Approach
The rights, freshness, and confidence rules behind every data row.
Every price, capacity signal, and provider row on ComputeTape carries an explicit evidence label. The labels are a public contract: they tell you where a value came from, how recently a human checked it, and how much weight to put on it. This page is the reference for all of them. Rows whose display rights have not been editorially reviewed — and derived aggregates that have no single primary source — are hidden rather than shown unverified.
Freshness
Provider prices and capacity signals change. A row that was reviewed once is not trustworthy forever. We keep a separate last-checked timestamp on every approved row so freshness is a human signal, not a scraper artifact.
When a reviewer most recently confirmed the source page is still valid and the row still matches it. For official-public-retail-prices rows, this is set automatically on each retail-API run. It is shown on each provider profile and in the data export as the last-verified timestamp.
If last-checked falls more than 30 days behind the page generation time, the row falls back to hidden until a reviewer reconfirms it. The cap is a starting position; it may tighten if operational load is wrong.
Confidence
Confidence reflects how reliable a row is for planning, given the source quality, freshness, and the join used to produce it. It is a separate signal from rights status: an approved row can still be Medium confidence, and a derived aggregate can still be High.
Official provider source, recent check, and a direct read on the listed price or capacity claim. Suitable as the primary input to a planning calculation.
Recent and source-backed, but with at least one normalization step, segment-level estimate, or noisier underlying source. Useful for comparison, less so as a single-source plan.
The source is older, indirect, or the join required several assumptions. Treat as directional only and prefer a higher-confidence row when one exists.
Observed vs illustrative
For any current price, capacity claim, availability statement, or delivery milestone, we record its source and observation timestamp, plus the ingestion timestamp and a methodology caveat where applicable. Observed data is kept separate from illustrative calculations, estimates, and market interpretation so a reader can verify what changed and why it matters. Worked examples in Compute College lessons are illustrative by default: they show the math, not a quoted market price, and are labelled as such.
A value read from a named source on a given date — a provider price page, a public capacity statement, a power filing. It carries a source link, an observation timestamp, and a last-checked time.
A calculation, worked example, or estimate built to explain a concept. It uses round or assumed inputs, is not a live quote, and is captioned as illustrative wherever it appears.
Market reads and "what this signals" commentary. Clearly separated from the observed values they discuss so a reader can weigh the data independently of the take.
Where the labels appear
The same labels are used everywhere they apply. When you read a number on ComputeTape, you can ask the same questions of it on any page.
Renders only rows with approvedForDisplay = true and a non-aggregate rights status. The grid links each source and shows how long the current list price has held in our records (the No change in column). The full last-checked time and confidence are recorded on every row and shown on each provider profile.
Capacity-led: availability and access signals plus a directional H100-equivalent market rate. Per-provider list prices and their evidence labels live in the Provider Directory.
Power signals carry the same source, last-checked, and confidence fields as provider data, and the regional tables show a Confidence column. Aggregated capacity claims are labeled, not implied.
If a row has slipped through wrong, the corrections page records the issue and the fix. Spotting one is welcome.
