AI-readiness check over Connect

AI-readiness check over Connect

Naming note. This workflow is an artificial intelligence (AI)-readiness check, not a certification. The tool gives the document a graded score across four dimensions. The check does not state that the document is “certified” for any purpose. The slug keeps the historical ai-ready-certification segment for continuity, but the workflow remains a readiness assessment that an independent process can confirm or override.

At a glance

This recipe scores a document across four dimensions: integrity (forensic), longevity (long-term-validation health), structure (tagged/accessible content), and machine-readability (text extraction). It runs the AI-readiness tool over the Connect transports. The tool is Enterprise-tier. A class_exists() probe discovers it, and the tool registers only when nextpdf/premium is installed alongside the server. Document creation and rendering still use core tools.

The graded result, a letter grade plus per-dimension scores, is the tool’s own readiness assessment. It is not a certification. A “Grade A” result means the tool’s checks found no problem within their scope. It does not mean that an authority has certified the document.

Install

composer require nextpdf/server

Confirm that the Enterprise AI-readiness tool is present with a tools/list call. See /connect/tool-catalog/.

Conceptual overview

The structure dimension rewards tagged, natural-language structure that makes content machine-extractable in reading order (ISO 32000-2 §14.7). The longevity dimension reflects long-term-validation material. PAdES baseline levels are distinct, and B-LT/B-LTA longevity material is an Enterprise-only capability (ETSI EN 319 142-2 §5.5). A standard states requirements for a checker to evaluate; producing software does not self-certify (PDF/A-4 §6.2.3). The score is therefore a readiness signal, not a conformance or legal attestation.

API surface

Verify tool names against the running registry with tools/list. The catalog of record is /connect/tool-catalog/. This recipe does not restate a tool count.

Code sample — Quick start

Create and render a document with core tools, then run the readiness check:

{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "tools/call",
  "params": {
    "name": "ai_ready_certify",
    "arguments": { "document_id": "<id of a rendered document>" }
  }
}

The tool name ai_ready_certify is the registered protocol name. The workflow it performs is a readiness check, as the naming note above states.

Code sample — Production

curl -sS -X POST https://connect.example.com/v1/tools/ai_ready_certify \
  -H 'Authorization: Bearer '"$NEXTPDF_CONNECT_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"document_id":"<id>"}' \
  -o /tmp/ai-ready.json -w '%{http_code}' > /tmp/ai-ready-status || {
    echo "transport failure invoking ai_ready_certify" >&2; exit 1; }

The response includes a top-level grade and per-dimension scores (integrity, longevity, structure, machine_readability), plus findings and recommendations. Treat the grade as a readiness signal to act on, not as a certificate.

Edge cases & gotchas

• Document not rendered. The check requires a finalized PDF; render the document first.
• Scanned image-only PDF. The machine-readability dimension scores low; run optical character recognition (OCR) before checking.
• No signatures present. The longevity dimension is advisory, not a failure. The tool reports the absence of long-term validation (LTV) material; it does not penalize it as tamper.
• Tool absent. Without nextpdf/premium, the Enterprise AI-readiness tool is not registered; the call fails with an unknown-tool error.

Performance

The front-matter budget is a documentation cap. Large documents may hit the tool’s analysis timeout; split them rather than raising a global timeout.

Security notes

The integrity dimension reuses the forensic analysis, so its “clean” sub-verdict carries the same caveat. It means detected absence of modification within the analysis performed, not a tamper-proof guarantee. Do not log the document path or the full report at an externally shipped log level.

Conformance

Claim	Clause	reference_id
A checker evaluates the document; software does not self-certify	PDF/A-4 §6.2.3
Tagged natural-language structure → machine-extractable in reading order	ISO 32000-2 §14.7
B-LT/B-LTA longevity material is distinct and Enterprise-only	ETSI EN 319 142-2 §5.5

Support for an AI-readiness check is not a certification. The graded result is a readiness assessment. An independent process makes any certification determination.

Commercial context

The AI-readiness tool is Enterprise-tier and registers only when nextpdf/premium is installed alongside the server.

Connect specifics

Transport availability (MCP / REST / gRPC)

Invoke the call the same way over Model Context Protocol (MCP) tools/call, the Representational State Transfer (REST) tool endpoint, and the gRPC service through the shared tool executor.

HITL risk tier

Document creation and rendering are caution-level; the readiness check is read-only. None is approval_required by default. See /connect/hitl-risk-tiers/.

Confirmation gate JSON envelope

These tools do not trigger the gate unless an operator override raises one to approval_required. The envelope and single-use token contract is in /connect/hitl-risk-tiers/.

See also

• /cookbook/connect/compliance-check/ — targeted named-standard check.
• /cookbook/connect/forensic-analysis/ — the integrity dimension in detail.
• /cookbook/connect/ltv-health-check/ — the longevity dimension in detail.
• /connect/tool-catalog/ — per-tier tool set computation.