How every number is made. Computed, tested, replayable.

No number you see in Othor AI is written by a model: not on a page, not in an email, not in a ranking. Every number comes from your data, computed by code, and is placed into the text from the computed result. If a model’s wording ever contains a number, it is thrown away and the computed value is used.

Models only draft: your setup, a new brief, a change you describe in words, and the wording of a brief from sentences that are already complete. You review a draft before it applies, and wording is checked against the computed numbers before it is shown.

Reading, computing, testing, forecasting, ranking and the brief’s numbers all run without a model. If the model provider is down, the brief still arrives on time, with every number, in plain template wording.

Open any finding or number, and Othor AI shows:

• The definition from your Ontology, and the version it was computed with.
• The exact SQL that read it, which you can copy and run yourself.
• The test, its parameters and its result.
• The run it came from, with its date and its number.

Each night’s run is recorded and sealed. Replaying it, from the same inputs, gives the same numbers byte for byte. Nothing about a past run changes later: your snoozes, mutes and Ontology changes apply from the next run, never to the one already made.

Othor AI reads your sources with read-only logins and refuses one that can write. It reads only what changed since last night, keeps the results rather than copies of your tables, and never reads columns that hold personal data. No row of your data is sent to a model. More on the security page.