PRODUCT / ShipProof SUBJECT / Small AI-written web app METHOD / Inside your environment · 10 min
Surface the questions your next handover will get.
A machine reads your fast, AI-written code and returns what it still can’t prove — on a few pages, in 10 minutes. What we didn’t fix, we write down as not fixed. No score. We don’t bend the numbers.
Unproven / Before
23
not visible
Unproven / After
3
visible · agreed
Handover docs
7
added in After
Test coverage
22%
not raised
Before / After — a sample you can check From an app built in 30 minutes to one you can hand over.
Same repository, a branch diff. Code and history are public. What we changed — and what we left untouched — is right there to inspect.
Handover docs
README only
7 handover docs
Test coverage
22%
22% not raised
Env variables
5 undocumented
14 with purpose noted
AI-written code
12 spots, no record
4 marked “unreviewed”
Rollback
no procedure
procedure added
Unproven items
things not yet proven
23items — invisible
3items — visible · agreed
Left unfixed on purposeWhat we did not fix here
- U-01Test coverage stays at 22%. We did not raise it.
- U-0218 dependencies stay outdated. Noted as “update separately.”
- U-03The admin API permission check is marked “unverified.” We did not fix it.
The existence of this “did-not-fix” list is itself the proof that we sell a state you can explain, not a quality bump. The conversation only starts once the other side can say “ours is 22% too” and make it their own.
What you get A few pages. No score.
We return a list of “things not yet proven,” not an overall grade. What we couldn’t detect, we write as “we couldn’t find it” — never as “it isn’t there.”
P1Count and type of unproven items
Not a score — a count and a type distribution. What isn’t proven, and how much.
P2Questions likely at handover
For “is this tested?”, what you can show today. In priority order.
P3Where maintenance handoff stalls
Env vars, dependencies, deploy, who to call on an incident. Can one person stand it up?
P4Which docs to write first
Highest effect for lowest cost — plus what this diagnosis could not cover, stated plainly.
This is not a service that raises quality. It’s a service that makes “what’s proven and what isn’t” explainable to your customer. We don’t make what isn’t done look done.
Getting ahead of the doubts We remove your reasons to say no, up front.
Q1Isn’t that analysis script dangerous?+
You get the full script. Read it before you run it. We’ll show you, on the spot, that it runs with the network cut off (--network none). It holds no key to send anything to an external AI service. Not a promise not to send — a structure that can’t.
Q2Couldn’t I do this myself with AI?+
You can generate it. What we sell is the operation of bundling it into a form you can submit and stand behind — and the fact that an outside party verified it. We don’t argue on features.
Q3Security won’t allow it.+
The analysis completes inside your environment. We never receive the code, not once. So when you decline, there’s nothing on our side to delete either.
Q4How much is it?+
The diagnosis is free. The handover-evidence pack after it is ¥150,000. We say it up front. We don’t blur it.
The same thing, on your repository.
Just read access to the target repo, or a ZIP. We return the spots likely to trip up your next handover, on a few pages. This is not a request for a sales meeting.
Request a free diagnosis → ShipProof is the first product of Mananect. We’re building Mioca — an AI Operating Company (an OS where AI runs a company’s operations).