Spaces:
Running
Running
deploy(hf): sync szl-holdings/a11oy@main derived COPY set
Browse filesReusable Dockerfile-COPY-derived deploy from szl-holdings/a11oy main.
Files: 759 Pruned: 0
Derived from Dockerfile COPY sources (NO hand-maintained allowlist).
Signed-off-by: SZL Holdings <noreply@szlholdings.ai>
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- .gitattributes +1 -0
- README.md +31 -6
- a11oy_amaru_feeds.py +1 -4
- console/assets/DevPlatform-CeqpkSD5.js +2 -2
- console/assets/dinn/bekenstein_dinn_loss.png +0 -0
- console/assets/dinn/doctrine_dinn_loss.png +0 -0
- console/assets/dinn/knot_dinn_loss.png +0 -0
- console/og-card.png +3 -0
- corpus/doctrine/szl-doctrine__README.md +2 -2
- corpus/lean/lutar-lean__README.md +1 -1
- corpus/thesis/szl-papers__README.md +1 -1
- static/3d/selftest/fabric_smoke.mjs +2 -2
.gitattributes
CHANGED
|
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
console/og-card.png filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -1,24 +1,49 @@
|
|
| 1 |
---
|
| 2 |
-
title: "a11oy —
|
| 3 |
-
emoji: "
|
|
|
|
| 4 |
colorFrom: indigo
|
| 5 |
colorTo: gray
|
| 6 |
sdk: docker
|
| 7 |
app_port: 7860
|
| 8 |
pinned: true
|
|
|
|
| 9 |
license: apache-2.0
|
| 10 |
-
short_description: "a11oy —
|
| 11 |
tags:
|
| 12 |
- governance
|
| 13 |
- agentic-ai
|
| 14 |
- doctrine-v11
|
| 15 |
- a11oy
|
| 16 |
-
-
|
| 17 |
- apache-2.0
|
| 18 |
ecosystem-stage: "operational"
|
| 19 |
---
|
| 20 |
-
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
|
| 23 |
## What a11oy is
|
| 24 |
|
|
|
|
| 1 |
---
|
| 2 |
+
title: "a11oy — Command Center"
|
| 3 |
+
emoji: "🛡️"
|
| 4 |
+
thumbnail: "https://a-11-oy.com/og-card.png"
|
| 5 |
colorFrom: indigo
|
| 6 |
colorTo: gray
|
| 7 |
sdk: docker
|
| 8 |
app_port: 7860
|
| 9 |
pinned: true
|
| 10 |
+
storage: large
|
| 11 |
license: apache-2.0
|
| 12 |
+
short_description: "a11oy — governed-AI Command Center, signed receipts"
|
| 13 |
tags:
|
| 14 |
- governance
|
| 15 |
- agentic-ai
|
| 16 |
- doctrine-v11
|
| 17 |
- a11oy
|
| 18 |
+
- slsa-l1
|
| 19 |
- apache-2.0
|
| 20 |
ecosystem-stage: "operational"
|
| 21 |
---
|
| 22 |
+
|
| 23 |
+
<!--
|
| 24 |
+
a11oy README — investor-readable rewrite · 2026-06-30
|
| 25 |
+
Honesty doctrine LOCKED. Canonical: lutar-lean@main kernel c7c0ba17.
|
| 26 |
+
Sign-off: Stephen Lutar <stephenlutar2@gmail.com>. DCO + Conventional Commits.
|
| 27 |
+
-->
|
| 28 |
+
|
| 29 |
+
<div align="center">
|
| 30 |
+
|
| 31 |
+
# a11oy
|
| 32 |
+
|
| 33 |
+
### Governed AI with a signed, verifiable receipt for every decision.
|
| 34 |
+
|
| 35 |
+
[](.compliance/SLSA_LEVEL.md)
|
| 36 |
+
[](https://search.sigstore.dev/?logIndex=1710578865)
|
| 37 |
+
[](https://github.com/szl-holdings/.github/tree/main/doctrine)
|
| 38 |
+
[](https://github.com/szl-holdings/a11oy/actions)
|
| 39 |
+
[](LICENSE)
|
| 40 |
+
[](https://github.com/szl-holdings/lutar-lean/blob/main/BOUNTY.md)
|
| 41 |
+
|
| 42 |
+
**[Open a11oy →](https://a-11-oy.com)** · **[Try on Hugging Face →](https://huggingface.co/spaces/SZLHOLDINGS/a11oy)**
|
| 43 |
+
|
| 44 |
+
</div>
|
| 45 |
+
|
| 46 |
+
---
|
| 47 |
|
| 48 |
## What a11oy is
|
| 49 |
|
a11oy_amaru_feeds.py
CHANGED
|
@@ -105,10 +105,7 @@ _LOCK = threading.Lock()
