chore: upgrade afdocs to ^0.17.1

[SahilAujla]

Summary

This PR upgrades the afdocs dependency from 0.14.0 → 0.17.1, incorporating three minor releases of fixes and adapting the codebase to the breaking API changes introduced in v0.17.0.

It includes:

• Minimal refactors in scoring logic to support the new nullable CategoryScore.score
• Updates to renamed and split check IDs across both scoring code and marketing pages

Key user-facing improvement:
Sites that have both an apex llms.txt and a docs-specific llms.txt (e.g., Cloudflare, Supabase, Vercel, Resend, Pinecone, Clerk, Sentry, Datadog, MongoDB, Auth0, LaunchDarkly, etc.) will now correctly use the docs version as canonical. This comes from afdocs#54 and prevents scores from being skewed by marketing-site indexes.

---

What Changed in afdocs Since 0.14.0

Release	Key Changes
0.15.x	Canonical llms.txt selection (prefers most specific file). Adds --llms-txt-url CLI flag and llmsTxtUrl config.
0.16.x	CategoryScore becomes an object instead of a number. Programmatic API improvements.
0.17.0	Spec v0.5.0. Renames llms-txt-freshness → llms-txt-coverage. Splits llms-txt-directive into HTML + Markdown variants. Adds nested index handling, strips <script>/<style> before sizing, introduces nullable CategoryScore.score, and validates options in createContext() / runChecks(). See the migration guide.
0.17.1	Patch fixes on top of 0.17.0.

---

Code Changes in This PR

Scoring Core

• lib/scoring.ts

  • Introduces normalizeCategoryScores() helper
  • Handles new type:
    Record<string, number | { score: number | null }>
  • Drops null scores (new in v0.17 for low sample sizes), allowing fallback to existing local calculations
  • Result: preserves previous behavior with no UI changes

• app/api/score/route.ts and scripts/score-single.ts

  • Replace inline conversions with the new helper
  • Add required specUrl: '' field to reports (now enforced by computeScore())

---

Renamed / Split Check IDs

• scripts/debug-score.ts

  • Updated WEIGHTS map:

    • llms-txt-directive →

      • llms-txt-directive-html (High / 7)
      • llms-txt-directive-md (Medium / 4)

    • llms-txt-freshness → llms-txt-coverage

• app/master-plan/page.tsx

  • Updated methodology table
  • Rebalanced Discoverability weights (2 + 1) to maintain total = 100

• app/page.tsx

  • Updated Discoverability check count: 1 → 2

---

Dependency Updates

• package.json

  • afdocs: ^0.14.0 → ^0.17.1

• package-lock.json

  • Regenerated to resolve afdocs@0.17.1

---

Stored Data: Rescore Required

Existing Supabase rows still reference old check IDs:

• llms-txt-directive
• llms-txt-freshness

After this upgrade:

• computeScore() will ignore these outdated checks
• Result: slightly inaccurate scores for any non-rescored entries (effectively treated as 0-weight)

Recommended Post-Merge Steps

# 1. Preview score changes (no writes)
export $(grep -v '^#' .env.local | xargs)
npx tsx scripts/compare-scores.ts --changed

# 2. Rescore all sites
npx tsx scripts/rerun-all.ts

From the v0.17 migration guide:

• Typical score movement: ±5 points

• Directive split may reduce discoverability by 3–14 points

• Gains often offset by:

  • HTML size improvements (script/style stripping)
  • Better coverage scoring (especially for nested llms.txt setups)

Net effect: generally small changes, with some significant upward movers.

---

Verification

Check	Result
npm install	Resolves afdocs@0.17.1
npm run build	Succeeds; all 15 routes generated
tsc --noEmit	Passes
Touched files (lib/scoring.ts, updated scripts)	Clean
npx tsx scripts/debug-score.ts https://www.alchemy.com/docs	Runs end-to-end; new check IDs appear; score = 97 (A)

Note:
Pre-existing type errors in the following files are unrelated and out of scope:

• scripts/dedup-supabase.ts
• scripts/reclassify-categories.ts
• scripts/migrate-to-supabase.ts

[vercel]

@SahilAujla is attempting to deploy a commit to the fern Team on Vercel.

A member of the Team first needs to authorize it.

[SahilAujla]

@kgowru could I get your eyes on this pls 🙏🏻