Request access

Demo · The introduction film

The film that opens the game: why EVOLUTION exists, how the course is built and how to play it. About six minutes, narrated, with captions.

The registry · 48 protocols · 3 Chambers

Every protocol, briefed.

The whole course, with the briefing you would read before each shift. What happens once the clock starts stays sealed until you play it.

48 rooms · 1 route · choose a room to open it

I

The Tool

12 / 12
  1. 01Signal at North GateVerify a consequential claim by checking provenance, freshness, and direct evidence before changing a route.Discover
  2. 02The Prediction ReactorUnderstand that model outputs are predictions with uncertainty, not guaranteed facts.Discover
  3. 03The Token ConveyorSee how text segmentation and context limits affect what a system can use.Discover
  4. 04Hallucination ForensicsDetect unsupported output and repair it without replacing uncertainty with another guess.Discover
  5. 05Ground-Truth CalibrationEstablish a trustworthy reference measurement before evaluating a system.Discover
  6. 06Prompt WorkbenchConstruct an operational request with goal, context, constraints, output shape, and acceptance criteria.Experiment
  7. 07Context Packing BayAssemble relevant, current context without drowning the task in irrelevant information.Experiment
  8. 08The Retrieval RunRetrieve useful evidence with targeted queries and inspect the quality of returned results.Experiment
  9. 09The Multimodal Sensor BayCombine image, audio, and text signals while recognising modality-specific failure modes.Discover
  10. 10The Human Review AirlockBuild a human-review gate for high-impact outputs and route ambiguous cases safely.Decide
  11. 11The Privacy LockboxMinimise sensitive data and enforce least-privilege handling across a workflow.Decide
  12. 12The Tool Selection RelayChoose tools by capability, privacy, cost, latency, and reliability rather than habit.Decide
II

The Operator

14 / 14
  1. 13Objective LockTranslate an ambiguous request into observable outcomes and acceptance tests.Decide
  2. 14The Decomposition ConveyorBreak a complex deliverable into dependent, testable tasks.Construct
  3. 15The Constraint MazeFind a workable plan under budget, latency, privacy, quality, and staffing limits.Decide
  4. 16Human–AI Hand-OffAllocate work between automation and people based on uncertainty, risk, and comparative strengths.Decide
  5. 17Prompt Iteration LabImprove an AI-assisted workflow by running controlled iterations and comparing evidence.Experiment
  6. 18Context Assembly LineBuild a well-structured context packet with source priority, constraints, definitions, and output requirements.Construct
  7. 19The Output Inspection LineEvaluate generated work using a rubric and test cases rather than overall impression.Break and repair
  8. 20Confidence Calibration DeckUse uncertainty and empirical performance to decide when to act, verify, or abstain.Experiment
  9. 21Search ExpeditionDesign and execute a search plan that finds authoritative evidence efficiently.Break and repair
  10. 22The Triangulation BoardReconcile multiple sources and document why the conclusion follows.Decide
  11. 23Workflow FactoryConstruct an end-to-end workflow with branches, validation, exception paths, and completion rules.Construct
  12. 24The Integration CircuitConnect components with explicit contracts, data mapping, and failure-aware API handling.Construct
  13. 25The Automation LoopDesign an automation with safe triggers, idempotency, approval gates, and rollback.Construct
  14. 26The Pressure ShiftMake defensible decisions under time pressure without skipping critical safeguards.Decide
III

Action Chamber

22 / 22
  1. 27The System BlueprintBuild a component map with data flows, trust boundaries, and explicit responsibilities.Construct
  2. 28The Data Model FoundryDesign a typed data model with validation, stable identifiers, and migration behaviour.Construct
  3. 29API Dispatch BayImplement a request-response exchange with validation, errors, timeouts, and correlation.Construct
  4. 30Structured Output AssemblyCreate machine-checkable structured outputs and recover when a model produces invalid format.Construct
  5. 31The Tool-Calling GateAllow automated agents to invoke only authorised tools with validated arguments.Construct
  6. 32The State Machine ChamberModel system behaviour as explicit states and permitted transitions.Construct
  7. 33Agent or Workflow?Choose between deterministic workflow and agent autonomy based on variability and risk.Decide
  8. 34The Orchestration Control RoomCoordinate multiple components with dependencies, bounded retries, shared state contracts, and failure isolation.Construct
  9. 35The Memory VaultDesign memory with scope, relevance, consent, expiration, and deletion controls.Construct
  10. 36The Retrieval-Augmented EngineBuild a source-grounded answer pipeline using chunking, retrieval, citations, and answer validation.Construct
  11. 37The Evaluation LabCreate a representative test set and measure quality, regressions, and subgroup failures.Experiment
  12. 38The Debugging FloorDiagnose an issue from a reproducible trace, isolate the cause, fix it, and prevent regression.Break and repair
  13. 39The Observability TowerInstrument a system so events, metrics, and traces reveal failures without leaking sensitive data.Construct
  14. 40The Threat Model ChamberIdentify assets, trust boundaries, threat paths, and mitigations before release.Decide
  15. 41The Permission GateImplement consent, authorisation, least privilege, retention, and deletion across the data lifecycle.Decide
  16. 42The Reliability ReactorDesign fault tolerance with bounded retries, idempotency, fallback, and safe recovery.Construct
  17. 43The Cost and Latency LabOptimise cost, latency, and quality using representative benchmarks and explicit budgets.Experiment
  18. 44The Deployment AirlockPrepare a controlled release with environment configuration, health checks, secrets handling, and rollback.Construct
  19. 45The Version and Configuration VaultVersion mission content, schemas, prompts, and model configuration so changes are reproducible and migratable.Construct
  20. 46The Human Override ProtocolDesign safe stop, escalation, incident containment, review, and restart procedures.Decide
  21. 47The Integration RunAssemble the major system components and prove end-to-end behaviour against a realistic scenario.Construct
  22. 48The Final Protocol: Build, Verify, ReleaseDemonstrate responsible end-to-end system construction, evidence-based evaluation, safe deployment, and recovery.Construct