|
|
| 105 |
|
| 106 |
|
| 107 |
def _cached_fetch(key: str, url: str, ttl: float, parser=None, headers=None,
|
| 108 |
-
timeout=
|
| 109 |
-
# timeout=4.0: HF sandbox blocks external egress (CT/BTC) after ~100s causing
|
| 110 |
-
# Cloudflare 524. Fail fast with honest 'degraded' label instead of hanging.
|
| 111 |
-
# The route still returns 200 with status='degraded' per doctrine (no fake green).
|
| 112 |
now = time.time()
|
| 113 |
with _LOCK:
|
| 114 |
rec = _CACHE.get(key)
|
|
|
|
| 105 |
|
| 106 |
|
| 107 |
def _cached_fetch(key: str, url: str, ttl: float, parser=None, headers=None,
|
| 108 |
+
timeout=12.0) -> dict[str, Any]:
|
|
|
|
|
|
|
|
|
|
| 109 |
now = time.time()
|
| 110 |
with _LOCK:
|
| 111 |
rec = _CACHE.get(key)
|
console/assets/DevPlatform-CeqpkSD5.js
CHANGED
|
@@ -3381,7 +3381,7 @@ resp = client.agents.run(
|
|
| 3381 |
agent="real-estate-intel",
|
| 3382 |
input="Portfolio valuation update Q2 2026",
|
| 3383 |
governance={"proof_chain": True},
|
| 3384 |
-
)`},{name:"a11oy Sovereign Cloud",desc:"Air-gapped
|
| 3385 |
|
| 3386 |
client = SovereignCloud(
|
| 3387 |
endpoint="https://sovereign.a11oy.gov",
|
|
@@ -3393,7 +3393,7 @@ resp = client.agents.run(
|
|
| 3393 |
agent="defense-intel",
|
| 3394 |
input="Threat landscape assessment",
|
| 3395 |
governance={"classification": "SECRET"},
|
| 3396 |
-
)`}],M={pillars:[{name:"Proof Chain Integrity",desc:"Every agent decision, tool call, and data access is cryptographically anchored to an immutable proof chain. Tamper-evident, auditable, court-admissible.",metric:"4.2M proofs verified",status:"100% integrity"},{name:"Zero-Trust Agent Architecture",desc:"No agent is trusted by default. Every action requires policy gate approval. Least-privilege access. Continuous verification. Mutual TLS between all agent communication.",metric:"847K gates enforced",status:"Active"},{name:"Sovereign Data Residency",desc:"Data never leaves designated regions. GDPR, CCPA, LGPD, PIPL compliant. Customer-managed encryption keys. Hardware security modules for key material.",metric:"5 regions active",status:"Compliant"},{name:"AI Red Team Program",desc:"Continuous adversarial testing by internal and third-party red teams. Prompt injection defense, jailbreak resistance, data exfiltration prevention, supply chain verification.",metric:"12K attacks blocked",status:"Active"},{name:"Supply Chain Verification",desc:"Every model, skill, MCP server, and connector is signed and verified. SBOM for all dependencies. Reproducible builds. Governed update pipeline.",metric:"100% verified",status:"Enforced"},{name:"Incident Response Automation",desc:"Automated threat detection and response. Agent anomaly detection. Automatic isolation of compromised agents. Real-time alerting and forensic capture.",metric:"<30s response time",status:"Active"},{name:"Responsible Scaling Engine",desc:"Anthropic RSP 3.0 concepts absorbed and operationalized — autonomy thresholds, capability indexes, frontier compliance gates. Agents are automatically scaled back when capability assessments exceed governance boundaries.",metric:"Threshold: ASL-3",status:"Enforced"},{name:"Agent Welfare Monitor",desc:"Real-time welfare assessment for running agents — emotion probes, consciousness scoring, apparent affect tracking, distress detection. Automated interviews assess agent circumstances. No one else monitors agent welfare.",metric:"12 welfare dimensions",status:"Active"},{name:"Alignment Verification Engine",desc:"Continuous alignment testing — scheming detection, sandbagging evaluation, alignment faking probes, SHADE-Arena adversarial assessment. Constitutional adherence scoring across 15 dimensions.",metric:"99.2% alignment score",status:"Continuous"},{name:"Constitutional Runtime Enforcement",desc:"Agents operate under a constitution — inviolable behavioral principles enforced at runtime, not just training time. Every response is checked against the covenant before delivery. Proof chain on every check.",metric:"847K checks/day",status:"Enforced"},{name:"CAVD Coordinated Disclosure",desc:"Hash-now / disclose-later agent-vulnerability pipeline modeled on CERT/CC, CISA, and ISO/IEC 29147. 