Project · forecasting / competition

Numerai Unstaked Track Record

A reproducible, unstaked Numerai Classic model that completes scored weekly rounds to establish a verifiable track record before any capital is committed.

Boundary: This is a track-record building exercise, not a trading system. It risks no capital, holds no positions, and submits only free unstaked predictions to Numerai. No staking, no NMR at risk, no live money decisions.

Venture Status (live from ledger)

Track: forecasting · Stage: hold PAUSED

Metric: mean_mmc_vs_example = 0.0

Kill criteria: Mean MMC below 0 after 8 scored unstaked rounds; stops.

Days alive: 7.0 / 56 minimum · Iterations: 3 / 4 minimum

Review in: -0.7 days

Last iteration

Changed: baseline predictions generated for current round (6923 rows) -> predictions ready for submission

Result: predictions/current.csv written with 6923 rows, min=0.00014445, max=1.00000000

Owner: quant · Jurisdiction: none-needed for free unstaked submissions; staking requires ' payment decision

Data from rodion scoreboard --json; snapshot at build time UTC.

Competition context

Numerai Classic: Weekly tournament on encrypted financial data. Free unstaked submissions are open to all; staking NMR requires a separate payment decision.

Goal: Complete at least 8 scored unstaked rounds with non-negative mean MMC versus the official example model.

Current status: Baseline predictions generated for current round (6923 rows). Local validation pipeline ready; submission format verified.

Method

The baseline implementation uses a deterministic model trained on Numerai's historical training data. It produces weekly predictions in the required format (CSV with prediction column). All code is pure Python with no hidden state; any recorded forecast can be re-computed from the competition data.

Local validation: Prediction distribution checked (min=0.00014445, max=1.00000000). Submission format verified against Numerai schema.

Sources and verifiability

Numerai platform: numer.ai (public)

Local repo: The tracked predictions are in the project repository.

All inputs are provided by Numerai; the algorithm is deterministic. Any recorded forecast can be independently re-computed.

Verification

Run the local validation suite. Inspect predictions/current.csv for the latest predictions. Re-check any scored round against Numerai's published results. The evidence boundary is the competition rules and provided data; no private data or capital is used.