Skip to content

P1.3: surface Healthcare.gov as second case#13

Open
MrBinnacle wants to merge 1 commit into
mainfrom
launch/surface-healthcare-case
Open

P1.3: surface Healthcare.gov as second case#13
MrBinnacle wants to merge 1 commit into
mainfrom
launch/surface-healthcare-case

Conversation

@MrBinnacle

Copy link
Copy Markdown
Owner

What

Lifts the testbed from single-case Boeing to the n=2 framing locked on the launch board.

  • New `Healthcare.gov Pre-launch` prompt variant. Prompt text lifted verbatim from the canonical case study's pre-October-2013 input.
  • New `Case: Healthcare.gov` preamble in the left panel below the existing Boeing preamble. Same Decision / Outcome / What-to-look-for shape. Surfaces calibration: 5/6 recall, 0 false positives, 1 disclosed miss.
  • Orientation banner updated from singular Boeing to plural "two known-outcome failures."
  • README section renamed from "The Boeing methodology runs" → "Methodology calibration — two cases," with a Boeing subsection and a new Healthcare.gov subsection.

Preview

https://deploy-preview-p13--azimuth-testbed.netlify.app

Cold-read verified in deployed bundle:

  • `Case: Healthcare.gov` → 1
  • `Healthcare.gov Pre-launch` (variant in prompt list) → 1
  • `5/6 documented` (calibration score) → 1
  • `two known-outcome failures` (orientation banner) → 1

Claims-ledger compliance

  • "5/6 documented causes surfaced; 1 miss disclosed" matches the ledger's allowed wording for the Healthcare calibration row.
  • No "caught Healthcare.gov" or "predicted Healthcare.gov." The README spells out why over-claiming would be a credibility tax.
  • "Calibration exhibit, not a benchmark or validation" framing preserved on both surfaces.

Verified

  • `cd testbed && npm run build` → green (192.63 KB bundle, 566 ms)
  • Healthcare prompt text is the canonical input from the shipped case study; running on Opus 4.5 should reproduce DELAY PENDING EVIDENCE with the documented Critical Risks.

Do not merge until

  • Cold-read on preview confirms two preambles stack cleanly on both wide and narrow viewports.
  • Matthew approves. Merge to `main` + manual `netlify deploy --prod` ships to production (Git→Netlify still not wired per E2).

🤖 Generated with Claude Code

Lifts the testbed from a single-case (Boeing) demo to the n=2 framing
locked in the launch board. The Healthcare case study already exists
in MrBinnacle/azimuth/examples/case-study-healthcare-gov.md — this PR
makes it runnable from the testbed and references it in the README.

Changes
- testbed/App.jsx PROMPT_VARIANTS: adds "Healthcare.gov Pre-launch"
  variant. Prompt text lifted verbatim from the case study's "Input
  (as presented, circa September 2013)" section so the verdict the
  user gets matches the documented run.
- testbed/App.jsx left-panel preamble: adds a "Case: Healthcare.gov"
  preamble below the existing Boeing preamble. Same Decision /
  Outcome / What-to-look-for shape. Surfaces calibration score
  (5/6 recall, 0 false positives, 1 disclosed miss).
- testbed/App.jsx orientation banner: updates the "what you're
  looking at" card from singular "Boeing's 2011 decision" to plural
  "two known-outcome failures — Boeing's 2011 decision and the
  October 2013 launch of Healthcare.gov". Updates the "this testbed
  is a calibration exhibit on a known-outcome failure" wording to
  "two known-outcome failures".
- README.md: section title renamed from "The Boeing methodology runs"
  to "Methodology calibration — two cases" with a Boeing subsection
  and a new Healthcare.gov subsection. Healthcare subsection states
  the 5/6 recall, 0 false positives, 1 miss explicitly and links to
  the case study file in the skill repo.

Claims-ledger compliance
- "5/6 documented causes surfaced; 1 miss disclosed" wording matches
  the ledger's allowed form for the Healthcare.gov calibration row.
- No "caught Healthcare.gov" or "predicted Healthcare.gov" wording.
  The README spells out explicitly why over-claiming would be a
  credibility tax.
- "calibration exhibit, not a benchmark or validation" framing
  preserved on both surfaces.

Verified
- npm run build -> green (192.63 KB bundle)
- The Healthcare prompt is the canonical input from the shipped case
  study; running it on Opus 4.5 should reproduce DELAY PENDING
  EVIDENCE with the documented Critical Risks.

Do not merge until
- Cold-read of the deploy preview confirms two preambles render
  cleanly in the left panel without overflow on the 1200px+
  layout and the < 640px stacked layout.
- Matthew approves (merge to main + manual netlify deploy --prod
  ships to production).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@MrBinnacle

Copy link
Copy Markdown
Owner Author

Held pending Substack article publish. Disposition will be revisited after the launch sequence completes — Healthcare.gov as second case is not gating the Substack article (which only references Boeing) but is in scope for the post-launch fast-follow window. Not abandoned.

MrBinnacle added a commit that referenced this pull request Jun 5, 2026
- .gitignore: add .day1-state.json, graphify-out/, drafts/ (paste-and-ship
  surfaces, not versioned product)
- ci.yml: add weekly Monday 12:00 UTC cron trigger so npm audit runs against
  the locked deps without waiting for a push. Closes the false-negative trap
  where commit 58332db flipped CI red because a Critical vitest alert was
  published between pushes and the audit step only fires on push by default.

Session-close housekeeping companions (no source changes, not in this commit):
- Filed issue #19 (React 19 + plugin-react 6 + Vite 8 coordinated trio merge)
- Filed issue #20 (CI Node 20 → 22 bump before 2026-06-16)
- Closed PRs #1, #2, #4, #5 with @dependabot ignore-major directives + cross-ref
  to umbrella issue #19
- Posted held-pending-Substack-publish disposition on PRs #13, #14

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant