- C++ 77.7%
- Python 19.7%
- CMake 1.3%
- Lean 0.9%
- Shell 0.3%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
|
All checks were successful
clang-format / check-format (pull_request) Successful in 1m12s
markdownlint / markdown-lint (pull_request) Successful in 42s
formal / definitions (pull_request) Successful in 2m22s
ci / smoke (pull_request) Successful in 35m8s
ci / sanitizers (pull_request) Successful in 1h21m38s
markdownlint / markdown-lint (push) Successful in 17s
clang-format / check-format (push) Successful in 29s
formal / definitions (push) Successful in 1m7s
ci / smoke (push) Successful in 31m48s
ci / sanitizers (push) Successful in 2h10m57s
|
||
| .forgejo | ||
| analysis | ||
| artifacts | ||
| cli | ||
| cmake | ||
| core | ||
| docs | ||
| formal | ||
| io | ||
| metrics | ||
| protocols/protocol-2.2 | ||
| research | ||
| schemas | ||
| sim | ||
| tests | ||
| tools | ||
| validation/sel4/environment | ||
| workload | ||
| .clang-format | ||
| .clang-tidy | ||
| .editorconfig | ||
| .gitattributes | ||
| .gitignore | ||
| .markdownlint.json | ||
| .mdl_style.rb | ||
| .mdlrc | ||
| .pre-commit-config.yaml | ||
| CMakeLists.txt | ||
| LICENSE | ||
| README.md | ||
AES — Authority Evolution Simulator
AES (Authority Evolution Simulator) is a research simulator for studying how authority evolves over time in capability-based systems.
Rather than modeling a concrete operating system or kernel, AES treats authority as a temporal graph and focuses on long-term dynamics such as revocation, provenance, epistemic limits, persistence, and collapse. The simulator is designed to support controlled, reproducible experiments that expose structural limits of authority management in long-lived systems.
What AES is (and is not)
AES is:
- a research simulator
- a temporal model of authority and delegation
- a tool for empirical studies accompanying academic publications
AES is not:
- an operating system or microkernel
- a kernel emulator
- a security product
- a performance benchmark
The goal is explanatory power, not deployment.
Core design invariants
AES is built around a small number of non-negotiable invariants:
- All authority changes occur via explicit events
- Authority state is fully reconstructible from history
- Simulation runs are deterministic given a seed
- Metrics are pure functions over history or snapshots
- There is a strict separation between ground truth and observation
These invariants are established in Phase 0 and preserved throughout all later phases.
Research structure
Development is organized into Phases 0–6. Phases are cumulative: each phase weakens an assumption made by earlier ones.
| Phase | Focus |
|---|---|
| Phase 0 | Core simulator infrastructure and invariants |
| Phase 1 | Revocation semantics under full knowledge |
| Phase 2 | Summary adequacy and metadata reclamation |
| Phase 3 | Epistemic limits of authority reasoning |
| Phase 4 | Revocation horizons and historical limits |
| Phase 5 | Memory, persistence, and authority residue |
| Phase 6 | Predicting authority collapse |
Phases are tracked as milestones in the issue tracker.
Tasks within a phase are tracked as issues, with checklist items
corresponding to verifiable work and acceptance outcomes.
The Phase 2 protocol fixes its scope, requirements
and evidence.
The legacy ledger records preserved
behavior. The policy scope documentation
describes the independent grant and withdrawal components.
Repository structure
The repository is organized by responsibility, not by phase:
cli/ Command-line tools
core/ Core data structures and invariants
docs/ Specifications and phase documentation
io/ Event logs, replay, and serialization
metrics/ Measurement and analysis code
sim/ Simulation engine and execution logic
tests/ Determinism, invariants, and correctness tests
tools/ Auxiliary scripts and helpers
workload/ Workload and stress-test generators
Workflow overview
- Milestones represent Phases (0–6)
- Issues represent reviewable outcomes or parents of linked tasks
- Issue checklists represent verified work and acceptance steps
- Branches and PRs correspond to individual issues, including closure records
- Labels describe semantics (e.g.
revocation,metrics,epistemic) - A single Kanban board is used for daily execution only
Reproducibility
AES is designed so that:
- identical seeds produce identical event traces,
- all results can be regenerated from raw logs,
- figures and tables are derived from scripts, not manual steps.
Reproducibility is treated as a research requirement, not an afterthought.
Workload-Family Experiments
AES sweep mode now emits one row per run in experiment_runs.csv in addition
to existing sweep artifacts. This enables workload-family reporting over
(strategy, depth, fan_out, seed) instead of single-graph snapshots.
Default Phase 1 workload-family matrix and analysis scripts:
- config:
tools/figures/phase1/sweep_workload_family_default.json - run + validate + aggregate + plots:
tools/regen-workload-family.sh - summary table output:
results/summary/by_strategy_depth_fanout.csv
License
AES is released under the MIT License.
See LICENSE for details.
Status
This repository is under active research-driven development.
Interfaces may evolve between phases, but core invariants are preserved
deliberately.
If you are reading this as part of an academic artifact evaluation:
the recommended entry point is
Phase 0 documentation in docs/.
Citation
A formal citation will be added once the first associated publication is available.