Status: needs a decision from us. Not blocked on anyone; not a defect.
The situation
|
value |
2025 Assessment flood source |
First Street Foundation (2020) |
| our source |
FEMA NFHL, SFHA_TF='T' |
| 100-yr floodplain conserved, assessment |
14% |
| 100-yr floodplain conserved, ours |
16.1% |
Two defensible national floodplain products; a 2.1 pp difference in the answer. Nothing is broken — we simply do not serve the layer the assessment used, and the app does not say so.
Options
- Keep FEMA NFHL and document the difference in
system-prompt.md, so the agent attributes the source correctly and does not present 16.1% as reproducing the assessment. Cheapest, and keeps the app on an open, redistributable source.
- Ingest First Street 2020 for exact parity. Previously deferred on access/licensing grounds, which have not been revisited since.
Recommendation: option 1, unless a partner specifically needs report parity on this row. First Street's licensing makes it a poor fit for a public app, and the honest framing ("we use FEMA NFHL; the assessment used First Street; here is the difference") is more useful to users than a number that silently matches.
Done when: either system-prompt.md carries the attribution note (option 1), or a focused ingest issue exists on data-workflows with the licensing question resolved (option 2).
Status: needs a decision from us. Not blocked on anyone; not a defect.
The situation
floodsourceSFHA_TF='T'Two defensible national floodplain products; a 2.1 pp difference in the answer. Nothing is broken — we simply do not serve the layer the assessment used, and the app does not say so.
Options
system-prompt.md, so the agent attributes the source correctly and does not present 16.1% as reproducing the assessment. Cheapest, and keeps the app on an open, redistributable source.Recommendation: option 1, unless a partner specifically needs report parity on this row. First Street's licensing makes it a poor fit for a public app, and the honest framing ("we use FEMA NFHL; the assessment used First Street; here is the difference") is more useful to users than a number that silently matches.
Done when: either
system-prompt.mdcarries the attribution note (option 1), or a focused ingest issue exists ondata-workflowswith the licensing question resolved (option 2).