Tighten chatbot grounding guardrails

[philhetzel]

Summary

• add shared grounding guardrails to the chatbot system instructions in both the API route and the shared chat runtime
• add a focused grounded-claims scorer to evals/sim.eval.ts to measure unsupported timelines and process promises
• validate the change with bounded and full Braintrust sim eval runs against the remote dataset

Validation

• npx tsc --noEmit
• bt eval --env-file .env --first 10 --json evals/sim.eval.ts
• bt eval --env-file .env --json evals/sim.eval.ts

Braintrust experiments

• Smoke run: l8r-sim-eval-2026-06-30-2af0b9c9
• Full run: l8r-sim-eval-2026-06-30-e61530a1
• Additional reruns from a shell-quoting mishap during PR creation: l8r-sim-eval-2026-06-30-1f65a08e and l8r-sim-eval-2026-06-30-37a4d46a

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
l8r	Ready	Preview, Comment	Jun 30, 2026 8:02pm