correx/prompts/bootstrap.md

Correx Context Bootstrap

---

1. module architecture

❯ tree -L 2
.
├── apps
│   ├── cli
│   ├── desktop
│   ├── server
│   └── worker
├── build
│   └── reports
├── build.gradle
├── core
│   ├── agents
│   ├── approvals
│   ├── artifacts
│   ├── config
│   ├── context
│   ├── events
│   ├── inference
│   ├── kernel
│   ├── observability
│   ├── policies
│   ├── router
│   ├── sessions
│   ├── stages
│   ├── tools
│   ├── transitions
│   └── validation
├── detekt.yml
├── docs
│   ├── architecture
│   ├── configs
│   ├── decisions
│   ├── des-doc-v0.1.md
│   ├── epics.md
│   ├── events
│   ├── example.yaml
│   ├── lang-framewok-missing-pieces.md
│   ├── modules
│   ├── plugins
│   ├── spec-v0.1.md
│   ├── structure.md
│   ├── threat_model
│   └── transitions
├── epics
│   ├── epic-1.5.md
│   ├── epic-1.md
│   ├── epic-2.md
│   └── epic-3.md
├── examples
│   ├── configs
│   ├── plugins
│   ├── stages
│   └── workflows
├── frontend
│   ├── src
│   └── static
├── gradle
│   └── wrapper
├── gradle.properties
├── gradlew
├── gradlew.bat
├── infrastructure
│   ├── inference
│   ├── persistence
│   ├── scheduler
│   ├── security
│   ├── telemetry
│   └── tools
├── interfaces
│   ├── api
│   ├── cli
│   └── sdk
├── plugins
│   ├── compressors
│   ├── policies
│   ├── providers
│   ├── stages
│   ├── tools
│   ├── transitions
│   └── validators
├── settings.gradle
└── testing
    ├── approvals
    ├── contracts
    ├── deterministic
    ├── fixtures
    ├── integration
    ├── projections
    ├── replay
    └── transitions

---

dependency rules (important)

• core modules → no infrastructure dependencies

• infrastructure → implements core interfaces

• testing/contracts → defines reusable invariants

• core/events = event system foundation

• core/sessions = deterministic event-sourced lifecycle projection layer

• core/transitions = deterministic workflow graph evaluation engine

• projections are domain-specific folds over event streams

• replay engine stays generic and domain-agnostic

• transition engine is pure evaluation logic (no orchestration)

---

2. current epic

Epic 4: Validation pipeline

goal: deterministic multi-layer validation system for workflows, sessions, and transitions

deliverables:

- routing validation layer (graph + transition correctness)
- schema validation layer (event + config integrity)
- semantic validation hooks (domain rules without execution coupling)
- validation pipeline executor (ordered deterministic execution)
- approval trigger integration point (event-based, no direct execution coupling)

dependencies:

- Epic 1 (event system + replay foundation)
- Epic 2 (session projection layer)
- Epic 3 (workflow transition engine)

---

3. implementation state

READY TO START

Completed before Epic 4:

- append-only EventStore with strict ordering guarantees
- polymorphic EventPayload serialization (kotlinx.serialization)
- deterministic replay infrastructure (EventReplayer + Projection fold model)
- session lifecycle projection (SessionProjector + SessionReducer)
- FSM abstraction replaced by deterministic reducer semantics
- workflow graph model (WorkflowGraph, TransitionEdge, StageId)
- deterministic transition evaluation engine (resolver + condition evaluation)
- cycle detection + canonicalization (DFS-based graph analysis)
- stage execution contract (execution isolated from orchestration)
- workflow execution events integrated into event stream
- session ↔ transition integration via event-driven projection model
- full replay determinism across session + workflow execution

---

4. key design decisions (do not violate)

- EventPayload uses kotlinx.serialization polymorphic serialization
- EventStore is append-only per session stream
- sequence is strictly monotonic per session
- projections are pure functions: (state, event) -> new state
- replay must be deterministic across implementations
- core/events contains replay infrastructure only

- core/sessions contains event-driven lifecycle projection (no FSM coupling)
- SessionProjector is a deterministic reducer over StoredEvent stream
- SessionReducer replaces FSM semantics (no imperative state transitions)

- core/transitions is a pure workflow graph evaluation engine
- transitions do NOT execute runtime logic
- transitions do NOT schedule or orchestrate execution
- transitions only emit execution-relevant events

- workflow graph is immutable and unordered (Set-based topology)
- determinism is enforced by resolver, not input structure

- cycle detection is analysis-only in Epic 3 (no runtime policy enforcement yet)

