# Credit Without Debit
Formal verification models and reproducible checks for ACH payment flows.
The artifact connects the paper’s payment-system obligations to executable formal models.
## What is this?
Artifact for the ACM CCS 2026 paper **Credit Without Debit: Formal Verification of Banking Payment Systems**, by Ali Hamza Malik and Muhammad Taqi Raza, University of Massachusetts Amherst. NuSMV/nuXmv models describe Automated Clearing House (ACH) payment flow; scripts merge, verify and summarize results in Markdown. Credit Without Debit (CWD) encompasses A1 creditBLEED (Φ1), A2 creditMINT (Φ2) and A3 creditRESET (Φ3).
- **Maintained version:** this repository,
- **Archived version:** Zenodo record [22701199](https://zenodo.org/records/22701199) (DOI [10.5281/zenodo.22701199](https://doi.org/10.5281/zenodo.22701199))
## At a glance
| Item | Value |
| --- | --- |
| Model fragments | 8 |
| Participants | 2 Originators, 2 Receivers, ODFI, ACH Operator, RDFI |
| Checks | 40: 3 INVARSPEC + 37 EF |
| Reference safety results | Φ1–Φ3 all violated |
| Reference EF results | 29 true, 4 false, 4 bounded-inconclusive |
The composition joins participant state machines through control and monetary channels.
```mermaid
flowchart TB
subgraph ADV_O["Adversarial Originators"]
direction LR
O0["orig_0
originator(0, …)"]
O1["orig_1
originator(1, …)"]
end
subgraph BANKS["Trusted, Nacha-compliant banks"]
direction LR
ODFI["odfi_i
ODFI"]
ACH["ach_i
ACH Operator"]
RDFI["rdfi_i
RDFI"]
end
subgraph ADV_R["Adversarial Receivers"]
direction LR
R0["recv_0
receiver(0, …)"]
R1["recv_1
receiver(1, …)"]
end
ADV_O -- "LINK_REQ · VERIFY_REQ
PAYMENT_INIT · REVERSAL_NOTICE" --> ODFI
ODFI -- "LINK_ACK · VERIFY_ACK
BATCH_ACK · RETURN_NOTICE" --> ADV_O
ODFI -- "BATCH_FORWARD" --> ACH
ACH -- "BATCH_ACK · RETURN_NOTICE" --> ODFI
ACH -- "SETTLEMENT_REQ" --> RDFI
RDFI -- "SETTLEMENT_ACK · RETURN_NOTICE
REVERSAL_NOTICE" --> ACH
RDFI == "c_dollar: FUND_CREDIT · FUND_DEBIT · FUND_REVERSE" ==> ADV_R
ADV_O == "c_dollar: FUND_CREDIT (MICRO deposit)" ==> ADV_R
ADV_R -. "GRANT_AUTH · REVOKE_AUTH · VERIFY_ACK" .-> ADV_O
ENV(["env_i · Nacha phase clock
window_count ≤ 3"])
G(["g · global_invariants
Φ2 settlement counters"])
ENV -.-> BANKS
BANKS -.-> G
classDef adv fill:#ddf4ff,stroke:#0969da,color:#0a3069
classDef bank fill:#eaeef2,stroke:#57606a,color:#1f2328
classDef aux fill:#fbefff,stroke:#8250df,color:#3e1f79
class O0,O1,R0,R1 adv
class ODFI,ACH,RDFI bank
class ENV,G aux
```
Instances and parameters from MODULE main in models/99_main.smv; message routes from the c_info_req_* / c_dollar_req_* assignments in each participant module. All c_info traffic shares one single-slot bus; all money movement shares the c_dollar bus.
The runner merges fragments, checks properties and compiles results.
```mermaid
flowchart LR
A["models/*.smv
8 ordered fragments"] -- "tools/merge_smv.sh
make merge check-merge" --> B["_main_model.smv"]
B -- "show_property" --> C{"index
discovery"}
C -- "3 × INVARSPEC" --> D["Φ1, Φ3: IC3
Φ2: incremental BMC, k = 60"]
C -- "37 × SPEC EF p" --> E["BMC of !(p)
een-sorensson, k = 50"]
D --> F["spec-{0,1,2}-INV.log"]
E --> G["spec-{0..36}-CTL.log"]
F --> H["tools/nuxmv_model_stats.py"]
G --> H
H --> I[("model_stats.md")]
classDef io fill:#ddf4ff,stroke:#0969da,color:#0a3069
classDef run fill:#fbefff,stroke:#8250df,color:#3e1f79
class A,B,F,G,I io
class D,E,H run
```
## Quick start
Run from the repository root with Docker, or use the bundled binary on Linux/macOS with Bash, make, Python 3 and diff:
```bash
./run_all_artifact.sh
# Native alternative:
./run_all_artifact.sh --no-docker
# Compose alternative:
ARTIFACT_CPUS=4 ARTIFACT_MEMORY=8g JOBS=3 docker compose run --build --rm artifact
```
A complete run can take hours. See [pipeline setup, modes, resources and troubleshooting](docs/verification-pipeline.md).
