detent
Ecosystem/08 of 08

Insurance and the Loss Dataset

Why it is in the ecosystem. The end of the assurance chain: insurable. Insurers currently price agent risk with no loss data at all; the de-identified labelled corpus is the accident book they will price from.

pipeline REALvolume SIMULATED SCALEwho pays: insurers and underwriters, data licence
SIMULATED SCALEVolume-gated.

This becomes real with fleet customers. Partners deliver the policies; DETENT delivers the data. What is below is the real pipeline run at demo volume: de-identified aggregates from the testnet corpus, no owner, no address, no amount.

What an insurer sees

Labels in the bookSIMULATED SCALE
reading the corpus
source · real pipeline, demo volume
Confirmed lossesSIMULATED SCALE
reading
source · the only label that costs money
Loss shareSIMULATED SCALE
reading
source · confirmed_loss over all labels
Attack classes seenSIMULATED SCALE
reading
source · rules-tier classes on labelled receipts

reading the corpus

Losses by label type

SIMULATED SCALE
count per outcome label · real pipeline, demo volume

the axes are real and the bars are empty. tour 2 fills the first one.

By attack class

SIMULATED SCALE
the rules tier's class on each labelled receipt

Over time

SIMULATED SCALE
labels per day · the 0 most recent

What the extract holds

in the file
  • counts by outcome label and by attack class
  • labels per day over the window
  • the five-label taxonomy, so a reader can check the columns
  • the scale tag, in the file, not only on the page
not in the file
  • no owner, no agent key, no address
  • no amounts: receipts do not carry them
  • no receipt hashes: aggregates only
what it is for

Insurers price agent risk today with no loss data at all. This is the shape of the first dataset that could change that: an accident book where every entry traces to a signed receipt and a stamped outcome. The shape ships now. The volume comes with fleets.

the aha: This is the first loss book the agent economy has ever had.