parse-dec — optional per-deployment seam (NOT built in the base template)

AI dec-page parsing is a per-deployment add-on. The base template deliberately does not ship it, and you should not build it unless explicitly handed the source implementation (upload path, model, exact prompt, and output shape). Do not invent the prompt or output shape.

The slot

When a deployment adds it, create src/pages/api/parse-dec.ts:

export const prerender = false;
// Vercel function config — multi-page PDF + LLM latency needs longer runtime:
export const config = { maxDuration: 60 };

Intended contract (documented for whoever adds it)

multipart PDF in → base64 → Claude document block → structured JSON (named insured, carrier, policy number, coverages/limits, effective dates, premium) → prefill form fields and/or append to the AgencyZoom lead notes.

Serverless caveats to design around

• ~4.5 MB Vercel function payload limit. Large/scanned dec PDFs must not be POSTed directly. Route them through blob storage first (client uploads to Vercel Blob / S3, then hands the function a URL); the function fetches and parses. Base the interface on “accepts either a small inline file or a blob URL.”
• Set maxDuration high enough (start at 60s) for multi-page OCR + model latency.
• Model key from env (ANTHROPIC_API_KEY), never config, never source.

Status

Pending source. The reference McWright Next.js build contains no dec-parse code (confirmed in the Phase-1 audit). When the separate implementation is provided, fill this slot from it.