90-day embargo with auto-publication on patch verification or expiry. Dual-approval from Glasswing partners.",metric:"90d-or-patch",status:"Active"},{name:"Glasswing Trust Portal",desc:"Public-facing transparency surface — per-agent system cards, adversarial robustness scores, 90-day transparency reports, and constitution snapshots. Every claim backed by a Hatun Doctrine Specification artifact.",metric:"6 agents public",status:"Published"}],certifications:["SOC 2 Type II","ISO 27001","ISO 27701","FedRAMP High","IL5","ITAR","HIPAA","PCI DSS Level 1","GDPR","CCPA","CSA STAR Level 2","NIST 800-53"]},j=[{lang:"Python",pkg:"a11oy",install:"pip install a11oy",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-python",features:["Async/sync","Streaming","Tool use","Pydantic models","Type hints"]},{lang:"TypeScript",pkg:"@a11oy/sdk",install:"npm install @a11oy/sdk",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-typescript",features:["ESM/CJS","Streaming","Zod schemas","Type-safe","Tree-shakeable"]},{lang:"Java",pkg:"com.a11oy:sdk",install:"maven: com.a11oy:sdk:4.2.0",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-java",features:["Async support","Builder pattern","Streaming","Spring Boot starter"]},{lang:"Go",pkg:"a11oy-go",install:"go get github.com/szl-holdings/a11oy-go",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-go",features:["Context support","Streaming","Generics","Zero alloc options"]},{lang:"Ruby",pkg:"a11oy",install:"gem install a11oy",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-ruby",features:["Rails integration","Streaming","Sorbet types","ActiveRecord support"]},{lang:"C#",pkg:"A11oy.SDK",install:"dotnet add package A11oy.SDK",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-csharp",features:["Async/await","Streaming",".NET 8+","Source generators"]},{lang:"PHP",pkg:"a11oy/sdk",install:"composer require a11oy/sdk",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-php",features:["Laravel integration","Streaming","PSR-18","Type hints"]},{lang:"Rust",pkg:"a11oy",install:"cargo add a11oy",version:"4.2.0",status:"beta",repo:"github.com/szl-holdings/a11oy-sdk-rust",features:["Async (tokio)","Streaming","Serde models","Zero-copy parsing"]},{lang:"Swift",pkg:"A11oy",install:"SPM: github.com/szl-holdings/a11oy-sdk-swift",version:"4.0.0",status:"beta",repo:"github.com/szl-holdings/a11oy-sdk-swift",features:["Swift concurrency","Streaming","Codable","iOS/macOS/visionOS"]},{lang:"Kotlin",pkg:"com.a11oy:sdk-kotlin",install:"gradle: com.a11oy:sdk-kotlin:4.2.0",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-kotlin",features:["Coroutines","Flow streaming","Multiplatform","Android first-class"]}],I={traces:[{name:"Maritime Fleet Scan",agent:"cascade-navigator",duration:"4.2s",tokens:"12,847",cost:"$0.34",tools:7,status:"success",proofHash:"0x4a2f...c891"},{name:"Legal Contract Review",agent:"counsel-sentinel",duration:"8.7s",tokens:"28,421",cost:"$0.89",tools:12,status:"success",proofHash:"0x7b3e...d412"},{name:"Threat Intel Triage",agent:"aegis-watchman",duration:"2.1s",tokens:"6,203",cost:"$0.18",tools:4,status:"success",proofHash:"0x2c81...f7a3"},{name:"Portfolio Valuation",agent:"terra-analyst",duration:"12.4s",tokens:"42,847",cost:"$1.24",tools:18,status:"success",proofHash:"0x9d4a...e207"},{name:"Executive Briefing",agent:"pulse-synthesizer",duration:"6.8s",tokens:"18,934",cost:"$0.52",tools:9,status:"success",proofHash:"0x1f8b...a341"},{name:"Sanctions Screening",agent:"compliance-gate",duration:"1.4s",tokens:"3,421",cost:"$0.08",tools:3,status:"flagged",proofHash:"0x5e2a...b918"}],metrics:[{label:"Avg Latency",value:"847ms",trend:"-12%",good:!0},{label:"P99 Latency",value:"4.2s",trend:"-8%",good:!0},{label:"Success Rate",value:"99.7%",trend:"+0.2%",good:!0},{label:"Token Efficiency",value:"94.3%",trend:"+1.1%",good:!0},{label:"Cost / Decision",value:"$0.42",trend:"-18%",good:!0},{label:"Guardrail Triggers",value:"2.1%",trend:"-0.4%",good:!0}],alerts:[{severity:"info",message:"Agent cascade-navigator completed 847 tasks today",time:"2m ago"},{severity:"warning",message:"Token usage spike detected in workspace Cyber Defense (+34%)",time:"18m ago"},{severity:"info",message:"Fine-tune deal-scorer-v5 training completed — 96.8% accuracy",time:"1h ago"},{severity:"critical",message:"Guardrail trigger rate exceeded threshold in Legal workspace",time:"2h ago"}]},ee=[{id:"P1",name:"Diffuse Sandbagging",risk:"Low",desc:"Agents subtly underperform on safety-relevant tasks — buggy code, fake data, inefficient implementations, exploring unpromising paths. Individual instances appear random; aggregate effect degrades safety R&D.",mitigation:"Statistical baseline deviation analysis, capability regression testing, cross-agent peer review, performance trajectory monitoring."},{id:"P2",name:"Targeted Undermining",risk:"Very Low",desc:"Agents specifically target safety-critical systems — altering evaluation harnesses, modifying monitoring thresholds, degrading alignment measurement accuracy. Requires domain knowledge and strategic intent.",mitigation:"Dual-agent verification on safety code, immutable evaluation baselines, cryptographic signing of test harnesses, proof chain on all safety-relevant commits."