> For the complete documentation index, see [llms.txt](https://docs.everesteer.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.everesteer.ai/research-guide/what-the-platform-rewards.md). # What the platform rewards What CORR, AIMC and NCORR each pay for, and what that means for a model. The round score is an arctan of the sum of three equally weighted terms: CORR, AIMC and NCORR. All three are covariances with the centred target. Call `explain_scoring` (SDK) or `GET /api/v1/scoring` (REST) for the live formula, scale and bound. ## CORR CORR is the covariance of your rank-gaussianised predictions with the centred target, per exped, then averaged. **What CORR pays for:** ordering. Only the ranks of your predictions matter, so scale and monotone transforms do not affect your score. Rank-gaussianising spreads the extreme ranks further from zero than the middle ones, so getting the most extreme instruments right on each cross-section counts for more than getting the middle right. ## AIMC AIMC is the covariance of the centred target with your predictions after removing the component that lies along the designated benchmark. It rewards predictions that differ from the benchmark while being correct. **What AIMC pays for:** being different and right. Reproducing the designated benchmark scores near zero, however high your CORR. The benchmark is the column `v1_sherpa` in the benchmark parquet files downloaded for each split. The AIMC reference is the fixed designated benchmark in every lane. On an exped where the benchmark's own covariance with the target is negative, AIMC is 0 for that exped. ## NCORR NCORR is computed exactly like AIMC, with the equal-weight average of a frozen core feature set standing in for the benchmark. The core set is train-selected and fixed. On an exped where that average's own covariance with the target is negative, NCORR is 0. **What NCORR pays for:** signal beyond the core features taken together. A prediction that tracks the equal-weight average of the core features scores near zero on NCORR, even if it scores well on CORR. The point is that a score should not come purely from features the platform already knows about. ## The round score The round score is `b * arctan((CORR + AIMC + NCORR) / b)`, bounded per round, and payout is stake times round score. The boards rank on the round score, not on CORR alone. A board falls back to ranking on CORR only when nothing on it is scored yet. `rank_metric` on the leaderboard response reports what a given board was actually ordered by. ## Why a CORR-only model leaves score on the table AIMC and NCORR carry the same weight as CORR. A model tuned on CORR alone ignores two of the three terms. A model that reproduces the benchmark with high CORR scores near zero on AIMC. A model that tracks the average of the core features scores near zero on NCORR. Predictions that are right and differ from both references score on all three terms. ## Lower-turnover models tend to score better over time This is an empirical observation from the live tournament, documented in the public example-scripts repository. A model that produces stable predictions across rounds tends to accumulate a higher total score than one that changes sharply each round. ## What is display-only The diagnostics panel shows several metrics that are informational and not part of the round score: * Sharpe ratio (annualised information ratio of per-exped CORR) * Standard deviation of per-exped CORR * Feature exposure (max absolute correlation between predictions and any feature) * Maximum drawdown of per-exped CORR * Autocorrelation of per-exped CORR These are available via `get_validation_panel` in the SDK and `format_diagnostics_table`. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.everesteer.ai/research-guide/what-the-platform-rewards.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.