Predictions: ptrs & timeline

The CLI exposes Gosset's ML models for a single trial by NCT id.

gosset ptrs: probability of technical & regulatory success

gosset ptrs <NCT_ID>

Returns the model's success probability plus the per-phase transition chain and the SHAP feature drivers.

gosset ptrs NCT05147220

{
  "probability": 0.2153,
  "target_phase": "REGISTRATION",
  "transitions": [ ... ],
  "feature_influence": [ ... ]
}

Pull just the headline number:

gosset ptrs NCT05147220 | jq '.probability'
# 0.2153

Interpreting PTRS

The model estimates a trial's own-phase progression. For a pre-Phase-3 trial, treat the single number as the transition probability, not an end-to-end approval odds; compose it with an external late-phase LOA when you need a full approval estimate.

gosset timeline: predicted primary-completion date

gosset timeline <NCT_ID> [--as-of YYYY-MM-DD]

Predicts when an ongoing trial will actually reach primary completion, with an 80% interval, versus the sponsor's posted date.

gosset timeline NCT06651281        # a running TL1A Phase 3

{
  "trial_status": "running",
  "expected_completion_date": "2030-08-09",
  "sponsor_estimated_completion": "2037-12-17",
  "lower_bound_days": 541,
  "upper_bound_days": 3464
}

Here the model predicts primary completion ~2030, well ahead of the sponsor's posted 2037, the kind of gap the timeline model is built to surface.

--as-of anchors the prediction to a past date (useful for backtests):

gosset timeline NCT05599191 --as-of 2024-06-01

Chaining from a search

Find the trials, then predict each: a two-line landscape of readout odds:

for nct in $(gosset trials --target TL1A --phase 3 | jq -r '.[].nct_id'); do
  echo "$nct  $(gosset ptrs "$nct" | jq -r '.probability')"
done