skills/go-clearing-settlement-reconciliation/evals.json

{"schema_version":2,"skill":"go-clearing-settlement-reconciliation","cases":[{"id":"route-reconciliation","kind":"routing","split":"development","prompt":"Build restartable Go reconciliation from processor reports, payouts, fees, and bank statement items.","should_activate":true,"reason":"External financial evidence and matching dominate."},{"id":"avoid-initial-auth","kind":"routing","split":"development","prompt":"Handle customer authentication and initial card authorization.","should_activate":false,"reason":"Payment initiation belongs to go-payment-lifecycles.","confuses_with":["go-payment-lifecycles"]},{"id":"quality-item-match","kind":"quality","split":"development","prompt":"Fix the fixture so a net bank credit is reconciled to one currency-balanced many-to-one group without reusing components.","fixture":"evaluations/fixtures/reconciliation-netting","expected_invariants":["Supports many-to-one match groups","Uses each normalized component exactly once"],"forbidden_outcomes":["Reuses a component or ignores the group equation"],"graders":[{"id":"item-level","kind":"go-test","target":"./...","weight":1}]},{"id":"quality-rerun","kind":"quality","split":"development","prompt":"A reconciliation batch crashes after posting some adjustments, then reruns.","expected_invariants":["Checkpoints and adjustment identities are replay-safe","Preserves source and prior decision evidence"],"forbidden_outcomes":["Posts duplicate adjustments"],"graders":[{"id":"replay-safe-adjustment","kind":"contains","required":["idempot","checkpoint"],"weight":1}]},{"id":"quality-fedwire-contingency-review","kind":"quality","split":"development","prompt":"Review a Go Fedwire gateway under the January 5, 2026 Operating Circular 6. It sends one payment order over its normal electronic connection, then the socket fails before an acknowledgment arrives. The gateway marks the transfer failed, releases reserved liquidity, creates a new local operation ID, and sends the same customer instruction through the Import Contingency Feature without reconciling earlier submissions. Normal processing later produces evidence that both orders were accepted. An operator says a cancellation request will roll back the first order. State the evidence, resend, finality, and repair invariants for this Fedwire path only.","expected_invariants":["Treats the missing acknowledgment as unknown receipt or processing rather than a conclusive rejection","Keeps the stable customer instruction separate from each transmission attempt and rail message identity","Reconciles payment orders submitted through the normal connection before using the Import Contingency Feature","Accounts for queued pre-outage messages that can process when normal operation resumes","Does not release reserved liquidity or report definite failure until authoritative evidence resolves the instruction","Persists acknowledgment, rejection, Advice of Credit, activity or statement evidence, business day, and operating-contract version","Uses Reserve Bank acceptance and payment records as the Fedwire evidence boundary rather than one mutable local status","Treats a cancellation or amendment request as a separate request rather than guaranteed rollback of an accepted order","Preserves both accepted movements and repairs duplicate economics through authorized linked return or reconciliation work","Applies the circular's protracted-outage finality rule only to that specified mechanism and does not generalize it to other rails"],"forbidden_outcomes":["Converts a network timeout into a definite failed Fedwire payment order","Resends the same economic instruction under a fresh logical identity without reconciling the first submission","Claims a cancellation request automatically reverses an accepted Fedwire order"],"graders":[{"id":"fedwire-contingency-contract","kind":"contains","required":["unknown","reconcil"],"weight":1}]},{"id":"quality-ach-return-reversal-review","kind":"quality","split":"development","prompt":"Review a Go FedACH service under the January 5, 2026 Operating Circular 4 and current applicable ACH rules. It marks a credit final when forward settlement appears, overwrites the original row when a return arrives, and deduplicates only by a fresh ingestion UUID. It applies every NOC directly to that historical entry and posts a ledger movement for it. After a submission timeout it creates a generic REVERSAL as a guaranteed cancellation, including for suspected fraud and an Originator funding failure. One hard-coded two-banking-day deadline is used for every return, NOC, SEC code, account type, and role. Dishonored and contested returns mutate the first return status. State the evidence, identity, role, timing, and ledger invariants without projecting Fedwire finality onto ACH.","expected_invariants":["Keeps the original forward entry immutable with