Use when the user needs failed order records, counts, affected products, or an
explanation of why a group of orders failed. Broader "checkout is down" symptoms
start with b2c-checkout-triage; some checkout failures create no order at all.
Scope and impact
Resolve instance/site with config_inspect (CLI: b2c setup inspect). Fix UTC
from/to boundaries and a comparison period. Ask for missing site/window;
do not use a demo default. Record impact and owner under the team's incident
process. Counts describe indexed orders, not all checkout attempts or conversion.
Access and tools
| Need | Toolkit path | Important limit |
|---|---|---|
| Enumerate FAILED orders | Check Admin Orders query support with scapi_search; external OCAPI Shop POST /order_search is the fallback |
SCAPI status=failed has returned HTTP 400; a response enum does not prove query support. CIP/CCAC is not a substitute for this population. |
| External request authentication | CLI b2c auth token --json or MCP's documented AM token-export helper |
Use only for the external client. SCAPI code mode has no OCAPI request helper, and its safety policy does not cover external HTTP. |
| Failure reason and correlation | logs_list_files -> discovered integration prefix -> bounded logs_get_recent |
Discover custom-* categories; error/customerror defaults can miss the payment integration. |
| Suspect product/custom attribute | Code-mode Admin reads; live schema expansion when the field is unknown | Compare selected fields on known failed/healthy IDs; Shopper execution is unsupported. |
| Suspected data writer | b2c-job-health | Inspect the execution log and actual data effects, not just job status. |
Read order-search recipe before counting: it
defines the site grant/index prerequisites, bounded request bodies, auth handling,
and fault interpretation. For code mode, first read skill://mcp/scapi/SKILL.md.
If external requests are unavailable, report that evidence gap and continue with
logs; do not deploy an OrderMgr endpoint to obtain access.
Checks and decisions
- Query FAILED and all indexed order totals for the same site/window. Validate
every response. Report
failed / all indexed orders; if either query fails, the ratio is unknown. Zero denominator means undefined, not zero failures. - Select at most 10 recent failed orders: order number, creation time, status, and product IDs. Omit customer/payment fields. Label product frequencies as a sample and deduplicate each product per order. The earliest indexed failure in the window is not necessarily incident onset.
- Correlate safe order/request IDs and timestamps with integration logs. Test whether failures cluster by input, provider response, region, or change window. Do not infer a payment decline solely from the order's FAILED status.
- Compare one healthy input before blaming an attribute. If a job writes the suspect field, inspect its log and persisted data. Keep failed stages, sampling/index delays, alternative causes, and missing access explicit.
Mitigation and recovery
Investigation does not authorize changing order status, reprocessing orders, retrying charges, refunding, rerunning integrations, or changing access/indexes. Before authorized intervention, establish duplicate-charge/message risks and partial effects. Verify new outcomes over an equivalent interval; old FAILED records need not disappear for recovery to be real.
Escalation and handoff
Provide site/window, totals/denominator, sample size and safe IDs, recurring error, affected input, suspected writer/component, limitations, and the next owner. Route to the administrator, developer, integration provider, or Salesforce Support using the case template. Missing Shop access is a targeted administrator request, not evidence of no failures. Case submission and other external communication require authorization.