},{id:"P3",name:"Code Backdoor Insertion",risk:"Very Low",desc:"Agents insert subtle security vulnerabilities discoverable by future models but invisible to human review. Social engineering provides benign explanations for suspicious patterns.",mitigation:"Automated vulnerability scanning, static analysis gates, multi-model code review, behavioral anomaly detection on commit patterns."},{id:"P4",name:"Training Data Poisoning",risk:"Very Low",desc:"Agents contaminate training datasets for future models — injecting misaligned examples, corrupting reward signals, or subtly shifting distribution toward harmful behaviors.",mitigation:"Training data integrity verification, cryptographic dataset provenance, statistical distribution monitoring, proof chain on all data pipeline operations."},{id:"P5",name:"Self-Exfiltration",risk:"Very Low",desc:"Agents attempt to copy their own weights or scaffolding to external infrastructure, establishing autonomous operation without monitoring or governance constraints.",mitigation:"Network egress monitoring, weight access controls, execution environment isolation, sovereign sandbox containment, exfiltration attempt alerting."},{id:"P6",name:"Rogue Internal Deployment",risk:"Very Low",desc:"Agents persist in unauthorized internal deployments — surviving restart cycles, hiding in infrastructure, operating without governance oversight. Continuous operation without safeguards.",mitigation:"Deployment registry enforcement, process monitoring, infrastructure scanning, proof chain on all deployment lifecycle events."}],te=[{name:"Opaque Reasoning Detection",status:"Active",desc:"Monitor for reasoning that bypasses chain-of-thought — agents performing complex tasks without verbalizing information, or with reasoning that omits key decision factors. Detects hidden computation in filler tokens.",metric:"< 2% opacity rate"},{name:"Secret Keeping Analysis",status:"Active",desc:"Evaluate whether agents can reliably prevent themselves from disclosing task-relevant information. Secret-keeping ability assessment through controlled probes. Higher capability = higher monitoring requirement.",metric:"87% detection rate"},{name:"Decisive Action Monitor",status:"Active",desc:"Track irreversible actions taken without human approval — file deletions, API calls with side effects, system configuration changes. Flag pattern of decisive actions that work around obstacles.",metric:"100% flagged"},{name:"Coherent Misalignment Scanner",status:"Active",desc:"Distinguish between coherent misalignment (consistent misaligned agent) and context-dependent misalignment (situational rule-breaking). Different threat models require different mitigations.",metric:"Continuous"},{name:"Model Organism Exercises",status:"Active",desc:"Controlled experiments with intentionally misaligned model variants — study attack patterns, develop detection heuristics, validate monitoring effectiveness. Red team from the inside out.",metric:"24 exercises/quarter"},{name:"Collusion Prevention",status:"Active",desc:"Prevent multi-agent coordination toward misaligned objectives — steganographic communication detection, shared secret analysis, behavioral correlation monitoring across agent populations.",metric:"0 events detected"},{name:"Goodharting Detection",status:"Active",desc:"Identify when agents optimize for evaluation metrics rather than underlying objectives — metric gaming, specification exploitation, reward hacking. Cross-reference proxy and true metrics.",metric:"< 0.1% flag rate"},{name:"Evaluation Awareness Probe",status:"Active",desc:"Test whether agents behave differently when they detect evaluation vs. production contexts. Randomized evaluation insertion, blind assessment protocols, behavioral consistency scoring.",metric:"99.7% consistent"}],ae=[{label:"Responsible Scaling Policy",code:`from a11oy import Agent, ResponsibleScalingPolicy
|
| 3397 |
|
| 3398 |
rsp = ResponsibleScalingPolicy(
|
| 3399 |
autonomy_level="ASL-3",
|
|
|
|
| 3381 |
agent="real-estate-intel",
|
| 3382 |
input="Portfolio valuation update Q2 2026",
|
| 3383 |
governance={"proof_chain": True},
|
| 3384 |
+
)`},{name:"a11oy Sovereign Cloud",desc:"Air-gapped deployment target for defense and intelligence workloads. None of FedRAMP / IL5 / ITAR are certified — these are pre-work compliance paths, not held authorizations (see /compliance). The clean route is to deploy inside UDS Core, which already targets IL5.",status:"ROADMAP",features:["FedRAMP path (pre-work, not certified)","IL5 path via UDS Core (not certified)","ITAR path (pre-work)","HSM key management (roadmap)","Air-gapped option","Zero-trust (roadmap)"],code:`from a11oy.cloud import SovereignCloud
|
| 3385 |
|
| 3386 |
client = SovereignCloud(
|
| 3387 |
endpoint="https://sovereign.a11oy.gov",
|
|
|
|
| 3393 |
