A live 0-100 index of the physical supply-demand balance behind each segment's rates. Not a price forecast — direct rate levels don't beat random walk over any horizon we have. This is the pressure that creates them: how much open tonnage there is, how much cargo it's chasing, and how hard owners can push.
Where we can prove out-of-sample skill on top, we attach a probability the rate moves >15% in two weeks. Where we can't yet, we say so — the model waits for history, not opinion.
Overlay waiting for evidence: insufficient labeled history (25 obs). We don't publish a forecast until its backtest beats naive.
Overlay waiting for evidence: no target series. We don't publish a forecast until its backtest beats naive.
Overlay waiting for evidence: no target series. We don't publish a forecast until its backtest beats naive.
Supply tension combines the share of laden tonnage, the share NOT idle, the share NOT ballasting home, and the speed index — each weight documented in the method meta on every observation. High supply tension means owners have pricing power.
Demand pull is the rolling z-score of segment ton-miles (transport work performed). Positive z amplifies the composite — a pull that stretches an already-tight supply.
The index is the rolling percentile rank of that composite over the recent history, mapped to 0-100. Regime bands: ≥70 tight, 30-70 balanced, ≤30 slack.
The rate-spike overlay is a logistic model of the probability the TC-daily moves >15% within two weeks, features drawn from tightness and its 7-day delta. It publishes ONLY when a rolling-origin backtest clears AUROC 0.65 with at least five observed spikes. Otherwise the field stays null and the gate reason is shown — we don't publish forecasts that fail their backtest.
Refreshed twice daily. Values, weights and gate outcomes carried verbatim from the intel job; nothing is hand-tuned per segment.
