Inside the AI parser & text generator

SuspendWizard is a fully automated Micro-SaaS. This page describes how the software pipeline turns a pasted suspension notice into a structured Plan of Action text block — no human reviewer is involved at any stage.

0. The pipeline, end to end

Everything below runs inside the web application as a single unattended request chain. There is no queue a person picks up, no draft that waits on an editor, and no delivery step performed by staff:

1. Input — you paste the raw suspension notice and select the marketplace (seconds).
2. Policy matching — the parser classifies the notice against the policy index and returns a diagnosis preview before any payment (typically 10–30 seconds).
3. Generation — the language model drafts the three-part document under a fixed schema (typically 20–60 seconds).
4. Sanitisation — automated post-processing strips filler connectives and any guarantee-style wording, then runs the checks in section 3.
5. Export — the PDF is rendered in your own browser and downloaded directly by you.

Typical end-to-end time from paste to downloadable PDF is under two minutes. Support staff can see billing and access records only; they do not write, review, edit or approve the generated text.

1. Policy source index

The Policy Analysis API runs against an internal index of published marketplace policy pages. The index is refreshed automatically on a schedule:

• Amazon — Seller Code of Conduct, Section 3 (ABSA), Anti-Counterfeiting Policy, Restricted Products, POA guidance.
• Etsy — Seller Policy, Intellectual Property Policy, Handmade Policy, Trust & Safety appeals.
• eBay — User Agreement, Selling Practices, MC011 / Below Standard messaging.
• Shopify — Acceptable Use Policy, Terms of Service, Payments risk-hold rules.

2. Three-part output structure

The document structuring engine enforces the three-part layout marketplace integrity systems expect. Every generation returns:

1. Root cause block.
2. Corrective action block, indexed against uploaded evidence.
3. Preventive measures block.

3. AI drafting engine

A large language model produces the text against the parsed intake. The prompt is constrained by the indexed policy sources and the three-part schema — the model cannot return free-form output. Automated policy checks run on every draft before download:

• Grounding check — every factual claim is linked to an uploaded evidence file.
• Tone check — structural analysis for the concise, accountable register.
• Contradiction check — output is diffed against the current policy index.

4. Self-serve dashboard access

Starter AI Credits include 30 days of self-serve dashboard access after purchase. Inside the dashboard, the same automated endpoints remain available:

• Automated POA regeneration endpoint.
• Escalation-letter template generator.
• PDF and plain-text export tools.

All actions are software calls. Nothing is queued for a human.

5. What the software will not do

• Guarantee reinstatement — only the marketplace decides.
• Publish invented success rates.
• Encourage misleading a marketplace or fabricating evidence.

6. Refund policy

Because the parser shows you its classification before any credits are consumed, you always know what the tool detected before paying. Orders are fully refundable at any time before the first download; once the document is downloaded, digital delivery is complete. Failure to deliver, unclassifiable input and duplicate charges are refunded in full — see the Refund Policy.

7. Launch cohort

The first 50 activations get the launch price of the Starter AI Credit pack locked in. After that, standard pricing applies. This is a real infrastructure cap while the automated processing queue scales.