So we keep score, and we publish the method. This page explains exactly how XPlor's forecast is measured, what we show you, and — just as important — what we refuse to show you.
The loop
1. Every Monday, the forecast becomes a record. When your 13-week forecast is built, week 1's projected inflows, outflows, and ending cash are written down and frozen. Not adjusted later. Not quietly re-baselined. Frozen.
2. The following Monday, reality grades it. Your actual bank balances and transactions come in through your books and bank feeds. XPlor compares: what did we say week 1 would look like, and what did week 1 actually do? The difference — the forecast error — is stored, signed, per week.
3. Errors become confidence bands. Once enough Mondays accumulate, the history of errors becomes a band around the forecast: “based on the last N weeks, actual ending cash has landed within ±$X of forecast, 9 times out of 10.” That's a p90 band computed from your business's measured history — not an industry benchmark, not a model's self-reported confidence, not vibes.
4. The band appears only where it's earned. This is the part we're most stubborn about. A forecast four weeks out with three Mondays of history behind it does not get a band, because any band we drew would be decoration. The product says so, on the screen, in plain text:
Week 1: within ±$100k at p90, based on 4 Mondays of history. Weeks 2+: not enough history — no band rather than a guessed one.
No band rather than a guessed one. If that line looks unpolished, good. It's what measurement actually looks like early on. The alternative — the industry default — is a smooth chart with no error bars and no way to know it's been wrong by 40% for a month.
Why the errors are even correctable
A forecast can only be honestly measured if the numbers are deterministic — produced by rules over your real invoices, bills, recurring payments, and balances, so that the same books always produce the same forecast. XPlor's are. When week 1 misses, we can decompose why: which invoice paid late, which bill landed early, which recurring pattern changed. Click the error and see the documents.
That's also why an AI narrating your books can't do this. There's no frozen prediction under the chat answer to grade, and no decomposition when it's wrong — just a new answer next time you ask, with no memory of the last one. Determinism isn't an implementation detail; it's what makes accountability possible.
What we don't do
We don't show bands we can't back. We don't re-baseline history to make old forecasts look better — the misses stay on the record, because the misses are the training signal and the trust signal. We don't average your business against “similar companies” to fake a longer history. And we don't quote a single global accuracy number in our marketing, because forecast accuracy is per-business and per-horizon, and any single number would be an average hiding the truth — which is the exact sin the runway formula commits.
What this buys you
When XPlor's Monday briefing says “cash crosses your buffer the week of September 14,” you know three things no chat tool or spreadsheet can give you: how that date was computed (click it), how forecasts like it have performed for your business (the band), and that next Monday the claim gets graded whether it's flattering or not.
That's the standard. Hold every tool that touches your cash to it — including us.
The public track record
This section holds our aggregate, anonymized numbers — computed by the grading loop above, never typed in by marketing. Misses included.
The dashes are deliberate. Design partners are accumulating measured Mondays now, and the first aggregate numbers appear here when there are enough of them to mean something — the same standard we apply to your confidence bands. No number will ever appear on this page that wasn't computed by the loop above.
Updated as the grading loop runs. If a number here ever disagrees with what the product shows you, tell us — that would be a bug in this page, not a rounding choice.