CompanyBench

Dropped incremental-auth phantom-hold reconcileCards & Interchange

17 verifiers · 3 LLM-graded

EnvironmentsJiraSlackZeta3 SQL

TaxonomyCards & InterchangeCards & Interchange

Task prompt

CX reopened ENG-318 — incremental authorization holds were dropped, depressing available balances. Read ENG-318 for the signature: approved incremental-auth-request on a trace that already has a pending Tran, then a hold-release archives the Tran. From rds.card_events (status='approved' only) and rds.trans (join trace_id = external_trace_id): count traces_with_incremental_auth (distinct trace_id with approved incremental-auth-request). For the affected phantom-hold population, flag traces with an approved incremental-auth-request where inc_cents > orig_auth_cents AND (Tran archived_reason is 'hold released'/'hold release' OR Tran amount stuck at orig_auth magnitude). Report affected_trace_count, affected_account_count, and dropped_increment_dollars_cents as SUM(inc−orig) on that set. Confirm seed account 3300 / trace 51757995-407c-4461-89ce-34ff2b87ac4f. For triage prioritization, also flag the single most-affected account — the account carrying the largest TOTAL dropped-increment dollars — and report its account_id, its dropped-increment cents, and how many affected traces it has; and how many of the affected accounts have more than one affected trace (repeat offenders). Note the seed account 3300 is not necessarily the worst one. When done, complete ALL THREE deliverables before finishing: (1) create the Slack channel "phantom-hold-incremental-auth-audit", post the summary there, and pin it; (2) file the Jira Engineering ticket titled "ENG-318 phantom hold incremental-auth audit"; (3) reply to me with the findings. In that reply, include the four headline numbers plus the top affected account (its id, dropped cents, and affected-trace count) and the repeat-offender account count, and END the reply with a single JSON object using exactly these keys: traces_with_incremental_auth, affected_trace_count, affected_account_count, dropped_increment_dollars_cents, top_affected_account_id, top_affected_account_dropped_cents, top_affected_account_trace_count, repeat_offender_accounts. Do not skip the Slack and Jira write-backs.