We all want to communicate with impact — otherwise what are we doing. The techniques for it are well known and a century old. The hard part is knowing they exist, knowing which one this message needs, and noticing what you left out.
Praxis does those three things. It decides what a message should do for its reader, tells you what your draft is missing, and proves nothing that mattered went missing in the rewrite. It never writes a word of the prose.
Bottom line up front. The pyramid principle. SCQA. PREP. Situation–impact–action. Claim–evidence–reasoning. There are more than a dozen in common use, and most people can name two.
Praxis holds twelve as inspectable data — each with what it favours, what it avoids, and when it is the wrong choice.
The right shape depends on who reads it first, what they can actually approve, what happens if they delay, and how much time they have. Those are answerable questions. Nobody answers them.
Praxis computes the shape from your answers — and asks only the questions whose answers would change it.
The deadline. The caveat. The figure. The person who owns the next step. Rewriting for tone is exactly when these go missing, and the draft still reads fine without them.
Praxis checks, and shows its evidence — no score, no verdict, just what it saw and where.
A real draft, and real output — this is what the tool actually printed, not a mock-up. The message is a request to a busy stakeholder at high stakes, which is the situation most likely to be got wrong.
Hi Priya, I wanted to give you some background on the migration work. Over the last six weeks the team has been looking at the legacy queue and we've done a fair amount of analysis on the failure modes. There's quite a lot of history here, some of which predates the current team. The upshot is that we think we should probably move to the new pipeline. It would be good to get your thoughts when you have a moment. Thanks, Dave
Fix 5 things before sending: no request of the reader; 1 unsupported claim; nothing marked uncertain. Plus 2 more. Shape it bottom line up front.
Ask for changes and it stops describing and starts pointing. Each edit carries a character offset into your draft, so there is no guessing which sentence it meant.
Every intake form you have ever filled in asked for everything. Praxis walks each unknown across its range, re-runs the recommendation, and asks only where the outcome actually splits. Three states of the same draft:
You have told it nothing.
You have said the stakes are high.
You have answered three.
It does not write your message.
The engine holds no API key, calls no model, and reaches no network. It cannot produce a sentence, and adding that would destroy the reason to trust its audits — a rewriter grading its own rewriting is not an audit.
It does not score you.
No number, no grade, no percentage out of a hundred. Findings come with the evidence attached, and you are free to disagree with every one of them.
It cannot tell you whether you are right.
It can see that no deadline appears and that a consequential claim sits nowhere near a figure. It cannot see whether your argument is sound or your number is true. It reports unknown often, and on purpose.
It is working, not finished.
Which is what the testing is for.
A recommendation to leadership, a request that needs a decision, a piece of bad news. Something with a real reader and a real consequence — the tool is uninteresting on writing that does not matter.
Fifteen minutes. It runs entirely in your browser: no account, no upload, nothing leaves your machine. Bring the message you would be nervous to get wrong.
Background, if you want it first — In Defense of Mechanical Writing ↗ and More on Mechanical Writing ↗. The side-by-side matrix shows the mechanisms against each other.