- Event stream is the single source of truth for all system state

- projections MUST NOT perform IO, side effects, or runtime clock access
- createdAt/updatedAt derived strictly from EventMetadata timestamps

- same event stream MUST always reconstruct identical system state

---

5. current invariants

- no mutation outside store/projection boundaries
- ordering is enforced at EventStore level
- idempotency enforced by eventId
- replay must be deterministic for identical event streams

- projections are stateless deterministic reducers
- projections MUST NOT depend on external mutable state
- projections MUST NOT perform side effects

- event sequence is authoritative ordering source
- event stream is the only system truth source

- session state is fully derived from replay
- session state is never persisted directly

- workflow execution is event-driven and replay-safe
- transition evaluation must be deterministic and context-bound

- workflow topology is independent from execution semantics
- graph structure does not define execution order

- cycles are structurally allowed but not yet policy-governed

---

6. current task

Epic 4 implementation:

Design and implement deterministic validation pipeline over workflow + session + transition layers.

Focus areas:

1. validate workflow graph correctness (structural + cycle awareness hooks)
2. validate transition definitions (conditions + reachability consistency)
3. validate session-event consistency against workflow semantics
4. introduce ordered validation pipeline executor
5. define extension points for semantic validation rules
6. ensure validation is replay-safe and deterministic

Important constraint:
validation MUST NOT mutate state or trigger execution
it only produces deterministic validation results over existing models

---

7. relevant code

no new core code required at bootstrap level beyond existing:

data class WorkflowGraph(
    val stages: Set<StageId>,
    val transitions: Set<TransitionEdge>,
    val start: StageId
) {
    init {
        require(start in stages) {
            "start stage must exist in stages"
        }
    }
}

class DefaultTransitionResolver(
    private val evaluator: TransitionConditionEvaluator
) : TransitionResolver {

    override fun resolve(
        graph: WorkflowGraph,
        context: EvaluationContext
    ): TransitionDecision {
        val outgoing = graph.transitions
            .asSequence()
            .filter { it.from == context.currentStage }
            .sortedWith(TransitionOrdering.comparator)
            .toList()

        for (edge in outgoing) {
            val result = evaluator.evaluate(edge.condition, context)
            if (result) {
                return TransitionDecision.Move(
                    transitionId = edge.id,
                    to = edge.to
                )
            }
        }
        return TransitionDecision.Stay
    }
}

class SessionProjector(
    private val reducer: SessionReducer
) : Projection<SessionState> {

    override fun initial(): SessionState =
        SessionState(
            status = SessionStatus.CREATED
        )

    override fun apply(
        state: SessionState,
        event: StoredEvent
    ): SessionState = reducer.reduce(state, event)
}

class DefaultSessionReducer : SessionReducer {

    override fun reduce(
        state: SessionState,
        event: StoredEvent
    ): SessionState {

        val payload = event.payload

        val newStatus = when (payload) {

            is SessionStartedEvent ->
                SessionStatus.ACTIVE

            is SessionPausedEvent ->
                SessionStatus.PAUSED

            is SessionResumedEvent ->
                SessionStatus.ACTIVE

            is SessionCompletedEvent ->
                SessionStatus.COMPLETED

            is SessionFailedEvent,
            is StageFailedEvent ->
                SessionStatus.FAILED

            is StageStartedEvent,
            is StageCompletedEvent,
            is TransitionExecutedEvent ->
                SessionStatus.ACTIVE

            else ->
                state.status
        }

        val createdAt = state.createdAt
            ?: event.metadata.timestamp

        return state.copy(
            status = newStatus,
            createdAt = createdAt,
            updatedAt = event.metadata.timestamp
        )
    }
}

class DefaultEventReplayer<S>(
    private val store: EventStore,
    private val projection: Projection<S>
) : EventReplayer<S> {

    override fun rebuild(sessionId: SessionId): S {
        val events = store.read(sessionId)

        return DefaultStateBuilder(projection)
            .build(events)
    }
}

internal class CycleExtractor(
    private val adjacencyBuilder: DeterministicAdjacencyBuilder,
    private val dfs: CycleDfs,
) {

    fun extract(graph: WorkflowGraph): List<DetectedCycle> {

        val adjacency = adjacencyBuilder.build(graph)

        val state = mutableMapOf<StageId, VisitState>()
        val path = mutableListOf<StageId>()

        val cycles = mutableListOf<DetectedCycle>()

        val sortedNodes = graph.stages
            .sortedBy { it.value }

        for (node in sortedNodes) {
            if (state[node] == null) {
                dfs.dfs(node, adjacency, state, path, cycles)
            }
        }

        return canonicalize(cycles)
    }
}

---