Distillation

Date: Aug 5, 2026

• 🔴 (13:37) User specified that the document text-trimming result must always be <= maxBytes bytes.
• 🔴 (13:37) User specified that trimmed output must always be valid UTF-8 within the byte budget.
• 🔴 (13:37) User specified that trimming must never split a multi-byte code point.
• 🔴 (13:37) User specified that trimming must prefer whitespace boundaries and indicate that text was trimmed.
• 🔴 (13:37) User specified that trimming must never collapse output to near-empty, including when processing one enormous token.
• 🔴 (13:37) User stated in the blog post that they always have their phone.
• 🟡 (13:37) The src/lib/standard-site.ts implementation defines RECORD_KEY_RE = /^[A-Za-z0-9._~:-]{1,512}$/ and documents site.standard.document/publication fields with maxSize: 1000000.
• 🟡 (13:37) The text conversion path in src/pages/standard-site.json.ts includes an htmlToText(html: string) helper that normalizes whitespace before document generation.
• 🟡 (13:37) scripts/publish-standard-site.mjs reads dist/standard-site.json, uploads blobs through com.atproto.repo.uploadBlob, resolves the DID’s PDS, lists existing records with com.atproto.repo.listRecords, and verifies the .well-known publication URI against manifest.publication.uri.
• 🟡 (13:37) Investigation identified the failure as a record-key strategy problem rather than a post-content problem: the implementation uses the human-readable post slug, while site.standard.document requires a deterministic 13-character TID. The available AT Protocol tooling and lexicon details are being checked before selecting a mapping that preserves stable links and publish idempotency.