Sovereign realizability topos and non-RLHF agentic AI kernel. 254 domains. λ < 1 λ<1 contraction-to-truth. Merkle-anchored deterministic toolchain
1
stars
2,097
commits
Python
primary language
Aug 11, 2026
updated
A proof-carrying Python repository implementing 254 domain invariants under the Yeshua Standard — exact arithmetic (Fraction-only), capability-gated kernel, SHA-256 Merkle anchoring, and 20 CI workflows.
git clone https://github.com/aidoruao/orthogonal-engineering.git
cd orthogonal-engineering
python3 tools/verify_all.py # 10-check verification suite
python3 audit/popperian_audit.py # audit all domain invariants

Every function returns Tuple[bool, ProofObject]. Every claim has a
falsifies_if condition. Every number is a Fraction, not a float.
Every file is SHA-256 anchored in a Merkle tree. 20 CI workflows enforce
these rules on every push and PR.
src/domains/STANDARDS_REGISTRY.jsonmerkle/global_root.json)kernel/run_all_invariants() passingEight non-negotiable axioms. Every function, every commit, every artifact obeys these:
| # | Axiom | Engineering Meaning |
|---|---|---|
| 1 | Every truth is derivable | No bare assertions — use ProofObject |
| 2 | Every derivation is reproducible | Deterministic: same input = same output |
| 3 | Every mutation is re-verifiable | Hash-anchored artifacts, append-only logs |
| 4 | No authority without proof | No assert — return Tuple[bool, ProofObject] |
| 5 | No hidden state | Glass-Box: all state inspectable |
| 6 | No unverifiable dependency | Standard library preferred; all deps documented |
| 7 | No economic gatekeeping | No paid-only features gate correctness |
| 8 | Every artifact is hash-anchored | SHA-256 on all evidence |
| Path | Purpose | Count |
|---|---|---|
src/domains/ | Domain invariant packages (each with invariants.py, implementation.py) | 254 |
src/noways/ | Impossibility proofs (what cannot be done) | 3 files |
src/enumerations/ | Bounded catalogs (magic numbers, hidden failures, black-box antipatterns) | 7 files |
kernel/ | Capability-gated OS kernel (boot, IPC, MMU, scheduler, commonwealth) | 15+ modules |
axioms/ | Formal foundations (Peano, logic, category theory, topology, etc.) | 37 modules |
tools/ | Utility scripts (consent, standards check, taxonomy, frontmatter audit) | 16 scripts |
tests/ | Pytest test suite | matches src/ |
audit/ | Popperian audit engine + reports | 2 files |
automation/ | CI enforcement scripts (PR #49 guard, full audit with trace) | 5 files |
pr47_stewardship/ | Append-only witness chain + consent log | JSONL files |
documentation/ | Glass-Box Boundary HTML blueprint + SHA-256 manifests | 2 dirs |
.github/workflows/ | CI: fraction enforcement, no-floats, frontmatter, determinism, state witness | 20 workflows |
Run these before and after every change. If any fails, stop and fix before proceeding.
# 1. Feed integrity (hash chain)
python tools/state_witness/generate_feed_entry.py --verify
# 2. Popperian audit (all domain invariants)
python audit/popperian_audit.py
# 3. Standards registry compliance
python tools/standards_check.py --verify
# 4. Full test suite
pytest tests/ -q
# 5. Complete 10-check verification suite (PR #152)
python tools/verify_all.py
New AI instances and GitHub Copilot agents should start here:
One-command onboarding:
python tools/onboard_agent.py --agent <type>
Every domain in src/domains/ follows the same pattern: invariants.py defines
falsifies_if conditions, implementation.py provides the proof-carrying
functions, and tests/test_<domain>.py verifies them.
The 1B/1Qi LOC generators exist in generators/. They demonstrate deterministic
fractal expansion from minimal seed definitions. See
generators/README.md for details.
See repository license file for details.