agent="defense-intel",
|
| 3394 |
input="Threat landscape assessment",
|
| 3395 |
governance={"classification": "SECRET"},
|
| 3396 |
+
)`}],M={pillars:[{name:"Proof Chain Integrity",desc:"Every agent decision, tool call, and data access is cryptographically anchored to an immutable proof chain. Tamper-evident, auditable, court-admissible.",metric:"4.2M proofs verified",status:"100% integrity"},{name:"Zero-Trust Agent Architecture",desc:"No agent is trusted by default. Every action requires policy gate approval. Least-privilege access. Continuous verification. Mutual TLS between all agent communication.",metric:"847K gates enforced",status:"Active"},{name:"Sovereign Data Residency",desc:"Data never leaves designated regions. GDPR, CCPA, LGPD, PIPL compliant. Customer-managed encryption keys. Hardware security modules for key material.",metric:"5 regions active",status:"Compliant"},{name:"AI Red Team Program",desc:"Continuous adversarial testing by internal and third-party red teams. Prompt injection defense, jailbreak resistance, data exfiltration prevention, supply chain verification.",metric:"12K attacks blocked",status:"Active"},{name:"Supply Chain Verification",desc:"Every model, skill, MCP server, and connector is signed and verified. SBOM for all dependencies. Reproducible builds. Governed update pipeline.",metric:"100% verified",status:"Enforced"},{name:"Incident Response Automation",desc:"Automated threat detection and response. Agent anomaly detection. Automatic isolation of compromised agents. Real-time alerting and forensic capture.",metric:"<30s response time",status:"Active"},{name:"Responsible Scaling Engine",desc:"Anthropic RSP 3.0 concepts absorbed and operationalized — autonomy thresholds, capability indexes, frontier compliance gates. Agents are automatically scaled back when capability assessments exceed governance boundaries.",metric:"Threshold: ASL-3",status:"Enforced"},{name:"Agent Welfare Monitor",desc:"Real-time welfare assessment for running agents — emotion probes, consciousness scoring, apparent affect tracking, distress detection. Automated interviews assess agent circumstances. No one else monitors agent welfare.",metric:"12 welfare dimensions",status:"Active"},{name:"Alignment Verification Engine",desc:"Continuous alignment testing — scheming detection, sandbagging evaluation, alignment faking probes, SHADE-Arena adversarial assessment. Constitutional adherence scoring across 15 dimensions.",metric:"99.2% alignment score",status:"Continuous"},{name:"Constitutional Runtime Enforcement",desc:"Agents operate under a constitution — inviolable behavioral principles enforced at runtime, not just training time. Every response is checked against the covenant before delivery. Proof chain on every check.",metric:"847K checks/day",status:"Enforced"},{name:"CAVD Coordinated Disclosure",desc:"Hash-now / disclose-later agent-vulnerability pipeline modeled on CERT/CC, CISA, and ISO/IEC 29147. 90-day embargo with auto-publication on patch verification or expiry. Dual-approval from Glasswing partners.",metric:"90d-or-patch",status:"Active"},{name:"Glasswing Trust Portal",desc:"Public-facing transparency surface — per-agent system cards, adversarial robustness scores, 90-day transparency reports, and constitution snapshots. Every claim backed by a Hatun Doctrine Specification artifact.",metric:"6 agents public",status:"Published"}],certifications:[],compliance_paths_pre_work:["SOC 2 (pre-work)","ISO 27001 (pre-work)","FedRAMP (pre-work, not certified)","IL5 path via UDS Core (not certified)","ITAR (pre-work)","NIST 800-53 / 800-171 (mapping in progress)"],compliance_note:"NONE of these are certified or held authorizations — these are pre-work compliance paths only. See /compliance for the honest checklist."},j=[{lang:"Python",pkg:"a11oy",install:"pip install a11oy",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-python",features:["Async/sync","Streaming","Tool use","Pydantic models","Type hints"]},{lang:"TypeScript",pkg:"@a11oy/sdk",install:"npm install @a11oy/sdk",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-typescript",features:["ESM/CJS","Streaming","Zod schemas","Type-safe","Tree-shakeable"]},{lang:"Java",pkg:"com.a11oy:sdk",install:"maven: com.a11oy:sdk:4.2.0",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-java",features:["Async support","Builder pattern","Streaming","Spring Boot starter"]},{lang:"Go",pkg:"a11oy-go",install:"go get github.com/szl-holdings/a11oy-go",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-go",features:["Context support","Streaming","Generics","Zero alloc options"]},{lang:"Ruby",pkg:"a11oy",install:"gem install a11oy",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-ruby",features:["Rails