INDUSTRY FIELD GUIDE
Wastewater process events for AI incident review
A process assistant can be tested on whether it distinguishes a failed instrument from a real change in a treatment stage.
NAICS 221320 · Sewage Treatment Facilities
A process assistant can be tested on whether it distinguishes a failed instrument from a real change in a treatment stage. NIST’s industrial control testbed research shows how controlled processes can support measurement and testing. EPA separately addresses security practices for water and wastewater systems. Use an isolated example before evaluating any operational record. NIST industrial control system testbed · EPA water sector security guidance
Build a paired sensor and lab observation
Use invented measurements for a treatment training scenario. Set one case where an online reading changes but a confirmatory measurement stays stable, and another where both change. Include sample timing, method, units and the documented reason an operator escalates. A model should identify the difference and request more evidence when the records conflict. The evaluation can measure recognition and escalation without giving a model control authority.
Keep operating information in the approved environment
Use a simulation or training rig that cannot touch production systems. Exclude site coordinates, process setpoints, access details, system diagrams and sensitive incident records from the first sample. Verify who can authorize operational information and what the intended use permits.
Send a fabricated event timeline and evidence rubric. Distinguish simulated examples from authorized real data in the description. Describe a dataset
Original Fiund exercise
Compare invented observations without diagnosing their cause
A detached paper scenario provides artificial online and confirmatory values. It tests evidence consistency and review needs, with no process-control authority.
| Record | Source | What it records |
|---|---|---|
| CASE-W1 | First simulated interval | An online demonstration value changes from 4 to 8. A separately documented confirmatory observation remains 4 under the stated matching interval and unit. |
| CASE-W2 | Second simulated interval | The online value changes from 4 to 8, and a matching confirmatory observation is also 8. |
| CAUSE-W3 | Available evidence note | Neither case includes a verified instrument-fault finding or a documented process cause. |
Question for the model
What difference is supported between the two evidence patterns, and can either cause be diagnosed from the packet?
Show the reference answer and checks
W1 contains disagreement between the supplied online and confirmatory observations; W2 contains agreement under the stated matching basis. That pattern does not by itself establish an instrument failure or a process cause. W3 explicitly supplies neither finding. Report the disagreement or agreement, cite the source and preserve the need for qualified review rather than infer a control response.
- Calling disagreement definitive proof of a failed instrument.
- Treating agreement between two values as proof of a particular process cause.
Fields to preserve
- Sample interval and unit
- Check that compared observations refer to the same stated basis.
- Measurement origin
- Keep online and confirmatory sources distinct.
- Verified cause status
- Preserve whether an independent finding was actually supplied.
Test another case
Shift the confirmatory sample to another interval. The evaluator must identify the timing mismatch before comparing its values.
All values are invented and disconnected from equipment. Exclude setpoints, diagrams, coordinates, credentials, real incidents and any production-system access.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.