## Results summary
The nuXmv 2.1.0 reference run contains all 40 results. Bounded-inconclusive EF results do not establish unreachability.
| Property | Attack | Unified index | Engine | Result | Counterexample | Log |
| --- | --- | --- | --- | --- | --- | --- |
| Φ1 Continuous Authorization | A1 creditBLEED | 37 | IC3 | false | 10 states | [`spec-0-INV.log`](saved-reference-results/local-run-20261029-artifact-reviews/spec-0-INV.log) |
| Φ2 Zero-Sum Invariant | A2 creditMINT | 38 | incremental BMC, k = 60 | false | 30 states | [`spec-1-INV.log`](saved-reference-results/local-run-20261029-artifact-reviews/spec-1-INV.log) |
| Φ3 Settlement Finality | A3 creditRESET | 39 | IC3 | false | 9 states | [`spec-2-INV.log`](saved-reference-results/local-run-20261029-artifact-reviews/spec-2-INV.log) |
The Φ2 counterexample replays the creditMINT sequence: the micro-deposit credit settles and the account is linked, then the recovery debit is returned `R16`, which leaves one settled credit and no settled debit in the verification cycle. See the [step-by-step walkthrough](saved-reference-results/README.md#creditmint-witness) and the [property semantics](docs/properties.md).
## Documentation
| Document | Purpose |
| --- | --- |
| [Model architecture](models/README.md) | Participants, channels, assumptions and extensions |
| [Verification pipeline](docs/verification-pipeline.md) | Commands, environment, outputs and troubleshooting |
| [Properties](docs/properties.md) | Safety obligations, reachability and verdict semantics |
| [Reference results](saved-reference-results/README.md) | Provenance, all 40 results and integrity checks |
| [Tools](tools/README.md) | Utility interfaces and binary licensing |
## Repository layout
Source fragments, tools and retained results occupy separate repository directories.
```text
ach-model-checking/
├── README.md ← you are here
├── run_all_artifact.sh # one-command pipeline: merge → nuXmv checks → stats
├── Makefile # merge · check-merge · verify · smoke · test · clean
├── Dockerfile # debian:bookworm-slim, linux/amd64, nuXmv + python3
├── compose.yaml # resource-limited Compose service "artifact"
├── docker/entrypoint.sh # container entry → run_all_artifact.sh --inside-docker
├── models/ # the formal model, merged in filename order → models/README.md
│ ├── 00_types_channels.smv # c_info / c_dollar buses, env phase clock
│ ├── 10_originator.smv # Originator FSM + transaction_slot
│ ├── 20_odfi.smv # ODFI
│ ├── 30_ach_operator.smv # ACH Operator
│ ├── 40_rdfi.smv # RDFI
│ ├── 50_receiver.smv # Receiver
│ ├── 60_global_constraints.smv # global observer for Φ2
│ └── 99_main.smv # composition, Φ1–Φ3, 37 EF witnesses, JUSTICE
├── tools/ # merge, statistics, smoke test, nuXmv binaries → tools/README.md
├── tests/fixtures/minimal/ # two-fragment model used by `make test`
├── docs/ # layered documentation + figures (docs/assets/)
└── saved-reference-results/ # complete reference run with logs and provenance → saved-reference-results/README.md
```
## Data and privacy
Our empirical evaluation uses real consumer bank accounts owned by members of the research team. Raw account screenshots and transaction logs cannot be shared publicly because they contain personally identifiable financial information. The paper provides the redacted screenshots documenting the user-visible state changes for each CWD instance (A1: creditBLEED Attack, A2: creditMINT Attack, and A3: creditRESET Attack).
## Citation
```bibtex
@inproceedings{malik2026credit,
title = {Credit Without Debit: Formal Verification of Banking Payment Systems},
author = {Malik, Ali Hamza and Raza, Muhammad Taqi},
booktitle = {Proceedings of the ACM Conference on Computer and Communications Security},
year = {2026}
}
```
To cite the artifact itself, use the Zenodo record: .
## License
[MIT](LICENSE) covers model sources and scripts. Bundled nuXmv binaries are FBK’s and are shipped as provided under their own licensing terms.