Sampling
Where to sample first
Replay a proposed ranking on past monitoring records before considering it for a real route. A replay of already sampled sites is not proof about unsampled water.
Water warning lab · Monterey Bay
Test which water
to check first.
We test whether the monitoring records you already collect can support one clear sampling or warning decision. Each test names its place, time window, comparison rule, and evidence limits before a public claim is considered.
Most monitoring programs cannot sample every site every day. We test whether one specific decision improves on a simple rule — and say when it does not.
Early warning
Test the warning time
Test one precisely defined future event at a real calendar lead time. If the data support only same-day detection, we call it detection, not warning.
Signal checks
Does it improve this held-out test?
Compare a new measurement with a simple rule on observations kept out of model fitting. That tests one defined decision; it does not prove a universal cause or law.
What the current evidence does not prove
Not route proof
Existing evaluation ranked observations that had already been sampled. It cannot establish what an alternative California route would have found at unsampled sites.
Not a calendar warning
The Florida analysis steps through later observed samples, while missing weeks remain missing. It does not establish a fixed warning time or distinct new outbreaks.
Not a fair comparison
The Ireland scoring used earlier Irish monitoring results, and published methods were not rerun on the same records. It does not support a no-local-data or best-published-method claim.
Read the machine-readable correction record or browse selected research reports and negative results.
“Corrections stay visible.”
Knowing what the evidence cannot support is part of the result.
How we work
01 — Baseline
It has to beat the simple rule
We define the simple comparison before interpreting a model. If the model does not improve the decision, the result stays negative.
02 — Red-team
Someone tries to break it
Headline claims receive a separate skeptical review before publication. Claims that do not survive are narrowed or removed.
03 — On the record
We show the correction
Withdrawn sales claims remain in a structured correction record, with the evidence paths and reasons they were removed.
Research across several kinds of water.
- Beach & estuary bacteria
- River & stream monitoring
- Lake & reservoir algae
- Shellfish toxins & harmful algae
- Low-oxygen water
- Marine heat & wildlife observations
Start a pilot
Start with one bounded question.
Pick one water system and one decision. First define what would count as useful evidence, what simple rule must be beaten, and what the available records cannot answer.
- 1Pick one decisionWhere to sample, what to watch, or a signal to check.
- 2Share your dataUse what you already collect; document what is usable and what is missing.
- 3Agree on the testSet the target, comparison rule, place, and time split before reading the result.
- 4Read the decision recordSee what held, what failed, and whether any next step is justified.
Tell us about your water.
Describe one water system, one decision, and roughly what records you hold. We will review whether a bounded test is possible and what evidence would still be missing.