In AEGOS, nothing is just "true." Every fact carries a grade for how strongly it's known and where it came from. That grade travels with the claim, so anyone — human or AI — can see at a glance whether a number was measured, observed, reasoned out, or simply assumed.
The strongest grade. A fact is measured when it comes from a direct, real reading of the real thing — a timer on the actual run, a byte count of the actual file, a status read from the actual machine, with no modelling or proxy in between. Measured facts are the bedrock consequential decisions rest on: when an action could touch production, money or a customer, AEGOS acts on nothing weaker. “The deploy returned HTTP 200” is the kind of measured check that clears a gate; “the deploy probably worked” is not — that is an assumption wearing a measurement’s clothes.
Seen first-hand, but not precisely measured. An observation is something a human or the system genuinely saw — a page that rendered, a job that appeared to finish — without a precise reading attached. It is real, credible evidence a notch below measured, and keeping the two distinct stops a soft impression being quoted as a hard number: “it looked fast” never quietly becomes “it took 200ms.” A screenshot confirming a layout rendered is an observation — enough to trust the page is live, not enough to claim a load time.
Reasoned from other evidence rather than seen directly. If A and B are both measured, C may follow — but the conclusion was computed, not observed, and AEGOS marks it that way. Inference is powerful and necessary, yet labelling it stops a chain of reasoning hardening into a “fact” no one actually checked. “The cache is warm because the last three requests were fast” is an inference: reasonable, and honestly flagged as a conclusion rather than a direct reading.
The weakest grade — stated but unverified. An assumption is a default, a guess, a “probably” with no evidence behind it yet. AEGOS does not forbid assumptions; it forbids hiding them. Most failures begin as a confident assumption dressed as a fact, so grading it “assumed” keeps it visibly weak — no gate clears on it, and no decision quietly comes to depend on it. “The API key is probably still valid” is an assumption; acting on it without a check is exactly the move AEGOS exists to catch.
Most failures start with a confident guess dressed up as a fact. Grading evidence makes that impossible to hide: weak claims stay visibly weak, strong ones earn trust, and a result produced by a stand-in is always labelled as simulated — not measured. It's how the whole system stays honest as it moves fast.