Three footnote cards sit at the bottom of your Overview page: Exploration, Cache savings, and Cost / merged PR. Together they tell you how much of your AI spend went to work that hasn’t been linked to a pull request yet, how much you saved by hitting the prompt cache, and what each merged pull request actually cost you compared to plain API pricing.

The three cards sit underneath the "Daily consumption" chart, near the bottom of the Overview page. Each card shows a headline number, a short progress bar (where one applies), and a line of supporting text explaining it.

Exploration is the percentage of this period’s spend that went to sessions ATTRIBUT hasn’t linked to any pull request yet. A session becomes linked as soon as one of its commits resolves to a pull request — merged or still open — and from that point its spend no longer counts toward Exploration. Only sessions with no pull request connection at all — experiments, dead ends, and work that hasn’t been tied to a PR yet — count here.

A high Exploration percentage isn’t a red flag on its own. It’s normal and expected for greenfield work, prototyping, or research spikes, where most sessions won’t map cleanly to a single shipped pull request.

Cache savings is the percentage of all tokens processed this period that were served from the model’s prompt cache instead of a full-price call. Cached tokens are billed at roughly one-tenth the standard rate, so a higher percentage means cheaper sessions on average.

These cached tokens are deliberately left out of the daily spend and token charts higher up the page — the Cache savings card is where that volume surfaces instead. If ATTRIBUT hasn’t recorded any token activity for the selected period yet, this card shows a dash ("—") instead of a percentage.

This is your total spend for the period divided by the number of pull requests that merged during it — a real, per-PR cost of shipped work. A "merged PR" here means a pull request that was actually merged, not just opened or linked to a commit.

Where the figure is available, it’s shown next to a second number labeled "vs $X API" — the cost of that same work if it had been metered at the public, per-token API rate. Comparing the two shows how your actual cost per shipped PR stacks up against straight metered pricing.

If no pull requests merged in the selected period, this card reads "$—" and the API comparison is left off.