keaxe21/fiscus-financial-workflow-state-fixtures

FISCUS AI Financial Workflow State Fixtures

This small, fully synthetic reference dataset demonstrates how finance-critical workflow events can be classified without confusing an AI suggestion, a user approval, an external-provider response, and a committed accounting effect.

The fixtures are intended for:

• validating deterministic workflow guards;
• testing human-in-the-loop finance assistants;
• reviewing retry and idempotency behavior;
• checking that user-facing status matches the source-of-truth state;
• verifying that audit events remain reconstructible.

It is not a training benchmark, a collection of real transactions, accounting advice, tax advice, or a substitute for provider and regulatory validation.

Data origin

All rows are hand-authored synthetic scenarios. They contain no customer records, bank data, invoice identifiers, taxpayer identifiers, personal data, model outputs from production, or confidential provider responses.

Schema

Field	Meaning
case_id	Stable synthetic fixture identifier
domain	Finance workflow surface
initial_status	Source-of-truth state before the event
event	Observed or requested transition event
external_result	Provider/system outcome when applicable
approval_state	Human or policy approval state
expected_status	Expected state after guard evaluation
financial_effect	Whether a committed financial effect exists
retry_policy	Safe retry behavior
audit_required	Whether the transition must create an audit event
reason_code	Stable machine-readable rationale
rationale_tr	Short Turkish explanation

Core semantics

The fixtures deliberately preserve these distinctions:

1. suggested is decision support, not an accounting fact.
2. approved does not imply that an external operation was committed.
3. unavailable means the result is unknown; it must not be presented as a negative result.
4. A retry must not duplicate an already committed financial effect.
5. Reversals, cancellations, and corrections remain explicit state transitions rather than destructive rewrites.
6. Live bank and e-document actions require the relevant plan, organization authority, business eligibility, provider readiness, and user approval.

Example

{
  "case_id": "bank_reconciliation_001",
  "domain": "bank_reconciliation",
  "initial_status": "suggested",
  "event": "user_approves_match",
  "external_result": "not_required",
  "approval_state": "approved",
  "expected_status": "approved",
  "financial_effect": "none_until_commit",
  "retry_policy": "not_applicable",
  "audit_required": true,
  "reason_code": "APPROVAL_IS_NOT_COMMIT",
  "rationale_tr": "Kullanıcı onayı eşleştirmeyi yetkilendirir; kalıcı muhasebe etkisi ayrı commit adımından sonra oluşur."
}

Responsible use and limitations

• Treat the rows as illustrative test fixtures, not universal accounting rules.
• Add jurisdiction, organization, role, amount, tax, document-type, and provider-specific guards before production use.
• Do not infer regulatory eligibility from a missing or unavailable provider response.
• Do not use these fixtures to auto-post ledger entries without explicit domain validation and authorization.
• A passing fixture suite does not prove legal, tax, accounting, security, or production readiness.

Project context

The state model reflects the controlled-automation principles used by FISCUS AI, an AI-supported financial operating system for Turkish SMEs. Product availability and live integrations remain conditional on plan, organization authority, business eligibility, provider readiness, and user approval.

License

The dataset card and synthetic fixtures are released under CC BY 4.0.