare the policy at exactly 45 days and 45 days plus one second at a fixed UTC clock. Check the safe result, the absence of a committed refund and the unchanged original payment.
R-04 uncertain provider resultSimulate provider acceptance with a lost response. Verify pending state and its receipt survive restart. Reconcile the original reference to one confirmed effect. An inconclusive lookup leaves pending state and produces no second send.

A hypothetical existing suite asserts two matching receipt IDs under serial retries. Audit R-02 as weak until it checks provider effect count and persisted state under the relevant retry and concurrency conditions. In a disposable fixture, make the provider recorder observe a second send and verify the strengthened check fails on that count. Restore the fixture and rerun the regression. For R-04, distinguish the case where the provider returns an authoritative result from the case where a status response is missing or stale. A green check that only proves the response says pending is insufficient evidence of restart recovery.

Close out and hand off

Record requirements revision, artifact revision, delivered and open variants, evidence paths, test grades, regressions, and consequential limitations. Distinguish accepted deferrals from verified behavior. Name the next phase, accountable owner, prerequisite and first authorized action for each open gap. For example, pass synthetic R-04 evidence to the adapter owner with the approved sandbox matrix and precise lookup behavior still to confirm. If the test audit reveals a product defect, reopen the affected implementation and delivery evidence, then re-audit the test coverage. Review and test evidence describe readiness. They do not grant separate authority for production changes or release.

Upstream license

MIT License

Copyright (c) 2026 Doug Kerwin

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OF OR OTHER DEALINGS IN THE SOFTWARE.

Install this skill directly: skilldb add enso-method-skills

Get CLI access →