Roll up loss adjuster reports after a major weather event

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

1. 1

   Planner, while planning

   Split or keep serialYes or no, with a probability

   While planning

   Are there enough reports in this region, across distinct damage types, that reading them in parallel groups beats one pass?

   Sees only: Report counts by region and stated damage type

   Why: Thinly hit regions are cheaper handled in one pass.

   00.40 and 0.601

   • Yes: 0.60 or higher→thenSplit the work
   • Unsure: 0.40 up to 0.60→thenKeep serial
   • No: below 0.40→thenKeep serial
2. 2

   Split or keep serialYes or no, with a probability

   While planning

   Does this adjuster report cover several buildings or locations that should each be extracted as a separate task?

   Sees only: One report's table of contents and location list

   Why: Multi-location reports muddle small workers unless split at the location.

   00.40 and 0.601

   • Yes: 0.60 or higher→thenSplit the work
   • Unsure: 0.40 up to 0.60→thenKeep serial
   • No: below 0.40→thenKeep serial
3. 3

   Scope checkYes or no, with a probability

   Before work starts on a unit

   Do the loss date and location in this report fall within the event's stated dates and regions?

   Sees only: The event definition and one report's header fields

   Why: Keeps unrelated losses out of the event picture used for reinsurance notices.

   00.30 and 0.601

   • Yes: 0.60 or higher→thenAccept
   • Unsure: 0.30 up to 0.60→thenAsk a person
   • No: below 0.30→thenSkip this unit
4. 4

   After workers, the judge checks

   Evidence checkYes or no, with a probability

   After a worker answers

   Does the quoted sentence from the adjuster's report state the cause or damage description that was extracted into this guideline field?

   Sees only: One extracted guideline field and the quoted report sentence

   Why: The event picture is only as good as each field's link to what the adjuster wrote.

   00.50 and 0.851

   • Yes: 0.85 or higher→thenAccept
   • Unsure: 0.50 up to 0.85→thenEscalate to a strong model
   • No: below 0.50→thenReject and retry
5. 5

   Reconciler, while merging

   Conflict checkYes or no, with a probability

   While reconciling

   Do these two passages from the same adjuster report give different causes or damage descriptions for the same location?

   Sees only: Two quoted passages from one report

   Why: A self-contradicting report is kept out of totals and returned to the adjuster.

   00.30 and 0.701

   • Yes: 0.70 or higher→thenMark unresolved
   • Unsure: 0.30 up to 0.70→thenMark unresolved
   • No: below 0.30→thenAccept
6. 6

   Accountable person, before anything is settled

   Person decidesYes or no, with a probability

   Before anything is reported as settled

   Does the adjuster's text raise a coverage question, or is the report incomplete or self-contradicting on a field the handler relies on?

   Sees only: One report's extracted fields, gaps and flags

   Why: Each file stays with its handler; the run only extracts and flags.

   Accountable: The catastrophe claims manager owns the event picture; each file stays with its handler.

   00.10 and 0.301

   • Yes: 0.30 or higher→thenAsk a person
   • Unsure: 0.10 up to 0.30→thenAsk a person
   • No: below 0.10→thenAccept