A provenance system must show that a registered capture device's credential authorized the submission of the content. In an ordinary design, whoever checks the device and records the content can link the two, and that link is a map of who photographed what. The Birthmark Protocol never builds the map: it splits the work across independently operated components so that no single component holds what it takes to link a device to a record.
This repository is the companion repository of the two papers in Docs/: "The Birthmark Protocol: Achieving Semantic Non-Assembly in Media Provenance" (the Protocol paper) and "Birthmark Protocol Integration: Capture Device, Credential Validator and Edit Workflow" (the Integration paper). The Integration paper cites it as https://github.com/Birthmark-Standard/Protocol.
| Folder | What it holds | Where it appears in the papers |
|---|---|---|
Docs/ |
The Protocol paper, the Integration paper and the Messaging Structure specification | |
ProverifModels/ |
ProVerif models of the privacy and integrity properties, each with a twin that carries one deliberate fault | Protocol paper, Appendix A and Table A.1 |
RoutingModels/ |
Small simulation and enumeration scripts for the device routing table: the validator's tag a first relay can infer, the skew of F and I, and how often roles coincide | Integration paper, Section 3.9; Protocol paper, Section 6.3 |
SNAExperiment/ |
A simulation of the protocol's full message flow that measures how often each component, and a passive observer, names the device behind a registry record | Protocol paper, Appendix D; Integration paper, Section 7 |
Benchmark/ |
A harness that measures the device-side cost of a capture and the cost of proving a SHA-256 hash in zero knowledge, for a Raspberry Pi 4. The committed results are dry runs on a virtual machine, not Pi measurements. | Protocol paper, Appendix B and Table A.3, and Integration paper, Section 8, which calculate these costs from published benchmarks |
Each folder has a README that says what it holds, how to run it and what the output means.
| Folder | Needs | Command | Time |
|---|---|---|---|
ProverifModels/ |
ProVerif 2.05 | proverif BM_Baseline_Noncorrelation.pv, one model at a time, from ProverifModels/ |
one model per command |
RoutingModels/ |
Python 3 and NumPy | python run_all.py > output.txt, then compare with expected_output.txt |
about 1 second |
SNAExperiment/ |
Python 3.10 or later, NumPy, SciPy and Matplotlib | python -m sna run from SNAExperiment/ |
about 8.7 hours on 8 worker processes |
Benchmark/ |
A Raspberry Pi 4 with the packages python3 check_prereqs.py lists |
python3 run_partA.py for Part A; python3 run_partB.py setup, prove and hash for Part B, from Benchmark/ |
not recorded for a Pi |
The smallest check of SNAExperiment/ is its test suite: python -m pytest tests -q from SNAExperiment/, about one minute. python -m sna run --runs 2 --out results/quick runs every step of the experiment on two runs per cell, in about 20 minutes on a four-core machine.
See LICENSING.md. The licence text is in LICENSE.