Queued · waiting for a slot
Four independent attempts at the same answer, so we can tell whether they agree.
The model doesn’t calculate one answer — it explores thousands of possible ones and keeps the ones your data supports. We run that search four times from different starting points. If they all end up in the same place, the answer is stable. If they don’t, we tell you, and we don’t pretend the number is solid.
Each line is one search wandering toward the answer. The first third is throw-away — it is the search finding its feet. What matters is that the right-hand end of all four lines is the same height. These lines are a diagram of the method, not this run’s searches — the sampler reports how far it has got, never the path it took. Whether yours actually agreed is measured at the end, and said either way.
—
weeks being explained · counted as your file is read
34
quantities being estimated
13
weeks hidden from it, to test on later
What is left to do
There is no half-answer to show you while this runs. The sampler reports where it has got to, not what it has found — so rather than draw ranges that would still be moving, here is exactly what is happening and what is left.
- Read your file and checked the columns
- Checked your data can answer the question at all
- Separating each channel's effect from ordinary demand
- Working out what each dollar came back as
- Testing it on the 13 hidden weeks, then trying to break it on purpose
We will tell you this at the end whether you ask or not
Some searches hit rough ground — corners of the answer the sampler cannot fully explore. We count them, divide by the 4,000 draws this run keeps, and hold the result against a 0.1% bar. Whatever that number turns out to be, it appears here and on your results — not in a log file nobody opens.
Stamped so this can be reproduced
- run
- 20260903-104512-a1c7f2
- data
- bd433ddb2822
- engine
- 0.1.0
- holdout
- 13 periods