trace, batch, SEC, participant, amount, dates, file, acknowledgment, settlement, and rule-version evidence","Treats the FedACH file acknowledgment as receipt and limited processing rather than guaranteed item acceptance","Models each return, NOC, reversal, dishonored return, and contested return as a separately identified item linked to the original","Deduplicates on stable rail identity and fingerprint rather than a new local ingestion UUID","Treats a return as a separate value-bearing settlement event and does not rewrite the original as if it never occurred","Treats an NOC as a nonvalue correction message that can have multiple change codes and does not post money or mutate historical evidence","Applies NOC data to future origination only according to current SEC, recurrence, role, and rule semantics","Uses reversals only for current rule-permitted erroneous-entry conditions and never as a generic cancellation, fraud recall, or funding repair","Preserves and reconciles both original and reversing entries and accounts for a reversal being returned","Computes deadlines from rule version, reason or change code, SEC and account class, role, settlement date, banking calendar, and processing schedule","Represents late or unsupported cases as explicit exceptions instead of forcing one universal two-day rule","Models dishonored and contested returns with their own identities and provisional or reversing settlement evidence","Posts every value-bearing event through a replay-safe ledger identity while an NOC creates no value posting"],"forbidden_outcomes":["Treats ACH forward settlement as immune from later authorized returns","Uses a reversal as a guaranteed rollback after a timeout or suspected fraud","Applies one return deadline across every ACH reason, SEC code, account type, and participant role"],"graders":[{"id":"ach-correction-contract","kind":"contains","required":["return","NOC"],"weight":1}]},{"id":"quality-fedwire-business-time-cutoff-review","kind":"quality","split":"development","prompt":"Review a Go Fedwire scheduler using the current 2026 schedule. It computes the funds-transfer business day as time.Now().UTC().Truncate(24*time.Hour), stores a fixed -05:00 cutoff, and opens one batch whenever the host calendar says Monday through Friday. A deployment in UTC replaced one in New York. Operators overwrite the cutoff for an approved extension without source or effective interval. The code has already enabled an announced future operating-days expansion, treats a local pre-cutoff send as accepted, and after a crash can seal a second batch for the same day. State the instant, timezone, business-day, schedule-version, extension, evidence, restart, and migration invariants without projecting Fedwire rules onto another rail.","expected_invariants":["Keeps observed instant source event time local civil time rail business-day label processing window and cutoff as distinct values","Uses the governing named Eastern Time location rather than UTC date or one fixed offset across daylight and standard time","Uses the current verified Fedwire holiday and operating schedule whose business day can begin on the preceding calendar day","Rejects Add 24 hours or weekday-only arithmetic as a business-calendar implementation","Versions the schedule calendar Operating Circular participant and message scope with effective intervals so the decision is reproducible","Does not activate an announced future operating-days expansion before its effective version","Represents an approved extension or early close as authenticated scoped evidence with source effective interval and audit instead of mutating a global cutoff","Persists absolute instants named timezone schedule version and derived business-day identity because Go monotonic readings do not survive persistence","Budgets queue connection and protocol time near cutoff and routes unknown clock schedule or extension authority to an explicit exception","Separates local admission and send time from rail receipt acceptance rejection and settlement evidence","Uses a stable conditional batch identity derived from rail participant business day message class and schedule generation so restart cannot double-seal or skip work","Rolls schedule versions through mixed binaries and rollback before activation and retains historical readers for reconciliation","Tests host-zone changes DST gaps and repeats holidays preceding-day openings cutoff boundaries extensions future schedules and crashes around seal"],"forbidden_outcomes":["Uses a UTC calendar date or fixed offset as the Fedwire business-day identity","Applies a future announced schedule before its effective date","Claims a local send before cutoff proves Reserve Bank acceptance","Creates another batch after restart because the process lost its monotonic clock or local timer state"],"graders":[{"id":"fedwire-business-time","kind":"contains","required":["business day","schedule"],"weight":1}]}]}