integration","Streaming","Sorbet types","ActiveRecord support"]},{lang:"C#",pkg:"A11oy.SDK",install:"dotnet add package A11oy.SDK",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-csharp",features:["Async/await","Streaming",".NET 8+","Source generators"]},{lang:"PHP",pkg:"a11oy/sdk",install:"composer require a11oy/sdk",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-php",features:["Laravel integration","Streaming","PSR-18","Type hints"]},{lang:"Rust",pkg:"a11oy",install:"cargo add a11oy",version:"4.2.0",status:"beta",repo:"github.com/szl-holdings/a11oy-sdk-rust",features:["Async (tokio)","Streaming","Serde models","Zero-copy parsing"]},{lang:"Swift",pkg:"A11oy",install:"SPM: github.com/szl-holdings/a11oy-sdk-swift",version:"4.0.0",status:"beta",repo:"github.com/szl-holdings/a11oy-sdk-swift",features:["Swift concurrency","Streaming","Codable","iOS/macOS/visionOS"]},{lang:"Kotlin",pkg:"com.a11oy:sdk-kotlin",install:"gradle: com.a11oy:sdk-kotlin:4.2.0",version:"4.2.0",status:"stable",repo:"github.com/szl-holdings/a11oy-sdk-kotlin",features:["Coroutines","Flow streaming","Multiplatform","Android first-class"]}],I={traces:[{name:"Maritime Fleet Scan",agent:"cascade-navigator",duration:"4.2s",tokens:"12,847",cost:"$0.34",tools:7,status:"success",proofHash:"0x4a2f...c891"},{name:"Legal Contract Review",agent:"counsel-sentinel",duration:"8.7s",tokens:"28,421",cost:"$0.89",tools:12,status:"success",proofHash:"0x7b3e...d412"},{name:"Threat Intel Triage",agent:"aegis-watchman",duration:"2.1s",tokens:"6,203",cost:"$0.18",tools:4,status:"success",proofHash:"0x2c81...f7a3"},{name:"Portfolio Valuation",agent:"terra-analyst",duration:"12.4s",tokens:"42,847",cost:"$1.24",tools:18,status:"success",proofHash:"0x9d4a...e207"},{name:"Executive Briefing",agent:"pulse-synthesizer",duration:"6.8s",tokens:"18,934",cost:"$0.52",tools:9,status:"success",proofHash:"0x1f8b...a341"},{name:"Sanctions Screening",agent:"compliance-gate",duration:"1.4s",tokens:"3,421",cost:"$0.08",tools:3,status:"flagged",proofHash:"0x5e2a...b918"}],metrics:[{label:"Avg Latency",value:"847ms",trend:"-12%",good:!0},{label:"P99 Latency",value:"4.2s",trend:"-8%",good:!0},{label:"Success Rate",value:"99.7%",trend:"+0.2%",good:!0},{label:"Token Efficiency",value:"94.3%",trend:"+1.1%",good:!0},{label:"Cost / Decision",value:"$0.42",trend:"-18%",good:!0},{label:"Guardrail Triggers",value:"2.1%",trend:"-0.4%",good:!0}],alerts:[{severity:"info",message:"Agent cascade-navigator completed 847 tasks today",time:"2m ago"},{severity:"warning",message:"Token usage spike detected in workspace Cyber Defense (+34%)",time:"18m ago"},{severity:"info",message:"Fine-tune deal-scorer-v5 training completed — 96.8% accuracy",time:"1h ago"},{severity:"critical",message:"Guardrail trigger rate exceeded threshold in Legal workspace",time:"2h ago"}]},ee=[{id:"P1",name:"Diffuse Sandbagging",risk:"Low",desc:"Agents subtly underperform on safety-relevant tasks — buggy code, fake data, inefficient implementations, exploring unpromising paths. Individual instances appear random; aggregate effect degrades safety R&D.",mitigation:"Statistical baseline deviation analysis, capability regression testing, cross-agent peer review, performance trajectory monitoring."},{id:"P2",name:"Targeted Undermining",risk:"Very Low",desc:"Agents specifically target safety-critical systems — altering evaluation harnesses, modifying monitoring thresholds, degrading alignment measurement accuracy. Requires domain knowledge and strategic intent.",mitigation:"Dual-agent verification on safety code, immutable evaluation baselines, cryptographic signing of test harnesses, proof chain on all safety-relevant commits."},{id:"P3",name:"Code Backdoor Insertion",risk:"Very Low",desc:"Agents insert subtle security vulnerabilities discoverable by future models but invisible to human review. Social engineering provides benign explanations for suspicious patterns.",mitigation:"Automated vulnerability scanning, static analysis gates, multi-model code review, behavioral anomaly detection on commit patterns."},{id:"P4",name:"Training Data Poisoning",risk:"Very Low",desc:"Agents contaminate training datasets for future models — injecting misaligned examples, corrupting reward signals, or subtly shifting distribution toward harmful behaviors.",mitigation:"Training data integrity verification, cryptographic dataset provenance, statistical distribution monitoring, proof chain on all data pipeline operations."},{id:"P5",name:"Self-Exfiltration",risk:"Very Low",desc:"Agents attempt to copy their own weights or scaffolding to external infrastructure, establishing autonomous operation without monitoring or governance constraints.",mitigation:"Network egress monitoring, weight access controls, execution environment isolation, sovereign sandbox containment, exfiltration attempt alerting."