Skip to content
pulsify
Pulsify
06 / When something goes wrong Menu +

06 / When something goes wrong

When something goes wrong

Start with your agent. It can read everything below and explain what it finds. Each section opens with a request you can paste, then lists what the agent checks, in order.

Data is missing

"I can't see stock for SKU ABC-1 in the UK. Find out why."

  1. connection_status: is the connection connected and active? Active means authorized with billing in good standing.
  2. list_streams: is the event type active? If not, the reason says why.
  3. list_events: did events of that type arrive for that ASIN, SKU and marketplace?
  4. list_listings: which sources does the listing have under data? A missing source is not zero stock. It may not have been reported yet.
  5. Prerequisites: an advertising profile for that country, Brand Analytics access for hourly sales and traffic, and a selling account rather than a vendor account for listings.

In the console: the account's connections in Settings, the event type's page under Streams, and Activity. Refreshes lists how often each source updates.

An automation didn't run

"My stock automation didn't pause anything yesterday. Work out why."

  1. list_automations: is it active, and on the account that owns the listing?
  2. list_streams: is its event type subscribed?
  3. list_events: did the event arrive, and which listings did it reach?
  4. Is the listing active? Offer events skip listings Amazon doesn't report as buyable or discoverable, and listings the crawl hasn't described yet.
  5. Was the event stale? An older event than one already processed for that listing runs nothing.
  6. For advertising events: is the campaign's Automation switch off?
  7. Is the event type REPORT_PROCESSING_FINISHED or DATA_KIOSK_QUERY_PROCESSING_FINISHED? Pulsify handles those internally, and automations on them never run.
  8. list_automation_actions: a warning row means the code read keys the payload doesn't have. missing_fields lists them, with the keys that were there.

An automation on a blocked listing does run: its requests are recorded as blocked and never sent. So are changes to a campaign whose Automation switch is off, whatever event asked for them. See emergency controls.

Amazon rejected a change

"Amazon rejected the price change on ABC-1. What did it say?"

  • Listing requests: list_listings shows pending requests and the latest terminal receipt under mutations, with accepted and Amazon's raw response. The listing's page in the console shows the same ledger, blocked requests included.
  • get_product_type_schema shows the shape Amazon expects for an attribute, which is the usual fix for a rejected patch.
  • Advertising changes: list_ads_entities shows mutations with the provider outcome, HTTP status, raw response and error_message. get_mutation_schema describes supported native fields.
  • Invalid targets or payloads produce failed automation decisions and validation errors in MCP or dry runs. Check each target's explicit type and local id. A decision marked applied means validation and admission; its receipt reports Amazon's result.

The outcome looks wrong

"The automation paused the wrong campaign last night. Show me why."

  1. Replay the event: run_automation with automation_id, event_id and, if the event reached several listings, listing_id. Compare its mutations and logs with list_automation_actions. The replay reads the listing as it is now, not as the original run saw it.
  2. Compare accepted with observed. Amazon can keep showing an old price for a while after accepting a new one. Ads data updates only from a matching, current entity Amazon returned or a later event or refresh. Inspect pending receipts before repeating a request. A partial or ambiguous result is uncertain and stops later requests for that target; do not assume a failed delivery means Amazon did nothing.
  3. A creation that stays uncertain is waiting on reconciliation, not stuck. Pulsify never sends it again. Within about 10 minutes the outcome becomes accepted with the new entity under created, absent when Amazon shows nothing from it, or unresolved when several entities could be it. After absent, ask again. After unresolved, check the candidates listed on the receipt; they appear as ordinary entities after the next refresh. "An identical creation is already pending" means the first request hasn't settled yet.
  4. Allow for restatements. Amazon revises hourly advertising metrics after the fact.

The code is yours, so the fix goes in the automation. Ask your agent to change it, dry-run it against the same event, and activate the new version.

What the console is for

  • Connections, accounts and billing.
  • Read-only inspection: Activity for events, tool calls and webhook deliveries; each listing's request ledger; campaigns; each event type's automations.
  • Activating, archiving and restoring automations.
  • Emergency controls: the listing block and the campaign switch.

Writing code, changing ads and asking questions happen through your agent.

Report a problem

If Pulsify itself is wrong, tell your agent. It files the report with report_bug, including what happened and what you expected. Ideas go through suggest_idea.