87ns
Per-record sensitivity propagation
Measured against a 4-arg typed DAG, warm cache, single-record perturbation. Reproducer: bench_compile_50_warm.py at commit 46a6328. Run it yourself: python bench_compile_50_warm.py.
bench · commit 46a6328 →Proof
Every number on this site
points to an artifact.
When the Heartbeat says 87 ns per sensitivity, this is the file the number comes from. When we say bit-identical, this is the test that checks it. When we say 100/100 rubric, this is what the rubric grades on.
This is what computation provenance means in practice: deterministic compilation plus a re-derivable record of how every number was reached. It is the architectural property at the core of Financial Reasoning Infrastructure.
Artifacts
4,000drivers
Largest compiled DAG in the benchmark suite
Cold + warm compile sweep across 50 / 200 / 500 / 1000 / 2000 / 4000 drivers, each step validated as a clean compile — every driver valued, zero errors. The 4,000-driver case: 7.7ms cold / 1.42ms warm. Synthetic model — not a production graph. The production NDO graph is 3,657 drivers. Reproducer: bench_scale_warm.py at commit 729aeb1, re-run clean 2026-06-05.
bench · bench_scale_warm.py · commit 729aeb1 · re-run 2026-06-05 →100/ 100
Institutional Excel-export rubric
Six dimensions, 100 points total, 90-point institutional pass threshold. Production exports score 100/100 on the SaaS, DTC, and LBO institutional test models.
rubric · six-sheet test →bit-identical
Same IR always produces same outputs
The compile pipeline is a pure function over the IR: same input, same output, every machine, every run. Verified across macOS arm64 and Linux amd64 in CI. No floating-point drift between cold and warm caches. Test: tests/test_determinism.py.
determinism · CI matrix →bothday one
MCP + REST surfaces
Durable per-call REST at api.flatlandfi.com/api/v2 · local MCP bridge (flatland-client mcp) over the same engine, two protocols. Your model is a file you own; the server holds nothing at rest.
API reference →50answers to start, free
Free tier
50 answers per API key to get started. No card required. One free API key per email address. After the 50, add a card and pay $0.10 per answer.
pricing detail →single-digit ms warm
Compile latency
50-driver compile, warm cache, runs in single-digit milliseconds on commodity hardware. Cold compile takes longer while the parser and type system warm up. Reproducer: bench_compile_50_warm.py at commit 46a6328.
bench · bench_compile_50_warm.py · commit 46a6328 →✓by architecture
Tenant isolation
Per-key authentication boundary. Each API key is a logical tenant. Cross-tenant access is architecturally impossible; no shared model store exists that two keys can both read from.
security posture →Flatland · index of everything
For
Learn
Comparisons
Disasters
© 2026 Flatland · Financial Reasoning Infrastructure · live pulse · awaiting first compile