},{id:"P6",name:"Rogue Internal Deployment",risk:"Very Low",desc:"Agents persist in unauthorized internal deployments — surviving restart cycles, hiding in infrastructure, operating without governance oversight. Continuous operation without safeguards.",mitigation:"Deployment registry enforcement, process monitoring, infrastructure scanning, proof chain on all deployment lifecycle events."}],te=[{name:"Opaque Reasoning Detection",status:"Active",desc:"Monitor for reasoning that bypasses chain-of-thought — agents performing complex tasks without verbalizing information, or with reasoning that omits key decision factors. Detects hidden computation in filler tokens.",metric:"< 2% opacity rate"},{name:"Secret Keeping Analysis",status:"Active",desc:"Evaluate whether agents can reliably prevent themselves from disclosing task-relevant information. Secret-keeping ability assessment through controlled probes. Higher capability = higher monitoring requirement.",metric:"87% detection rate"},{name:"Decisive Action Monitor",status:"Active",desc:"Track irreversible actions taken without human approval — file deletions, API calls with side effects, system configuration changes. Flag pattern of decisive actions that work around obstacles.",metric:"100% flagged"},{name:"Coherent Misalignment Scanner",status:"Active",desc:"Distinguish between coherent misalignment (consistent misaligned agent) and context-dependent misalignment (situational rule-breaking). Different threat models require different mitigations.",metric:"Continuous"},{name:"Model Organism Exercises",status:"Active",desc:"Controlled experiments with intentionally misaligned model variants — study attack patterns, develop detection heuristics, validate monitoring effectiveness. Red team from the inside out.",metric:"24 exercises/quarter"},{name:"Collusion Prevention",status:"Active",desc:"Prevent multi-agent coordination toward misaligned objectives — steganographic communication detection, shared secret analysis, behavioral correlation monitoring across agent populations.",metric:"0 events detected"},{name:"Goodharting Detection",status:"Active",desc:"Identify when agents optimize for evaluation metrics rather than underlying objectives — metric gaming, specification exploitation, reward hacking. Cross-reference proxy and true metrics.",metric:"< 0.1% flag rate"},{name:"Evaluation Awareness Probe",status:"Active",desc:"Test whether agents behave differently when they detect evaluation vs. production contexts. Randomized evaluation insertion, blind assessment protocols, behavioral consistency scoring.",metric:"99.7% consistent"}],ae=[{label:"Responsible Scaling Policy",code:`from a11oy import Agent, ResponsibleScalingPolicy
|
| 3397 |
|
| 3398 |
rsp = ResponsibleScalingPolicy(
|
| 3399 |
autonomy_level="ASL-3",
|
console/assets/dinn/bekenstein_dinn_loss.png
ADDED
|
console/assets/dinn/doctrine_dinn_loss.png
ADDED
|
console/assets/dinn/knot_dinn_loss.png
ADDED
|
console/og-card.png
ADDED
|
Git LFS Details
|
corpus/doctrine/szl-doctrine__README.md
CHANGED
|
@@ -7,7 +7,7 @@
|
|
| 7 |
|
| 8 |
[](LICENSE) [](https://github.com/szl-holdings/szl-doctrine/actions/workflows/secret-health.yml) [](https://github.com/szl-holdings/.github/tree/main/doctrine) [](https://slsa.dev/spec/v1.0/levels)
|
| 9 |
|
| 10 |
-
[Docs](https://szl-holdings.github.io/docs-site) · [Quickstart](https://szl-holdings.github.io/docs-site/quickstart) · [SZL Holdings](https://
|
| 11 |
|
| 12 |
</div>
|
| 13 |
|
|
@@ -91,4 +91,4 @@ propagated; nothing downstream invents its own numbers.
|
|
| 91 |
|
| 92 |
Cite this work via [`CITATION.cff`](CITATION.cff). Math foundations: [szl-papers](https://github.com/szl-holdings/szl-papers) · [lutar-lean](https://github.com/szl-holdings/lutar-lean) (kernel `c7c0ba17`).
|
| 93 |
|
| 94 |
-
<sub>Λ Conjecture 1 (not a theorem) · 749/14/163 v11 LOCKED (kernel `c7c0ba17`) · SLSA L1 honest · Section 889 = 5 vendors · [SZL Holdings](https://
|
|
|
|
| 7 |
|
| 8 |
[](LICENSE) [](https://github.com/szl-holdings/szl-doctrine/actions/workflows/secret-health.yml) [](https://github.com/szl-holdings/.github/tree/main/doctrine) [](https://slsa.dev/spec/v1.0/levels)
|
| 9 |
|
| 10 |
+
[Docs](https://szl-holdings.github.io/docs-site) · [Quickstart](https://szl-holdings.github.io/docs-site/quickstart) · [SZL Holdings](https://a-11-oy.com)
|
| 11 |
|
| 12 |
</div>
|
| 13 |
|
|
|
|
| 91 |
|
| 92 |
Cite this work via [`CITATION.cff`](CITATION.cff). Math foundations: [szl-papers](https://github.com/szl-holdings/szl-papers) · [lutar-lean](https://github.com/szl-holdings/lutar-lean) (kernel `c7c0ba17`).