This infrastructure is free forever guaranteed — not by law, but by architecture. See COVENANT.md for the architectural guarantee.
aidoruao manual: need to fix the issue with ai giving rm- rf whatever its called and me blindly pasting commands into terminal deleting 100 commits of history, I need to have the operational ai to give an entire huge kb of any file type of command plan and causal explanations and everything, which will be audited
1,289 commits
683 commits
88 commits
32 commits
Python
67.0%
Java
23.2%
HTML
6.4%
Sovereign realizability topos and non-RLHF agentic AI kernel. 254 domains. λ < 1 λ<1 contraction-to-truth. Merkle-anchored deterministic toolchain
1
stars
2,097
commits
Python
primary language
Aug 11, 2026
updated
A proof-carrying Python repository implementing 254 domain invariants under the Yeshua Standard — exact arithmetic (Fraction-only), capability-gated kernel, SHA-256 Merkle anchoring, and 20 CI workflows.
git clone https://github.com/aidoruao/orthogonal-engineering.git
cd orthogonal-engineering
python3 tools/verify_all.py # 10-check verification suite
python3 audit/popperian_audit.py # audit all domain invariants

Every function returns Tuple[bool, ProofObject]. Every claim has a
falsifies_if condition. Every number is a Fraction, not a float.
Every file is SHA-256 anchored in a Merkle tree. 20 CI workflows enforce
these rules on every push and PR.
src/domains/STANDARDS_REGISTRY.jsonmerkle/global_root.json)kernel/run_all_invariants() passingEight non-negotiable axioms. Every function, every commit, every artifact obeys these:
| # | Axiom | Engineering Meaning |
|---|---|---|
| 1 | Every truth is derivable | No bare assertions — use ProofObject |
| 2 | Every derivation is reproducible | Deterministic: same input = same output |
| 3 | Every mutation is re-verifiable | Hash-anchored artifacts, append-only logs |
| 4 | No authority without proof | No assert — return Tuple[bool, ProofObject] |
| 5 | No hidden state | Glass-Box: all state inspectable |
| 6 | No unverifiable dependency | Standard library preferred; all deps documented |
| 7 | No economic gatekeeping | No paid-only features gate correctness |
| 8 | Every artifact is hash-anchored | SHA-256 on all evidence |
| Path | Purpose | Count |
|---|---|---|
src/domains/ | Domain invariant packages (each with invariants.py, implementation.py) | 254 |
src/noways/ | Impossibility proofs (what cannot be done) | 3 files |
src/enumerations/ | Bounded catalogs (magic numbers, hidden failures, black-box antipatterns) | 7 files |
kernel/ | Capability-gated OS kernel (boot, IPC, MMU, scheduler, commonwealth) | 15+ modules |
axioms/ | Formal foundations (Peano, logic, category theory, topology, etc.) | 37 modules |
tools/ | Utility scripts (consent, standards check, taxonomy, frontmatter audit) | 16 scripts |
tests/ | Pytest test suite | matches src/ |
audit/ | Popperian audit engine + reports | 2 files |
automation/ | CI enforcement scripts (PR #49 guard, full audit with trace) | 5 files |
pr47_stewardship/ | Append-only witness chain + consent log | JSONL files |
documentation/ | Glass-Box Boundary HTML blueprint + SHA-256 manifests | 2 dirs |
.github/workflows/ | CI: fraction enforcement, no-floats, frontmatter, determinism, state witness | 20 workflows |
Run these before and after every change. If any fails, stop and fix before proceeding.
# 1. Feed integrity (hash chain)
python tools/state_witness/generate_feed_entry.py --verify
# 2. Popperian audit (all domain invariants)
python audit/popperian_audit.py
# 3. Standards registry compliance
python tools/standards_check.py --verify
# 4. Full test suite
pytest tests/ -q
# 5. Complete 10-check verification suite (PR #152)
python tools/verify_all.py
New AI instances and GitHub Copilot agents should start here:
One-command onboarding:
python tools/onboard_agent.py --agent <type>
Every domain in src/domains/ follows the same pattern: invariants.py defines
falsifies_if conditions, implementation.py provides the proof-carrying
functions, and tests/test_<domain>.py verifies them.
The 1B/1Qi LOC generators exist in generators/. They demonstrate deterministic
fractal expansion from minimal seed definitions. See
generators/README.md for details.
See repository license file for details.
This infrastructure is free forever guaranteed — not by law, but by architecture. See COVENANT.md for the architectural guarantee.
aidoruao manual: need to fix the issue with ai giving rm- rf whatever its called and me blindly pasting commands into terminal deleting 100 commits of history, I need to have the operational ai to give an entire huge kb of any file type of command plan and causal explanations and everything, which will be audited
1,289 commits
683 commits
88 commits
32 commits
Python
67.0%
Java
23.2%
HTML
6.4%