Help run a synthetic RedirX tutorial using the attached old-urls.csv and new-urls.csv. These are fictional URL inventories, not live sites. Read https://redirx.dev/llms.txt and inspect the connected RedirX tools before acting. If required tools are missing, stop and report the mismatch. Never request raw credentials.

This sample has 8 old URLs and 10 new URLs. Explain that starting it can use the account's one-time free migration; ask whether to use this sample or save that trial for a real project before starting a run. Do not buy anything, use a subscription allowance, approve overages, or create checkout. Stop if the service asks for payment or the account's applicable entitlement is unclear.

Once the user chooses the sample, plan a migration named “Synthetic first-run tutorial” from https://old.example.com to https://new.example.com. Import both attached CSV URL columns as explicit inventories, excluding headers. Do not crawl, discover, or fetch these fictional pages. Confirm both saved inventories are complete with exactly 8 and 10 unique URLs. Do not invent IDs. Save returned IDs and use stable, distinct idempotency keys for plan, each import, run, decisions, and export.

Run the URL-only migration with no confirmed_pairs. Preserve the same key when retrying an identical operation. Poll its returned operation, respecting retry timing. Read every proposed match and show a table containing each old URL, proposed destination, alternatives, confidence estimate, and unresolved state. A successful tool invocation is not proof of a completed run.

Pause for the user's explicit row-level review. Do not read a reference answer key or infer permission to approve from a high score. After review, apply only the chosen destinations with resolve_matches set_target and the current expected_revision. Check every per-row result. Leave anything uncertain unresolved; never automatically select 410, a homepage, or a category as a replacement.

Read current selection state. If any old URLs remain unresolved, report their count and ask whether to produce a partial CSV of approved rows. Only export with allow_partial=true after explicit approval. Read the artifact using its authorized MCP resource and save the actual returned content. Report the artifact ID, approved/exported count, excluded count, and unresolved URLs. Check every exported source and destination is in the supplied inventories, with no duplicate sources or self redirects.

Stop after downloading. Do not install redirects or call verify_redirects for these fictional domains. State that this was a synthetic tutorial, not a production migration, accuracy benchmark, or verified live installation. If an operation fails, preserve its IDs and explain the next recoverable action rather than claiming completion.