Risk scoring for high-impact cases

A usable risk score names the factor that put the case on the high-impact list. For this authorize step, that factor is one of three: hazard class, proximity, or financial exposure. If none of those is present in the application, the score field stays empty. An empty field is a result, not a gap to fill.

The score is not the decision. A human still signs. The model surfaces the cited factor so the authorizer can see why this file cannot be treated as ordinary. The authorizer then applies the same statutory and policy tests they would apply without a model.

A rank of high with no hazard class, no proximity trigger, and no exposure figure is a reason to send the file back, not a reason to sign. Ranking without a cited factor does not give an authorizer anything they can defend.

Permit and license records often live in Accela, Palantir, Tyler, or Microsoft systems. The vendor does not change the rule. If the record cannot point at a factor the table already defines, it is not scored.

Before scoring, confirm the file is complete enough to read. An incomplete application cannot support a cited factor. Use the application completeness checker when required fields or attachments are missing. Do not invent a factor to keep the queue moving.

Open the application and the factor table together. The table is the only source of triggers. Do not browse similar sites, prior years, or neighboring parcels looking for a risk the current file does not contain.

Read the application for the three factor types the table recognizes. Hazard class is the classification the applicant declared or that a required attachment already states, such as a dangerous-goods class on a storage license. Proximity is a measured relationship the file already records, such as distance to a school, water intake, or protected habitat in the application materials. Financial exposure is a figure the file already states: a bond, a throughput value, a guaranteed amount, or another exposure the form requires.

Match what you read to rows in the factor table. A match is a cite: the factor name, where it appears in the application, and which table row it satisfies. No match means the score stays empty.

A bulk-fuel storage license lists flammable liquids by hazard class and records the tank farm inside the proximity buffer for a school. The table has a row for that class combined with that buffer. The score cites both. A food-handler license on the same block lists neither a hazard class nor a proximity trigger nor a financial-exposure figure the table uses. That score stays empty. The authorizer still signs the food-handler file under ordinary rules. They do not receive a residual-risk percent, and they should not ask the model for one.

Do not derive an environmental risk from a similar site. The neighboring tank farm is not evidence in this application. Leave the score empty if this file has no proximity or hazard-class cite.

Write the score only if at least one factor cites. Stop when the cites are recorded. Do not add a composite rank, a residual-risk percent, or a narrative drawn from other files.

The score record should include the factor type, the value as it appears in the application rather than a paraphrase that upgrades it, the table row or rule identifier that treats that value as high-impact, and the reason this file is queued for mandatory human review.

If two factors cite, list both. Do not average them into a single unexplained rank. If the table requires a combination, such as class plus proximity, cite the combination. Do not fold it into a blended environment score the table never defined.

If no factor cites, leave the score empty. Do not substitute a percentile, a residual-risk percent, or a sentence about incidents at similar facilities. Those statements are not in the file or the table. They are not a score. Ordinary permits and licenses that never trip these factors should travel with an empty field. A populated score on an ordinary file trains reviewers to ignore scores.

Evidence in the file may still be thin even when a factor cites. Scoring is not the same as deciding that the evidence is enough to approve. When the cite exists but the supporting materials do not meet the standard of proof the statute requires, route through the evidence sufficiency assessor before sign-off, with the cited factor still visible.

Mandatory human review is the point of the score. The authorizer reads the cited factor, checks it against the application and the table, then applies the approval tests. The signature is the decision.

Confirm the cite is in the file. If the score names a hazard class the attachments do not show, reject the score and return the file. Confirm the table row applies. If staff used a row for a different program or a stricter buffer than the adopted table, correct the score or clear it. Then apply the ordinary legal tests: completeness, conditions, inspections, and any hearing or consultation the program requires. Sign, refuse, or return. The score does not pick among those three.

A high score is a queue label. It is not an approval, a denial, or a set of conditions. Conditions, if needed, are written after the factor is understood. The condition generator for approvals can draft conditions that attach to the same cited factor. It does not replace the signature.

A proximity or hazard-class cite is often the reason a first inspection is not optional. Hand that cite to the risk-based inspection scheduler rather than inventing a new risk story at the inspect stage.

Three failure modes produce files that look scored and are not usable.

A rank with no factor. The model returns high, medium, or low, or a numeric rank, without naming hazard class, proximity, or financial exposure. The authorizer cannot check the rank. Downstream staff cannot write a condition or schedule an inspection from it. Treat an unexplained rank as an empty score. Send it back to be rescored with a cite, or clear it and process the file as ordinary if no factor is present.

Treating the score as the decision. Staff auto-approve empty scores and auto-refuse high scores, or they paste the rank into the decision letter. The statute still requires a human sign-off. An empty score means no high-impact factor was cited, not that the activity is safe. A cited high-impact factor means review this, not deny this.

Inventing an environmental risk from a similar site. The model, or a reviewer filling a blank, imports contamination, flood, or habitat issues from a nearby facility, a prior operator, or a look-alike application. That is not a cite. If this application has no environmental factor in its own materials, that portion of the score stays empty. Staff can still require studies through a request for information. They cannot backfill a score to justify the request.

A filled field feels like progress when queues are long. An empty field on a sensitive street feels like neglect. The quality bar is the opposite: a score that cites the factor, and emptiness when the factor is absent.

The cited factor should survive the authorize step unchanged. Do not convert it into a residual-risk percent for the file jacket, the public register, or a dashboard. Percents the table does not produce are invention. They will be quoted later as if they were findings.

If the application changes after scoring (a new tank, a new bond, a new site plan), reload the application and the factor table and score again. A stale cite is as unusable as a rank with no factor.

Accela, Palantir, Tyler, and Microsoft systems can hold the record, the queue, and the signature. None of that substitutes for a cite the authorizer can read. Configure the authorize queue so a populated score without a factor cannot reach sign-off, and so an empty score on an ordinary file does not block it.