← Composa

What Composa won't tell you

Known limits and failure modes, published on purpose · Last updated 2026-07-10 · Ousios LLC

Most tools publish their strengths and let you discover the limits after purchase. We think that's exactly backwards — the fastest way to earn a skeptic's trust is to hand them the failure modes up front. This page is that list. If it ever feels incomplete, tell us what we missed and it goes on the page.

What the assessment can't see

What the ROI number can't know

Where the workflows themselves fail

Every one of the 44 recommended workflows ships with its own "pressure test" — the specific way it goes wrong. A few, verbatim in spirit, so you can calibrate the honesty level:

AI-drafted decision records go hollow

An ADR generated without your actual rejected-alternatives reasoning reads like a press release. The workflow refuses to skip that input — because without it, the output is worse than nothing.

Automated status reports mask the signal

A clean auto-generated report makes a bad week look like a good one. The value is the human exception-review on top — rubber-stamping the draft defeats the workflow.

Small-team "anonymous" surveys aren't

Under about six people, anonymized pulse surveys are identifiable in practice. Running one anyway erodes exactly the trust it's meant to build. We say this in the product, in the workflow caveat, and here.

AI compliance review is not a legal opinion

It's an internal screening that prioritizes what a qualified human should look at. Treating it as definitive creates false security — high-severity gaps still need a real reviewer.

Workload metrics are a poor proxy for burnout

Some people thrive at volumes that break others. Capacity dashboards flag anomalies; only a conversation confirms what they mean — and monitoring must always run with the team's knowledge.

What we can't prove yet

When you shouldn't buy Composa