Inspect it. Challenge it.
Reproduce it. Report it.
A reviewer workspace for engineers, technicians, researchers, investors and students. It links to real SWI source files, test suites, evidence packages, ZIPs and GitHub Actions. It does not execute or authorize the SWI kernel from the browser.
What the current README declares
Review path
Read the exact source, commit, equation, JSON and ZIP manifest.
Run the repository's documented Python / Z3 tests or inspect the linked CI run.
Record environment, source SHA, commands, outcomes, limitations and artifact hashes.
Repository map
Each project is shown according to its own README and declared role. Related repositories are not automatically treated as part of the V2 authorization spine.
Equation & test sources
Inspect the governing formula and the actual Python tests. The browser does not substitute a JavaScript imitation for the repository's Python implementation.
Governing permit formula
The formula requires all requested conditions to be covered by law, governance, system and human-authority sets, plus non-empty evidence. Unknown inputs must not be treated as success.
Reproduce the scoped equation tests
Run from the V2 repository root after installing its declared requirements. Record the exact commit SHA and environment.
Existing recorded result
The JSON record is historical and is bound to its own base commit. It is not a claim that 31 tests were just rerun on current main.
Inspect recorded JSON ↗Formal model (Z3)
The V2 repository contains a law-authority model and a dedicated adversarial workflow. Review the model and its workflow before interpreting any solver result.
ZIP, JSON & model artifacts
Open or download source artifacts directly. Upload a reviewer-supplied artifact below to calculate its SHA-256 and inspect JSON text where possible. ZIP files are fingerprinted, not executed or extracted in the browser.
V2 repository packages
Upload a review artifact for inspection
What the browser checks
Red-team review
Use the existing adversarial test suite to challenge the actual implementation. Every entry below is a repository file or workflow, not a fabricated browser simulation.
| Attack surface | Existing test | Review focus |
|---|---|---|
| Execution integrity | test_execution_integrity.py ↗ | Protected execution boundary, failures and recovery |
| Gate pass/fail matrix | test_gate_pass_fail_matrix.py ↗ | PASS / FAIL / system-error outcomes |
| Universal halt | test_universal_halt.py ↗ | Fail-closed halt behaviour |
| Authority non-escalation | test_authority_non_escalation.py ↗ | Observation, signature and capability do not create authority |
| New information boundary | test_new_information_boundary_attacks.py ↗ | Changed or newly supplied information |
| Observation / claim separation | test_observation_claim_separation.py ↗ | Observation is not admission or claim proof |
| M11 seal boundary | M11 test directory ↗ | Admission and seal boundary attacks |
One-click test workbench
Run allow-listed SWI V2 test profiles on your own device against an exact Git commit. The runner creates an isolated Python environment, installs the repository's declared requirements, records the environment and command output, then builds evidence files. It does not silently repair source or grant authority.
Local runner connection
CHECKINGStart on Termux/Linux: python run_swi_local.py · Windows: py run_swi_local.py. The runner binds only to 127.0.0.1 and never receives a GitHub write token.
Choose a real repository test
Each run checks out a detached commit in a separate workspace. It does not alter your existing SWI checkout. Test commands are fixed in the runner; arbitrary commands and uploaded code are not executed.
Run status & results
Repair findings — current main evidence
The CI test collector cannot import AdmittedRoute. A full recovery source exists on a repair branch, but that branch is not proof until its exact commit is tested.
The tracked file contains a placeholder string, while the test package imports inspector symbols. The repository contains a reconstruction patch; no historical implementation blob was found.
Reports & CI history
Live GitHub Actions history is fetched from the V2 repository API when this page loads. The report below is an existing source artifact tied to its own commit.
Latest Actions runs
Existing execution-integrity report
This report is bound to branch feature/universal-halt-error-and-reports and its stated commit—not the current main commit.
Read existing report ↗Prepare & submit a review
Create a structured review record bound to a repository, source commit, test command and artifact hash. The report is generated from what the reviewer enters; it never invents test results.
Review details
Generated review output
Publish the review
The suggested destination is public-reviews/inbox/. That folder is included as a starter template in this package but has not yet been created in the remote V2 repository. After the folder is added through an approved commit, reviewers can use GitHub's upload interface. GitHub login and a commit / pull request are required.
This opens GitHub's upload UI; it does not silently push. The static Pages site has no GitHub write token. Publication is complete only after GitHub confirms the commit or pull request.
Understand SWI
A quick explanation for people approaching the project from different disciplines.
For technicians
Treat SWI as a controlled workflow gate. A request is not executed just because it arrived, matches a rule, carries a signature, or has a passing test. Inspect inputs, boundaries, evidence and the exact failure path.
For developers
Trace each claim to source, tests, commit, environment and reproducible output. Keep the Python governing spine authoritative. Browser UI, Node display code and reports must not create a parallel permit path.
For researchers / students
Separate mathematical model assumptions from implementation behaviour. Reproduce one bounded claim at a time. Document counterexamples and unknowns; do not turn an experiment into a system-wide conclusion.
For investors / institutions
This portal exposes what is present, what has been tested, what remains open and what is not claimed. It is not a certification dashboard or a declaration of production readiness.
For security reviewers
Challenge replay, stale evidence, authority escalation, observation-to-claim conversion, bypasses, missing inputs and fail-closed behaviour. Report source SHA and reproducible steps.
For AI governance reviewers
The core distinction is that data, evidence, admission, authorization and action are separate. Model output is not authority. The human-authority boundary is not inferred from model confidence.
SWI's status vocabulary
Related repository boundaries
- V1.1 (repository slug SWI-V1-Module-1-10) is described as Modules 00–11; its README reports bounded seals, while Foundation Seal 5 is NOT READY and the Universal Gate is NOT PROVEN.
- Mathematical Evidence Engine is marked under development; its README asks for a Gemini API key, so it is not a production-ready authority path.
- Node Access Boundary describes a Stage 1 single-process simulation, not real network nodes.
- V4 Beagle is a separate audit / build / test control plane and explicitly does not authorize or seal SWI.
- V1.03 API Foundation is legacy/stub, not the active V1→V2 evidence path.
- Firefly Memory development is suspended and is not included as an active V2 runtime dependency.