Data model
Impact Map extracts what organizations actually do from their public materials (websites, program pages) and IRS 990 filings, then structures it into three layers: landscapes, taxonomies, and observations.
Landscapes
A landscape is a named set of organizations — a region, a funder's grantee
universe, or any other meaningful cohort. Membership is many-to-many: the same org can
belong to several landscapes. Every API query and MCP tool call resolves to exactly one
landscape (pass landscape=<slug>, or omit it for the default), so
results are always scoped to a defined universe rather than "everything we've ever
indexed."
Activity taxonomy
Each landscape has an activity taxonomy: a three-level tree of activity groups derived bottom-up from the organizations' actual activities — not from self-reported NTEE codes.
| Level | Meaning | Example |
|---|---|---|
2 | Sector — the broadest rollup | Education & Youth |
1 | Activity area | Youth mentorship & development |
0 | Program type — the most specific | School-based peer mentoring |
Groups carry a stable_id (e.g. group-8218f178709b), a label, a
description, keywords, and rollup counts. Taxonomies are versioned artifacts:
a taxonomy rebuild produces a new taxonomy_id rather than mutating the old one,
and every API response echoes the resolved taxonomy_id in meta so
results are reproducible across data releases.
Observations
Organizations carry typed observations extracted from evidence in their own materials:
| Type | What it captures |
|---|---|
| Activities | Concrete things the org does, each typed as direct_service / advocacy / research / capacity_building and assigned to activity groups. |
| Strategies | Named approaches ("housing-first model", "peer-led organizing") — grouped across orgs so you can see which approaches a field shares. |
| Populations | Who is served, with geography where stated. |
| Funding | Funders the org explicitly names, with funding type. |
| Relationships | Partners and coalitions the org publicly claims — a real, non-inferred network layer. |
An observation appearing here means the organization said it publicly — absence is not evidence of absence, and funder mentions are the org's claim, not a verified grant record.
990 financials
Where an organization files a Form 990, the org profile
(GET /api/v1/orgs/:id) joins the modeled IRS warehouse: revenue and expense
composition, balance-sheet liquidity, IRS-reported programs, and officer compensation for
the latest available tax year.
Provenance
Extracted rows are stamped with the pipeline version that produced them, and taxonomies
record how and when they were built. If you need to reconcile results across time, pin the
taxonomy_id from meta.
Caveats
This is a research preview. Extraction is AI-assisted and imperfect; coverage varies by landscape (some orgs have rich websites, some only a 990). Treat counts as strong signals, not census figures.