02 · Assess — For banks, insurers, and anyone holding Philippine property risk
Project Bedrock
The flood map, down to one front door.
Bedrock scores the address itself — what floods it, what shakes it, what it is like to live there, and an expected loss with its uncertainty attached. One lookup, or a whole loan book overnight.
- address, not a province
- 1
- numbers, kept separate
- 3
- uncertainty on every loss
- ±
- for a whole loan book
- Overnight
The problem
BSP Circular 1085 asks a question the hazard maps cannot answer.
Philippine banks are now required to identify and manage climate and environmental risk across their portfolios. The obligation is specific. The data available to meet it is not: national hazard layers are published at a resolution that describes a municipality, and a mortgage is written against one address on one side of one street.
So the gap gets closed by averaging — a provincial flood category stamped onto every property inside it. That is wrong in both directions at once. It overstates the risk on high ground the bank could have lent against, and it quietly understates the house at the bottom of the catchment, which is the one that defaults.
Bedrock does the join nobody had done: national hazard datasets and elevation, matched to a specific address, run through published engineering damage curves to produce an expected annual loss. Three numbers come back — hazard, livability, and loss — and they stay three numbers, because collapsing risk and desirability into one score produces something that looks decisive and means nothing.
We also cap the precision. An expected-loss figure carried to four decimal places implies a confidence the underlying elevation model does not have, and a disclosure built on it inherits the false confidence. Bedrock reports at the precision the data actually supports and shows where that limit sits.
How it works
Four moves, end to end.
- 01
Locate
Resolve the address to a parcel and its footprint on the ground.
- 02
Expose
Join PHIVOLCS, UP NOAH and HazardHunterPH layers against GLO-90 elevation.
- 03
Price
Run JRC depth-damage curves by structural class to an expected annual loss.
- 04
Report
Hazard, livability and loss — three numbers, at the precision the data supports.
- Flood
- 8.2 High
- Seismic
- 4.6 Moderate
- Landslide
- 0.9 Low
Hazard scoring
This address, not the province.
Flood, seismic, and multi-hazard exposure resolved to a specific address, drawn from PHIVOLCS, UP NOAH, and HazardHunterPH and joined against GLO-90 elevation. The output distinguishes between two houses four hundred metres apart, because the flood does.
Average annual loss
A loss number that survives an audit.
Expected annual loss built from JRC depth-damage curves and structural class, reported with its uncertainty rather than a false decimal tail. Every step is a published method you can hand to a regulator and have them follow it back to source.
8.2/10
Hazard
6.4/10
Livability
Livability scoring
Deliberately kept separate.
What it is like to live at an address is a different question from what threatens it, and the two answers move independently. Merging them into a single number destroys the information a credit committee actually needs — so we keep them apart and let you weigh them yourself.
// one address, at origination { "address": "Riverside lot, Naga City" } // three numbers back, kept apart { "hazard": 8.2, "livability": 6.4, "aal_php": { "mean": 38000, "pm": 11000 } }
API and batch pipeline
One address, or a book.
Score a property on demand at origination, or run an entire portfolio overnight for a disclosure cycle. Same methodology on both paths, so a spot check and a stress test cannot disagree with each other.
Where this stands
In development. The methodology is written down, including where it stops being precise. A backtest against Typhoon Kristine (2024) is running, and we will publish it whether or not it flatters us.
Built on
Push on Project Bedrock. We answer the hard ones ourselves.
Questions go to the engineer who built the part you’re asking about — not a sales sequence.