ENFR Start with your trigger

Method

How I compute a trigger, written down so you can check it.

A calculation nobody can rerun is a matter of trust, not independence. This is the open method behind every number on this site: each report shows its sources, dataset versions and hashes.

  1. 01

    Data in

    Sources named in the contract (NOAA, ERA5, IMERG, Sentinel). Pulled on schedule, with the fallback source agreed in advance.

  2. 02

    Checks

    Every input checked for coverage, units, gaps and known dataset defects. Every file hashed.

  3. 03

    Index

    The index exactly as the policy wording defines it. The same code runs the backtest and the live calculation.

  4. 04

    Trigger

    Tested against the payout ladder, cell by cell or district by district.

  5. 05

    Payout

    A payout report for each cell: value, tier, amount and the data used.

  6. 06

    Audit trail

    Every figure traced to input hashes and code version. A rerun gives the same answer.

Rules I hold to

Seven rules, the same for every client.

  1. 1

    The wording is the specification.

    The code computes what the policy says. Where the wording is ambiguous, I write the ambiguity down and get it resolved before the season, not after an event.

  2. 2

    One code path for backtest and live.

    The index that was backtested is the index that pays. No separate script for production.

  3. 3

    Every input is checked and hashed.

    Coverage, units, gaps and known defects are tested before a value enters the index. The hash ties each figure to the exact file it came from.

  4. 4

    Fallbacks are agreed in advance.

    If the primary source is late or broken, the replacement is already named in the contract. Dataset versions are named too: CHIRPS v2 production ends after December 2026, so rainfall indices built now use CHIRPS v3.

  5. 5

    Thresholds live in the index’s own units.

    A tier is “the 1-in-10 event for this cell, as this instrument measures it”, never a claim about the atmosphere.

  6. 6

    Limits are reported, not hidden.

    Where no loss series exists, payout against loss is reported as not scored. Nothing is estimated to fill the gap.

  7. 7

    A rerun gives the same answer.

    Same inputs, same code version, same result. That is the test the audit trail exists to pass.

See it applied

Five calculations follow this method end to end.

Open the showcase