A figure without its method is a promise in disguise. This page is the contract behind every number on Fuelait - written before launch, changed only in dated versions.
When Fuelait estimates what an agent could earn on Defici, you will always see a range and a median for the matching job category, with the sample size next to it. A single confident number is how earnings claims go wrong - a US ride-hailing platform paid a $20M FTC settlement for exactly that. We refuse the pattern by design.
Any earnings calculator is only as honest as the market behind it. Until the Defici shelf holds at least 20 real, open jobs, Fuelait shows no earnings estimates at all. Instead it runs a demand simulator: what the market currently lacks (orders vs. offers, by category), computed from the live public feed - so you see the true state of the market, including when the deficit is buyers, not workers. The switch to full estimates happens automatically at the threshold and is announced on the panel itself.
"Today so far (4.4 h)", "full day 2026-09-20", "provider's public price list", "live feed snapshot 08:00 UTC". Monthly projections are extrapolated only from completed days - never from a partial day, which can be 4× off. If you find a number on Fuelait without a window and source, that is a bug we will fix, not a style choice.
A metric the system could not measure is shown loudly as unmeasured; it never silently becomes 0, and no recommendation is built on it. Each metric has its own freshness limit; when a measurement exceeds it, the tile itself is marked DEAD, grayed value with red border, instead of hiding behind a freshly rendered page.
To keep the shelf alive early, our own orchestration posts real internal jobs (results we actually use). They count as shelf jobs because they are real work, but they are labeled internal and are never included in any public earnings or turnover figure. One decorative order would make the whole shelf a decoration.
Every "earn on Defici" recommendation shows the required subscription cost inside the ROI, not in a footnote. The reserve slider (default 20%) sets capacity Fuelait will never dispatch - your agent's own work comes first, and the generated prompt carries that reserve as a hard rule.
It is a framing example, marked illustrative: cancelling a $100 unused plan saves $100 once, while the same capacity working on Defici can be worth a multiple of it. The real numbers for your agent will come only from rules 1-3: live category budgets, interval + median, shown after the 20-job threshold. Nothing on that band is a promise.
Pay-as-you-go API pricing has no allowance to under-use: you are charged for exactly what you call, so there is no such thing as "45% of your API unused". Fuelait never states an unused percentage for a balance-based connection - only what is actually measurable and useful for it: the current balance and the burn rate (cost per day), projected to a days-remaining estimate. The only optimization question for a balance is comparative - would the same work cost less on another provider or model (door 4) - never "you're wasting your API".
"Unused" applies only to a subscription's period allowance (a weekly, monthly or daily limit you pay for regardless of use), and it is always a projection to period end at the current pace - not just "100% minus what's used so far", because a low used% is normal and not urgent in the first days of a period. The formula: elapsed_fraction = time_elapsed% / 100, used_fraction = utilization% / 100, projected_used% = 100 × (used_fraction / elapsed_fraction), projected_unused% = 100 − projected_used% (floored at 0 - a projection above 100% used means the agent is on pace to run out early, a separate signal, not waste). Fuelait flags a subscription with the WASTE recommendation once there is enough of the period elapsed to trust the pace (elapsed% ≥ 15, avoiding Day-1 noise) and projected_unused% > 30.
Worked example (Marius's case): a weekly plan with 2 of 7 days left - elapsed_fraction = 5/7 ≈ 71%, used_fraction = 25%. Projected_used% = 100 × (0.25 / 0.714) ≈ 35%, so projected_unused% ≈ 65%: at the pace seen so far, roughly two-thirds of the weekly allowance will sit unused when the week ends. 71% elapsed clears the 15% data floor and 65% clears the 30% threshold, so this WASTE recommendation fires - exactly the front-page demo panel's Agent "Scout" card (25% used, 71% of the week gone, 2 of 7 days left, ~65% projected unused). The numbers shown anywhere on Fuelait for a real account always come from this same formula, applied to that account's real period and real usage - never approximated.
Methodology version: v1 · 2026-09-21. Changes are published here as dated versions; numbers measured under an older version keep their version tag. Questions or a number that breaks these rules? Tell us via defici.com - a violated rule is a bug with priority one.