SABLE / Insights
Forecasts you can audit: registering predictions before the outcome
A forecast that can be edited after the fact is a story. A forecast that was written down first, and scored against a baseline afterwards, is evidence.
August 18, 2026 · 5 min read
Anyone can be right in hindsight. The value of a forecasting system is not that it produces numbers, it is that it produces numbers you can trust enough to act on. That trust has to be earned, and there is only one honest way to earn it.
Register first, score later
A forecast should be recorded before the outcome is known, with its date, its inputs, and its stated probability. When the outcome arrives, the forecast is scored against it, and against a simple baseline such as "things stay as they are". No edits after the fact. No quiet removal of the misses.
This sounds obvious. It is also rare. Most forecasting inside companies lives in spreadsheets and slide decks that get revised as reality unfolds, so by the quarter's end every forecast looks reasonable and nobody can say how good the process actually was.
Why the baseline matters
A forecast that is right seventy percent of the time sounds impressive until you learn that "no change" was right seventy-five percent of the time. Skill is performance above the baseline, not raw accuracy. A system that reports accuracy without a baseline is reporting a number that flatters itself.
Why probabilities beat point estimates
"Lead time will be 21 days" is a point estimate, and it will almost always be wrong by some amount. "There is a 30 percent chance lead time exceeds 28 days" is a probability, and it can be scored properly, it tells you how much to hedge, and it stays honest about uncertainty. Decision-makers do not need false precision; they need calibrated odds and the reasons behind them.
What auditability buys you
- Trust that compounds. After a few months you know which kinds of forecast the system is good at and which it is not, and you weight them accordingly.
- Better arguments. When two people disagree about a decision, the register shows what each forecast assumed and how similar forecasts have scored.
- Cleaner post-mortems. When something goes wrong, you can see whether the signal was there and was missed, or was never there.
How SABLE handles it
Every forecast the system produces is registered before the outcome, scored against baselines with no hindsight edits, and reported as measured rather than claimed. Every output carries its sources and its assumptions, and corroboration is counted from distinct sources rather than dressed up as a percentage. Forecasts and scenarios are probabilistic model outputs provided for informational purposes only, and the decision stays with the person who owns it.
That is a deliberately modest promise. It is also the only one a forecasting system can keep.
