How accurate is Covera, really?
Two honest scorecards. First, how the Monte-Carlo simulation holds up against the published MEPS aggregates it claims to reproduce and the ACA subsidy formula. Second, how candidate models compare when driving the real agent: on whether they cite real numbers, call the right tools, and what they cost.
Simulation accuracy
16/16 checks within tolerance · 80,000 simulated peopleMean spend by age band (vs MEPS)
Spend concentration (vs MEPS)
ACA subsidy formula
Read honestly: the engine reproduces adult age-band means, the ACA subsidy math, and (after adding a person-year frailty termthat correlates a year's care across service lines) the real spend concentration the top few percent of people who drive most healthcare cost. That heavy tail is the hard part to get right, and it is what the bad-year risk you rank on depends on. Source: Calibrated to the AHRQ Medical Expenditure Panel Survey (MEPS), Household Component.
Explore the distribution
5,000 simulated years, run live in your browser.
The long right tail is the whole point: a few people drive most spending, which is why the cheapest plan often is not the safest. Add a condition to watch the mean and tail move.
What real care costs, plan by plan
Each row is a fixed episode of care run through the real adjudication engine on the cheapest plan in each metal tier. Click any number to see the exact math.
Episodes are fixed reference bundles (round allowed amounts), so the math is checkable; the plans and cost-sharing are real CMS PY2026 data. Deductible + coinsurance + copays, capped at the out-of-pocket max, equals the number shown.
LLM model benchmark
The model benchmark (needs ANTHROPIC_API_KEY) hasn't been generated yet. Run npm run benchmark to populate this section.
Faithfulness = share of dollar figures in the reply that match a real simulated number (not hallucinated). Tool accuracy = share of questions where the model called the expected tool. Quality = LLM-as-judge rubric. Cost uses real token usage × published per-model pricing. The harness lives in scripts/benchmark/.