An unfinished MVP can look almost ready while its central customer workflow still depends on manual database edits. Before funding another development phase, establish what works reproducibly and what has only been demonstrated once.
This review is for founders deciding whether to continue, narrow or rebuild an existing product. The useful output is a decision with evidence, not a percentage-complete label that hides the remaining dependencies.
What does an unfinished MVP review measure?
The review compares the intended customer outcome with the current implementation. It separates implemented behaviour, demonstrated behaviour, missing work and assumptions that have not been tested with customers.
A feature checklist is useful background, but one feature may depend on several unfinished services. A completed payment screen does not establish that account access, subscription state and refunds behave correctly.
Trace one customer journey through the real system
Choose the narrowest valuable workflow and attempt it with a fresh test account. Record every manual intervention, preconfigured record and developer-only permission needed to reach the outcome.
| Review finding | Evidence to collect | Funding implication |
|---|---|---|
| Workflow completes independently | Repeatable test and resulting records | Candidate for pilot hardening |
| Staff intervention is required | Named step and responsible person | Budget operational work explicitly |
| Dependency is unavailable | Access or integration blocker | Resolve feasibility before expansion |
| Customer outcome is unclear | Missing acceptance criterion | Revisit scope before more features |
Distinguish a consciously manual pilot from a broken automation claim. A human-reviewed step can be a sensible early product decision when its cost and responsibility are visible.
Review the cost of continuing before proposing a rewrite
Inspect account ownership, deployment reproducibility, data structure and the ability to test a small change. Those constraints often determine whether another engineer can work safely in the existing codebase.
A rewrite adds migration and behaviour discovery work. Continuing with the existing system may preserve hidden assumptions that create repeated failures. Ask for a small repair experiment when the decision remains uncertain, rather than committing to either path from repository appearance alone.
The AI codebase inheritance checklist helps structure this inspection. Customer validation is a separate question: a maintainable product can still solve a problem that buyers will not pay to address.
Fund the next evidence-producing milestone
Define a bounded milestone around one usable outcome, the required integration tests and a reviewable operating procedure. Include explicit work that remains out of scope so the next estimate has a stable boundary.
For a product like my independent research platform, the review should follow the actual study creation and participant response workflow, rather than infer readiness from the landing page. Apply the same principle to your own product.
Can an engineer estimate completion from screenshots?
Screenshots help describe intent, but they do not expose dependencies, data quality or release work.
Is a rewrite always wasteful?
No. It may be justified when evidence shows that repair and migration constraints favour replacement.
What should we commission first?
Start with a scoped assessment and an evidence list in the external technical review brief.