What changes tomorrow?
Choose a site, hold a harvest, reopen water, or decide whether a new data feed is worth paying for.
Water decision science
Make the call before the result comes back.
Which water do you test first? Do you harvest today? Is a closure ready to lift? MBAL tests whether the observations already around you can make that decision better — then tries to prove the answer wrong before you rely on it.
- Open data first
- The free rule goes first
- Failure stays visible
The useful window
Decision due
Now
Result returns
Later
A useful system earns its place in this gap. If it cannot beat the rule you already use, we say so.
Start with the decision
One call. One clock. One cost of being wrong.
Not “build us a model.” Give us the choice somebody has to make while the evidence is still incomplete.
When must you know?
We fix the lead time before we look at the score. Knowing today is not the same as warning about tomorrow.
What does a miss cost?
Lost product, an unnecessary closure, a wasted sampling run, or false confidence in a tool that does not travel.
The test
Five ways we try to kill the idea.
A polished score is easy. A result that survives the way you actually work is much harder. These are the checks that matter before a tool reaches an operating plan.
The real choice
Score the decision, not the spreadsheet.
We test the order you can act on: by sampling run, crew day, harvest window, or closure review.
Fails ifthe apparent advantage exists only after unrelated places or dates are pooled together.
The ceiling
Work out what perfect knowledge could achieve.
We calculate the best possible result under the real schedule before judging any model.
Fails ifthe claim exceeds what even perfect foresight could deliver.
The free rule
Make the simplest fair rule go first.
Last result, published threshold, season, and site history all get a fair chance. They have beaten our own work.
Fails ifthe new method cannot beat the strongest rule you can run without buying anything.
The holdout
Ask it to work somewhere it has not seen.
We hold back a future period or a whole place, chosen before the result is known.
Fails ifthe improvement disappears outside the water or season used to build it.
The record
Let failed ideas stay failed.
Null results, broken assumptions, and withdrawn claims remain part of the evidence.
Fails ifthe answer depends on hiding a test that did not work.
Where it fits
The water changes. The decision shape repeats.
Dangerous algae and the toxins they produce. Unsafe bacteria. Water running out of oxygen. Flooding. Long-run habitat loss. Rivers, lakes, estuaries, coasts, and the open ocean: the work starts only when the evidence can support an honest call.
What to sample first
Whether a warning is early enough
When to close or reopen
Whether a tool is real
Whether new data changes the call
Ways to work
Use the lab at the scale of the decision.
A bounded test of one call, one lead time, and the evidence that already exists.
A check run separately from the people who built or sell a forecast, score, sensor package, or data subscription.
A repeatable way to gather, test, and update evidence as seasons and operating conditions change.
Start small
Bring us one decision.
Tell us the water, the call, when it has to be made, and what being wrong costs. You do not need a dataset. Working out whether the question is answerable is part of the job.
A useful first message fits in a few sentences.