Unreleased Python drift-governance library for ML ensembles
Nobody outside the owner uses this yet, and it has no revenue. It is listed on the strength of what was built, which passed Indiemaker's review before it went live. Read the owner's account of the work below before you weigh the asking price.
This listing has screenshots and a cover image. Create a free account to see how many.
What this is
Pitch
A fully tested, never-released safety kernel for ML ensembles and multi-agent systems, with documented benchmarks and honest limits.
A Python ML and agent-governance library that has never been publicly released. It computes omega (Ω) — an expert-disagreement drift signal — and uses it to govern a live ensemble: an adaptive controller with a Dirichlet entropy floor, a regime manager with hysteresis, ColdStartGuard, a safety-interlock trip, a RecoveryProtocol, a regressor, an AgentGovernor for multi-agent setups, and full telemetry. Benchmarks (honest, reproducible): Scenario A −85.6% MSE over the full window and −91.9% in the second half versus the same ensemble running static; the interlock fired in 20 of 30 batches.
Sign in to read the full description, the key features, and what the seller says about running it.
See it for yourself
- Live product
- Sign in, then unlock, to open it
- Walkthrough
- There is a walkthrough on file
It plays here once you unlock.
Financials
Create a free account to see the revenue breakdown, the expenses and the trend.