LAI Tracker
A living tracker of Long-Acting Injectable (LAI) drug development — semaglutide depot/LAI technologies first, then broader peptide/protein LAI delivery tech. Tracks both Korean domestic and global players.
Status
This repository contains the tracked data and a Vercel-native dashboard. Data is written and committed here automatically by Claude (via the lai-tracker-scan skill) as part of the LAI Tracker project's research pipeline. Every Vercel deployment rebuilds data/umbrellas/*.json, data/candidates.json, and data/meta.json into a browser-ready snapshot; routine data updates do not require dashboard-code changes.
The interface keeps three conditions distinct:
- Tracked intelligence accepted into the umbrella record
- Candidate entities pending analyst review, ordered by escalation level
- Daily/weekly monitoring and quality-control status from
meta.json
Candidate escalation
A discovered entity is not a one-time report. Every scan re-checks the unresolved candidates in candidates.json, appends new evidence to them, and scores them against a fixed four-condition promotion bar (named entity, technical claim, confirmed source, and a second dated event). A candidate that clears all four moves to ready_for_promotion: it sorts to the top of the review queue, is counted separately on the dashboard, and leads the digest email on every run until an analyst resolves it. A candidate that goes 30+ days without new evidence moves to stalled and is flagged for reject-or-snooze instead of aging silently.
Promotion itself stays manual. The scan proposes and escalates; creating, merging, rejecting and snoozing are admin actions taken by hand. scripts/build.mjs fails the build on a candidate status outside the controlled list, or on an escalated candidate with no written rationale.
Structure
data/
umbrellas/ one JSON file per tracked entity (company/platform/asset)
candidates.json discovered entities with their evidence, follow-up record and escalation status
meta.json last-run timestamps and source health
src/ browser application and data-normalization logic
scripts/ dependency-free static build + email digest renderer
tests/ Node tests for normalization, repository counts, and the email renderer
email/templates/ tokenized HTML template for the daily/weekly digest email
index.html Vercel dashboard entry point
vercel.json explicit static build and output configuration
Email digest
scripts/render-email.mjs fills email/templates/daily-digest.html from a per-run JSON input (see the lai-tracker-scan skill's "Email digest" section for the exact contract) and produces { subject, preheader, html, delivery }.
delivery is the send decision, made here rather than left to the run's judgment: "send" when the run logged at least one new finding or a new candidate, "draft" for a digest carrying only repeated escalations or late items. A "send" digest goes out unattended to a standing recipient list held in the daily routine's prompt — deliberately not in this repository, which is public. A "draft" digest waits in Gmail for a person. email/draft-input.json / email/draft-output.json are gitignored scratch files, regenerated every run.
Local build
Requires Node.js 20 or newer. No third-party packages are required.
npm test
npm run build
npm run serve
npm run serve runs a dependency-free static server over the generated dist/ directory at http://localhost:5173. Vercel runs the same build command and publishes dist/ automatically.
