Results
Country and site recommendations for the submitted query.
Submit a query from the Input tab or select a job from History.
Loading scoring weights...
No country recommendations for this query.
| # |
Country |
Score |
Evidence |
Precedent |
Details |
No site recommendations for this query.
| # |
Site |
Location |
Country |
Score |
Trials |
Completion |
Evidence |
Job History
Recent Compass recommendation queries fetched from the server.
No queries recorded yet.
| Time |
Query |
Phase |
Countries |
Sites |
Top Country |
Status |
|
Control
Live configuration, scoring weights, and runtime stats for the Compass engine.
Loading weight configuration...
Compass Pipeline Description
Scoring methodology, factor weights, and data sources
Country Scoring Factors
Final score = weighted sum of six normalized factors drawn from precedent density, success, recency, feasibility, regulatory speed, and patient-pool signals.
| Factor | Weight | Source | Description |
| density | — | AACT sites | Precedent trial count normalised across candidate countries |
| success | — | AACT overall_status | Wilson lower-bound CI on historical trial success rate |
| recency | — | AACT start_date | Exponential decay of precedent trial start dates (more recent = higher weight) |
| feasibility | — | World Bank / governance composite | 0.5 × governance + 0.5 × GDP per capita (normalised) |
| reg_speed | — | CTIS/EUCTR + AACT | Median regulatory approval days (60% CTIS + 40% AACT blend) |
| patient_pool | — | GBD DALY cache → WHO GHO → median fallback | Estimated disease-burden patient pool per country (source tagged per result) |
Weights shown are loaded dynamically from server config at page load.
Site Scoring
| Component | Description |
| Precedent overlap | Number of precedent NCT IDs the site appears in |
| Completion rate | Wilson lower-bound on historical trial completion rate |
| Recency decay | Exponential decay of most recent trial date at site |
| Stream cap | Firestore chronicle_sites streamed up to 10000 docs per request (CS-7 backlog item) |
Confidence Intervals
Bootstrap CI is opt-in (adds ~2–5 s). ci_reliable = false when n_precedent < 10 for a country.
| Parameter | Value |
| Method | Bootstrap (with replacement) |
| Resamples | 200 |
| Confidence level | 95 % |
| Minimum sample | 10 precedent trials (below this: ci_reliable = false) |
Data Sources
| Source | Content | Refresh |
| AACT (ClinicalTrials.gov) | Trial sites, countries, completion status | Monthly pipeline run |
| GBD cache (IHME) | Disease-burden DALY per country/condition (2 052 entries) | Static (GBD 2019) |
| WHO GHO | Global Health Observatory country-level statistics | API fallback when GBD cache misses |
| CMS cache | Site-level quality scores (74 K NCTs) for site scoring | Static build |
Firestore chronicle_sites | Materialised site stats (n_trials, completion rate) | After pipeline run |
Firestore country_profiles | Pre-computed country factor scores | After pipeline run |
Known backlog: CS-7 — shard chronicle_sites Firestore by country to reduce 10000 stream per request.
LambdaMART country ranker wired but not yet trained (pipeline/transforms/country_ranker.py).