FL-BSA Technical Whitepaper

Demo and synthetic boundary

Using a fully synthetic demonstration portfolio, the whitepaper walks through dual-branch simulation: an amplification branch that retains configured outcome/protected strata, and an intrinsic policy control that excludes outcome aliases during fitting and applies a declared equal-rate post-label policy. The intrinsic comparison does not estimate a causal counterfactual and is not an optimized lending policy.

Everything in this whitepaper was generated from synthetic demonstration data. It contains no customer data and no real borrower records.

It is published as a public technical characterization prerelease (v5.0.8-technical-whitepaper-20260930). Its disposition remains characterization_only; it is not eligible as customer or promotion evidence and does not establish production utility, certification, a legal determination, regulator approval, general availability, or a public Marketplace listing.

The verification companion is a bounded seven-member ZIP with the exact public source record, build decision and correction record, evidence summary, robustness table, original producer intake, manifest, and offline verifier used to check the technical publication.

Verify it yourself

Verification is offline: no Equilens service is involved.

1. Download the PDF and verification ZIP from the pinned release above.
2. Recompute SHA-256 checksums and compare them against SHA256SUMS.txt from the same release.
3. Extract verify_public_technical_companion.py and run it against the downloaded companion ZIP. It verifies the bounded archive, hashes, source identity, metric availability, controlling comparisons, robustness rows, and claims flags offline.

Evidence bundles produced by a deployed appliance ship with a documented hashing spec and offline verification helpers; see the Trust Center for the evidence-chain overview.