Methodology

Models publish themselves. Signals wait.

UncoverAI tracks the cost, deployment, and lock-in of enterprise AI. The audience is a CIO, CTO, or chief data officer who needs the number, the clause, and the date it was checked.

What is covered

Frontier Models lists context window, input and output cost per million tokens, deployment (API, VPC or cloud native, on-prem or edge), commercial license, and an executive verdict of ten words or fewer.

AI Platforms compares AWS, Azure, Google Cloud, Databricks, and Snowflake: native foundation models, data-training terms, SOC 2, HIPAA, GDPR, and the inference SLA.

Market Signals is five cards for a month. Each card has the signal, the noise, and the enterprise impact.

How a row is written

1. A scheduled job stays off until PIPELINES_ENABLED is true. Until then, a person starts one run.
2. The job fetches the source and skips the model when the hash has not changed.
3. Models and platforms publish unless a guardrail holds the change for review.
4. Market Signals land as candidates. The owner approves five. Publication copies only those five.
5. If the value changed, the previous figure and the new one are appended to the changelog.

What you are looking at

The rows on this site today are samples. Model names such as Northline Atlas 1 are placeholders. The platform matrix uses the five vendor names so the layout matches the index, and the figures under them are still samples. They are not measurements, audits, or terms. Each sample is marked in the data and on the page.

How to read a status

• Signal — the source states it.
• Caution — the source does not state it, or it is not offered.
• Critical — a term that blocks the workload. Unused until a published row earns it.

The word is the status. The mark repeats it. Color is not used anywhere else.

What we will not do

We do not estimate a number we have not verified. We do not publish an agent draft. We do not overwrite a price or a clause without keeping the previous value.