Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.pathprotocol.finance/llms.txt

Use this file to discover all available pages before exploring further.

Signal Mesh is Path’s prediction engine. Every 15 minutes it ingests ~117 signals across six categories, fans them into a multi-horizon model stack, and emits confidence-scored predictions per pool.

​
Signal inputs

On-chain protocol health

Utilisation curves, supply and borrow caps, oracle staleness, reserve factors, kink slopes, interest-accrual fragility.

TVL flow + cohort behaviour

Institutional vs. retail cohort flow, 24h / 7d / 30d TVL deltas, address-level rebalance event clustering.

Prediction-market + sentiment

Prediction-market-derived risk premia, sentiment on protocol-specific tickers, governance-vote dispersion.

Macro indicators

Stablecoin supply trajectories, central-bank rate prints, funding rates, category-level capital flow.

Per-protocol specialists

Mechanism-specific signals like AMM-yield discount evolution, vault-level curator action, parameter-drift detection, governance activity.

Cross-protocol correlation

Pair-wise APY co-movement matrix, regime-flip detection, leader– lagger graph from a 60-day rolling window.

​
Model stack

The active stack publishes predictions on three horizons:
Sub-day directional signal used for tight-cadence rebalance triggers. Sign-based accuracy, statistically significant against the random-walk baseline at standard test thresholds. Best for pools where utilisation changes quickly and rebalancing cost is low.
Primary directional model. The number we publish most prominently in product surfaces. Calibrated for confidence-score interpretability; outputs are usable as both a binary trigger and a confidence-weighted scaler.
Three-class output: trending up, trending down, stable. Used for long-horizon allocation context and to flag pools where the 1-day signal is noise around a stable mean rather than a real directional move.
In addition to the three directional heads, Signal Mesh emits:
  • A continuous APY estimator for the 7-day forward horizon, decomposed into base yield, reward yield, and demand components per protocol family. The decomposition is what lets the estimator handle protocols with structurally different rate mechanics—lending, AMM-style fixed yield, and governance-set rates—without forcing them through a single feature set.
  • A per-mechanism specialist for AMM-style yield where the underlying pricing is mechanically different from utilisation-curve lending. Specialists score ~20 percentage points higher than the generalist on their own protocol family on test data.
  • A hybrid composite score combining the directional output, the continuous estimator, native risk signals, and liquidity-impact context. The hybrid score is what drives the front-page allocation ranking.

​
Retrain cadence

The directional models retrain on a three-day cadence using a rolling window. A live-accuracy gate on every active model runs every 10 minutes and auto-deprecates any model that falls below the public DeFiLlama baseline on a 100+ sample window. The gate auto-files a ticket and pages the team on failure—calibration drift is one of the few classes of bug that can ship silently in a model pipeline, and we want that loop closed.

​
How predictions surface

Every prediction is published with:
  • The horizon (6 hours, 1 day, 7 days, 30 days)
  • A direction class (up, down, stable)
  • A calibrated confidence score (0–100)
  • A continuous APY estimate
  • A liquidity-impact context: “you would represent X% of this pool’s TVL with a $Y deposit”
The last item is the answer to the question every allocator asks first—“if I move size, do I cannibalise the yield I am chasing?”—and is documented in detail under Liquidity impact.