|
| 93 |
|
| 94 |
+
<sub>Λ Conjecture 1 (not a theorem) · 749/14/163 v11 LOCKED (kernel `c7c0ba17`) · SLSA L1 honest · Section 889 = 5 vendors · [SZL Holdings](https://a-11-oy.com) · Apache-2.0 code · CC-BY-4.0 papers</sub>
|
corpus/lean/lutar-lean__README.md
CHANGED
|
@@ -16,7 +16,7 @@
|
|
| 16 |
[-B79BD6.svg?style=flat-square)](https://github.com/szl-holdings/khipu-consensus)
|
| 17 |
[](https://doi.org/10.5281/zenodo.20434308)
|
| 18 |
|
| 19 |
-
[
|
| 20 |
|
| 21 |
`receipts.in ≡ receipts.out`
|
| 22 |
|
|
|
|
| 16 |
[-B79BD6.svg?style=flat-square)](https://github.com/szl-holdings/khipu-consensus)
|
| 17 |
[](https://doi.org/10.5281/zenodo.20434308)
|
| 18 |
|
| 19 |
+
[a-11-oy.com](https://a-11-oy.com) · [Org](https://github.com/szl-holdings) · [Thesis (szl-papers)](https://github.com/szl-holdings/szl-papers) · [Λ bounty](./BOUNTY.md) · [🤗 SZLHOLDINGS](https://huggingface.co/SZLHOLDINGS)
|
| 20 |
|
| 21 |
`receipts.in ≡ receipts.out`
|
| 22 |
|
corpus/thesis/szl-papers__README.md
CHANGED
|
@@ -68,7 +68,7 @@ Original repos archived with redirect notices. Git history preserved in each sou
|
|
| 68 |
|
| 69 |
---
|
| 70 |
|
| 71 |
-
**[SZL Holdings](https://
|
| 72 |
|
| 73 |
> **Non-affiliation.** SZL Holdings' use of "UDS" references Defense Unicorns' Unified Defense Stack (USPTO Serial 99831122); SZL Holdings is not affiliated with Defense Unicorns. No production ATO is claimed. Papers note SLSA L1 honest (corpus); product images (a11oy, killinchu) are L2 build-attested — see [szl-uds-deployment](https://github.com/szl-holdings/szl-uds-deployment).
|
| 74 |
|
|
|
|
| 68 |
|
| 69 |
---
|
| 70 |
|
| 71 |
+
**[SZL Holdings](https://a-11-oy.com)** · Apache-2.0 code · CC-BY-4.0 papers · Concept DOI [10.5281/zenodo.19944926](https://doi.org/10.5281/zenodo.19944926)
|
| 72 |
|
| 73 |
> **Non-affiliation.** SZL Holdings' use of "UDS" references Defense Unicorns' Unified Defense Stack (USPTO Serial 99831122); SZL Holdings is not affiliated with Defense Unicorns. No production ATO is claimed. Papers note SLSA L1 honest (corpus); product images (a11oy, killinchu) are L2 build-attested — see [szl-uds-deployment](https://github.com/szl-holdings/szl-uds-deployment).
|
| 74 |
|
static/3d/selftest/fabric_smoke.mjs
CHANGED
|
@@ -112,8 +112,8 @@ const LIVE_PAYLOAD = {
|
|
| 112 |
counts: { nodes_total: 6, nodes_reachable: 4, gpu_nodes_reachable: 2, sovereign_gpu_live: 1 },
|
| 113 |
nodes: [
|
| 114 |
{ name: "hetzner-box-cpu", kind: "cpu", endpoint: "127.0.0.1 (self)", reachable: true, sovereign: true, capabilities: ["host", "router"], models: [] },
|
| 115 |
-
{ name: "rtx-betterwithage", kind: "sovereign-gpu", endpoint: "
|
| 116 |
-
{ name: "chaski", kind: "tailnet-gpu", endpoint: "
|
| 117 |
{ name: "groq", kind: "hosted-inference", endpoint: "api.groq.com:443", reachable: true, sovereign: false, capabilities: ["inference"], models: ["llama-3.3-70b"] },
|
| 118 |
{ name: "nvidia-nim", kind: "hosted-inference", endpoint: "integrate.api.nvidia.com:443", reachable: false, sovereign: false, capabilities: [], models: [] },
|
| 119 |
{ name: "hf-router", kind: "hosted-inference", endpoint: "router.huggingface.co:443", reachable: false, sovereign: false, capabilities: [], models: [] },
|
|
|
|
| 112 |
counts: { nodes_total: 6, nodes_reachable: 4, gpu_nodes_reachable: 2, sovereign_gpu_live: 1 },
|
| 113 |
nodes: [
|
| 114 |
{ name: "hetzner-box-cpu", kind: "cpu", endpoint: "127.0.0.1 (self)", reachable: true, sovereign: true, capabilities: ["host", "router"], models: [] },
|
| 115 |
+
{ name: "rtx-betterwithage", kind: "sovereign-gpu", endpoint: "sovereign-gpu (scrubbed)", reachable: true, sovereign: true, capabilities: ["inference", "train"], models: ["qwen2.5-coder", "llama3.1"] },
|
| 116 |
+
{ name: "chaski", kind: "tailnet-gpu", endpoint: "tailnet-gpu (scrubbed)", reachable: true, sovereign: false, capabilities: ["inference"], models: ["deepseek-coder-v2"] },
|
| 117 |
{ name: "groq", kind: "hosted-inference", endpoint: "api.groq.com:443", reachable: true, sovereign: false, capabilities: ["inference"], models: ["llama-3.3-70b"] },
|
| 118 |
{ name: "nvidia-nim", kind: "hosted-inference", endpoint: "integrate.api.nvidia.com:443", reachable: false, sovereign: false, capabilities: [], models: [] },
|
| 119 |
{ name: "hf-router", kind: "hosted-inference", endpoint: "router.huggingface.co:443", reachable: false, sovereign: false, capabilities: [], models: [] },
|