# Rough Year - compare health plans side by side by what a quiet, a usual and a rough year would cost

> Open enrollment helper. Put two to four health plans on cards (premium per month, deductible, out-of-pocket maximum, coinsurance, copays, employer HSA/HRA money) and enter the household's usual year of care (visits, prescription fills, labs, planned care). The browser prices every plan in three years - a quiet year (one check-up and the regular prescriptions), the usual year, and a rough year (the usual year plus one ER visit and a $45,000 hospital stay) - draws each plan's cost curve as bills pile up, marks the break-even points where another plan overtakes the usual-year winner, flags blank fields and must-keep doctors or drugs not yet confirmed for a plan, and suggests a verdict from the numbers alone - free. Two metered tasks: `extract` reads a pasted benefits summary into a plan card, and `verdict` reads the cards and the three years and returns the pick, a read of each plan, a note per year and a before-you-enroll checklist. Every reply is checked against the page.

URL: https://rough-year.skillsafe.ai/
API: https://rough-year.skillsafe.ai/api.html
Built on the SkillSafe skill @jmsktm/insurance-analyst (insurance policy analysis, coverage assessment and risk management for individuals).

## What runs free in the browser (planmath.js, planview.js)

- Cost model (an estimate, not a quote): a copay plan charges its copay for primary-care, specialist, urgent-care and ER visits and for generic and brand prescriptions (a blank copay sends that service to the deductible); labs, imaging, planned care and hospital stays go to the deductible, then coinsurance. An HSA-type plan sends everything to the deductible first. The person's share stops at the out-of-pocket maximum; premiums never count toward it; the employer's HSA/HRA money comes off the year's net.
- Typical prices the page assumes, all estimates: primary-care visit $170, specialist visit $260, urgent care $210, ER visit $2,200, generic fill $20, brand-name fill $480, lab or imaging test $350; a rough year adds a $45,000 hospital stay and one ER visit.
- Suggested verdict from the numbers: `verify-first` when the usual-year winner has a blank premium, deductible, out-of-pocket maximum or coinsurance, or a must-keep item not confirmed for it; else `close-call` when the rough-year winner differs or the usual-year margin is under $400; else `clear-pick`.
- Must-keep lines: "Dr. Okafor (cardiologist): in P1, P2" - the plan ids after the colon are the plans already confirmed.
- Exports: Markdown, a years CSV (plan x year with premiums, copays, deductible, coinsurance, out-of-pocket, employer money, net), plans JSON.

## The metered tasks (model gpt-terra)

- `extract` - input `sbc_text` (clipped on whole lines past 9,000 characters, top and end kept) and optional `plan_label`. Returns `plan` (the card's fields as numbers or null; individual amounts when individual and family are both given), `found` [{field, quote}], `unclear` [fields], `gotchas` [{text, quote}], `summary`. The page checks that every number appears in the pasted text.
- `verdict` - inputs `plans`, `usage`, `years`, `checks`, `must_keep`, optional `note` (JSON-encoded strings). Returns `verdict` (`clear-pick`, `close-call`, `verify-first` - never less cautious than the page's suggestion), `pick`, `title`, `summary`, `plans` [{id, fit, read, watch_outs, cite}], `years` [{year, text, cite}], `before_enroll` [{item, why, cite}], `questions`, `caveats`. The page checks the pick against the cheapest plan, every cited id, and every dollar figure against the numbers it computed.
- House rules: only the input; no real insurer, employer, pharmacy or drug brands; no medical or tax advice; numbers only from the page.

## Notes

- Estimates from a simple model with typical prices - not a quote, and not insurance, tax or medical advice. Sample plans and people are made up.
- Saved comparisons go to the signed-in user's own collection (`comparisons`), only when they press save.