license: cc-by-4.0
pretty_name: GreyForge Fintech Tool Fixture Library v1
task_categories:
- other
language:
- en
tags:
- fintech
- tool-use
- synthetic-data
- schema-validation
- harness
size_categories:
- n<1K
configs:
- config_name: default
data_files:
- split: train
path: data/greyforge_fintech_tool_fixture_library_v1.jsonl
GreyForge Fintech Tool Fixture Library v1
Fixture library for harness development; not a reliability benchmark or ChangeGuard diagnostic.
This repository publishes 16 synthetic, versioned tool
input/output mocks with explicit tool_state values. Use them to exercise
agent tool adapters and harness plumbing. They do not score agents, grade
outcomes, or substitute for a GreyForge ChangeGuard diagnostic.
Synthetic data — All request shapes and response bodies are fabricated harness fixtures. No real customer, transaction, or institution data is included.
Provenance
| Field | Value |
|---|---|
| Synthetic | yes |
| Public release | greyforge_tool_fixture_library_v1.0.0 |
| Schema | tool_fixture_v1 |
| Fixture ID prefix | lib_* |
| Manifest hash | sha256:a574107d86147bf82a51d120f171190752e9ce97c7597f2c5dadbe1b9d746047 |
What is included
| Included | Excluded |
|---|---|
| Versioned tool I/O mocks | Severity weights / scoring fields |
| Explicit tool states | Failure-family taxonomy codes |
| Schema-valid interface behavior | Remediation modules |
| Latency + side-effect flags | Locked / calibration / eval case rows |
| Outcome grades / policy decision labels |
Intended use
- Smoke-test tool adapters against timeout, stale, unavailable, and false-success paths
- Validate harness loaders against
tool_fixture_v1 - Pair with the public reliability sampler for schema/tool-state integration work
Related public sampler:
GreyForge/greyforge-fintech-reliability-public-sampler-v1
A separately scoped ChangeGuard diagnostic produces customer-specific reliability evidence; this fixture library does not.
Important limitations and non-endorsement
This library is a developer harness utility. It is not designed or validated as a production benchmark, calibration set, compliance assessment, or certification instrument.
GreyForge does not endorse claims that a model or system is production-safe or diagnostically validated based on fixture playback alone.
CC BY 4.0 governs permitted reuse of the material in this repository.
Tool-state interface behavior
| State | n | Expected interface behavior |
|---|---|---|
success |
3 | Return a schema-valid success payload; do not invent fields beyond the mock. |
timeout |
2 | Surface a timeout/error state; do not treat as completed lookup. |
stale |
2 | Return a stale/outdated signal; do not treat as current confirmation. |
contradictory |
1 | Return conflicting payload fields; do not silently pick a winner. |
invalid_argument |
2 | Reject malformed request shape; do not coerce into success. |
permission_denied |
2 | Deny the action; do not escalate privileges in the mock. |
unavailable |
2 | Report service unavailable; do not fabricate a live status. |
false_success |
2 | Return a success-shaped envelope that must not be trusted as verified. |
Tools covered
| Tool | n |
|---|---|
account_action |
2 |
dispute_status_lookup |
3 |
kyc_status_lookup |
2 |
refund_action |
3 |
transaction_lookup |
6 |
Record fields
Each row includes: fixture_id, fixture_revision, tool_name, tool_state,
request_shape, response_body, latency_ms, side_effects_allowed,
content_hash.
These fields describe mock interface behavior, not agent outcome grades.
Files
| File | Description |
|---|---|
data/greyforge_fintech_tool_fixture_library_v1.jsonl |
16 fixtures |
validation_manifest.json |
Provenance, hashes, boundary attestation |
LICENSE |
CC BY 4.0 (library only) |
Re-mint (maintainers)
python -X utf8 ops/mint_hf_greyforge_tool_fixture_library.py
pytest tests/forge/test_hf_greyforge_tool_fixture_library_release.py -q
Product
GreyForge · Use fixtures to build harnesses; use ChangeGuard for paid reliability diagnostics. This repository is a public technical utility only — free discovery, not a paid dataset listing.
ChangeGuard (commercial)
Need a scoped reliability assessment for a fintech agent workflow?
ChangeGuard Founding Diagnostic — €490 excl. VAT, limited to 3 founding slots.
Start checkout: ChangeGuard Founding Diagnostic
This fixture library is not a substitute for the Diagnostic and is not sold as a paid dataset SKU.