Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

The laziest way to pick a 2027 health plan beat my model (82% vs 79%). So I built Second Look to tell you when lazy is safe

Description and use

Second Look lets employees assess whether choosing the cheapest 2027 health plan is truly cost‑effective by comparing their profile to real 2024 spending data.

Problem

People often choose the cheapest health plan without knowing if it is truly cost‑effective for their personal health profile.

Outcome

Second Look shows that the "just take the lowest premium" rule is truly cheapest for 82% of 758 surveyed people, but overpays on average $106; the "same as last year" rule is cheapest for 78% but overpays $149 on average.

Workflow

  • 1. User selects employer’s 2027 plan menu or uploads a Summary of Benefits and Coverage PDF.
  • 2. User answers five health‑related questions or selects a pre‑defined anonymous survey profile.
  • 3. Second Look parses the PDF with Gemma 4, extracts plan fields, and validates them against the document’s own examples.
  • 4. TabPFN predicts a range of next‑year costs for the user based on 3,033 real people from 2023→2024.
  • 5. Code applies plan rules (deductible, copays, coinsurance, out‑of‑pocket cap, HSA) to compute typical and bad‑year costs for each plan.
  • 6. Second Look outputs a verdict: "You can be lazy" with cost bars, or "Look closer" if predictions are uncertain.
  • 7. Users can tap any plan number to see the exact PDF line and page that provided the value.

Evidence

I priced the real 2024 medical care of 758 people (from a US government survey) under three employers' real 2027 plan menus.

Source

Second Look tells you whether the cheapest 2027 health plan on your employer's menu is safe for you.

Source

Built with PriorLabs-TabPFN. Plan documents are read by Gemma 4 running on your own computer.

Source

  • Record date: 2026-10-26T07:11:45.952Z