◈
01Simulation Before Write
No agentic write operation executes without a prior simulation pass. Imagining consequences is not optional — it is the prerequisite to acting.
Phase 24v0.24.0-rc1Release Candidate
Closure Hardening + Release-Candidate Conditioning
CodeWorldObservatory
A simulation-first control plane for agentic software engineering. Software agents should imagine code futures before acting.
Treating software repositories as dynamic causal worlds — not static text — enables counterfactual planning, branch evaluation, and prediction-before-write workflows that autoregressive approaches cannot achieve.
Observatory Thesis
Software is a world. Agents must learn to navigate it.
Autoregressive write-first is insufficient
Systems that generate code by predicting the next token cannot reason about causal consequences. They write first and discover problems later.
Repositories are observable worlds
A codebase is a deterministic, stateful world with defined laws (types, tests, contracts). It can be modeled, simulated, and reasoned over before any action.
Agentic IDEs are the substrate
Modern agentic environments provide state capture, tool access, artifact production, and execution verification - perfect conditions for a simulation control plane.
MCP bridges prediction to action
Model Context Protocol will serve as the bridge between the observatory's predictive intelligence and the tools that execute real interventions.
Live Observatory
Research control surface
Real-time visibility into world state, planned interventions, comparative timelines, experiment-centered lineage, replay-aware research consumers, evaluation persistence, advisory-only research prioritization, governed priority drift traceability, historically durable priority history, automated snapshot comparison surfaces, unified comparative governance synthesis, and closure-hardened release-candidate conditioning. Phase 24 active.
Current World State
Capturing repo snapshot…
WS
Pending
Scanning workspace…
Reading file metadata…
Analyzing imports…
Candidate Interventions
Generating branch plans…
INT
Pending
Reading repo signals…
Analyzing scope…
Building candidate branches…
Counterfactual Futures
Generating consequence projections…
CF
Pending
Predicting branch outcomes…
Uncertainty Surface
Checking structural limits…
USF
Pending
Analyzing prediction confidence…
Prediction vs Reality
Retrieving execution records…
PVR
Pending
Loading calibration data...
Artifact Ledger
Retrieving ledger entries…
ART
Pending
Reading immutable records...
Benchmark Harness [Sim-Sessions]
Initializing simulation session...
Phase 24 ActiveThe Observatory has completed closure hardening and release-candidate conditioning: canonical governance vocabulary unified across all services, inline styles eliminated from all panels in favor of Tailwind semantic tokens, institutional color scheme unified from fragmented info/warning to consistent accent, and shared UI primitives established for cross-panel visual coherence.
Operating Principles
Four laws of the Observatory
These are not guidelines. They are invariants. Any system claiming to implement CodeWorld Observatory must satisfy all four.
⟂
02Branch Before Intervention
Every proposed change is evaluated as a counterfactual branch in the world model. The main timeline is never the first casualty of exploration.
≈
03Visible Uncertainty
Uncertainty is first-class data. Where the model cannot predict with confidence, the system makes that ambiguity explicit and visible to the operator.
⬡
04Auditable Artifacts
Every plan, simulation, and outcome is recorded as an immutable, hashed artifact. The system can always trace the chain from intention to consequence.
Roadmap
24-phase build plan
Each phase is gated on the completion and verification of the preceding phase. No phase skipping.
✓
Foundation Scaffold
✓
State Capture Engine
✓
Intervention Planner
✓
Simulation Engine
✓
Evaluation Framework
✓
MCP Simulation Bridge
✓
SE-JEPA Prototype Layer
✓
Benchmark Harness
✓
Experiment Memory
✓
Latent State Approximation
✓
Research Export / Briefing Surface
✓
Comparative Research Timeline + Narrative Playback
✓
Experiment Registry + Historical State Capture
✓
Scenario Library + Dataset Infrastructure
✓
Reproducibility + Statistical Evaluation
✓
Experiment-to-Simulation / Benchmark Lineage Binding
✓
Replay-Aware Research Consumers + Experiment Detail Surface
✓
Research Evaluation Semantics + Evidence-Weighted Comparative Analysis
✓
Evaluation Persistence + Comparative Research Surfaces
✓
Research Prioritization Engine
✓
Priority Drift + Recommendation Governance
✓
Historical Priority Ledger + Comparative Governance History
✓
Priority Snapshot Automation + Historical Comparison Surfaces
✓
Comparative Governance Synthesis Layer
24