Infrastructure and open research for verifiable intelligent systems.

Missions · Authority · Durable Execution · Evidence · Verification · Agentic Institutions

Aftergraph is an engineering and research organization working on the systems layer around autonomous AI: how long-running agents can execute real work while preserving bounded authority, durable state, measurable cost, machine-verifiable evidence, and independently verified outcomes.

Intelligence that can act. Systems that can prove why.

📡 Public Knowledge Plane live: docs.aftergraph.org — architecture, contracts, claims with evidence strength, API reference, and agent-readable context packs, all provenance-stamped to exact source commits.

💬 Public deliberation: Aftergraph Discussions — research reproduction, prior art, RFC formation, roadmap input and architecture critique.

Direct public entry points: Discussions · Public roadmap · RFC process · Research papers · AIE specification · Contributing · Security

Most agent stacks are good at producing actions. Production systems need more than action generation:

human intent
    ↓
mission contract
    ↓
state + capabilities + authority + budget
    ↓
execution + recovery
    ↓
evidence
    ↓
independent verification
    ↓
verified outcome

The core design rule is simple: an agent saying “done” is not the same thing as the system proving the outcome.

The research program investigates a vendor-neutral systems contract connecting human intent to persistent, governed and verifiable outcomes across heterogeneous models, agents, tools and runtimes.

Current public work includes:

• SPEC-001 / Mission Contract — machine-readable mission, state, capability, authority, budget, trajectory, evidence and verifier semantics.
• MISSION-Bench — fault-injected evaluation of long-horizon intelligent systems.
• Evidence-gated completion — Complete ≠ Verified as a normative system invariant.
• Attenuated authority — scoped, expiring and purpose-bound delegation rather than ambient credentials.
• Cost Per Verified Outcome (CPVO) — evaluation of useful verified work instead of raw inference price alone.
• Cross-runtime conformance — portable semantics rather than framework-specific governance glue.

Scientific claims remain bounded by the evidence class stated in the research repository. Deterministic sandbox results, live-provider results, simulations and external reproduction are not treated as interchangeable evidence.

AIE investigates the engineering layer above graph coordination and runtime control:

Graph defines coordination. Control enforces execution. Institution resolves legitimate authority.

AIE focuses on portable semantics for principals, roles, authority, delegation, mission contracts, budgets, revocation, lifecycle, governed topology mutation and evidence. It is currently a Research Draft / experimental standards track, not a declared industry standard.

graphs → boundaries → authority → execution → evidence → verified outcomes

Aftergraph intentionally composes existing standards and protocols where they already solve a problem. The goal is not to invent another transport, identity system or telemetry format merely because the industry was apparently short on acronyms.

Private, temporary, legacy-transition and incubation repositories may participate in the topology without being public entry points. Repository existence never upgrades product maturity, conformance or evidence strength.

We value criticism that can falsify a claim, not merely decorate it with another opinion.

• Reproduce a benchmark or conformance result.
• Implement the specification independently.
• Test interoperability with another runtime.
• Submit contradictory prior art or a failing invariant.
• Review the threat model, evidence model or economics.
• Build against the public contracts and report semantic deviations.
• Start an [RFC] discussion before a cross-repository or normative change becomes implementation work.

Useful entry points: Discussions · MISSION-Bench registry · Prior-art challenge · Architecture RFC · Sentinel verdict · Roadmap thread · RFC intake

Aftergraph uses gates rather than ceremonial dates. Current public priorities are independent reproduction, contract/conformance hardening, AIE candidate-readiness work, enforceable runtime boundaries, Sentinel exact-HEAD verification, and a usable public community/knowledge plane.

Read the live coordination document: PUBLIC-ROADMAP.md.

⭐ Star the repositories you want to track.
👀 Watch research and standards repositories for releases and evidence updates.
🧪 Reproduce results instead of trusting screenshots, because civilization has suffered enough screenshots presented as benchmarks.
💬 Challenge claims and architecture in Discussions before they calcify into folklore.
🔗 Reference Aftergraph when using its specifications, benchmarks or open-source implementations so downstream work remains traceable.

• Organization Discussions
• MISSION-Bench registry
• Prior-art challenge
• Architecture RFC
• Sentinel verdict discussion
• Public roadmap discussion
• RFC intake discussion
• Discussion routing
• Public roadmap
• RFC process
• Contributing
• Security policy
• Support routing
• Code of Conduct

Runtime evidence, execution evidence, conformance evidence and scientific evidence are separate. No repository automatically inherits another repository's claims.

Canonical architecture, role allocation, contracts, naming, exact-head organization state and claim boundaries live in Aftergraph/after-graph-governance.

Open research · Responsible deployment · Verifiable progress

_{Brand status: masterbrand Aftergraph under Brand OS v1.1.0 (Aftergraph/brand release v1.1.0). Legacy “ABDE Intelligence” parent copy retired; Sentinel product identity pending naming review (Aftergraph/brand#19).}