Files
correx/docs/epics/epic-4.md
T

7.4 KiB

Epic 4: Validation Pipeline (Deterministic Analysis + Cross-Layer Consistency Engine)

status: completed date: 09.05.2026 (May) scope: :core:validation (depends on :core:events, :core:sessions, :core:transitions, Epic 3 analysis outputs)


context

With deterministic event sourcing (Epic 1), session projection (Epic 2), and workflow transition semantics (Epic 3) in place, Correx now has a complete execution and reconstruction model.

However, there is no formal layer that ensures:

  • graph integrity
  • transition consistency
  • session projection correctness
  • configuration validity (cycle policies, semantic rules)
  • cross-layer coherence under replay

Without a validation layer:

  • invalid workflow graphs can still be constructed
  • inconsistent transitions can pass unnoticed
  • session projections can diverge semantically from graph structure
  • cycle governance (policy binding) is not enforceably checked
  • system correctness is only implicitly trusted, not verified

Epic 4 introduces a deterministic validation system to close this gap.


goal

Introduce a deterministic, replay-safe validation pipeline:

a pure analysis engine that validates workflow structure, execution projections, and configuration consistency without influencing execution behavior

The system must remain:

  • deterministic
  • stateless
  • replay-safe
  • non-executing
  • side-effect free

It is NOT an orchestration engine and does NOT participate in runtime control flow.


scope (what IS included)


1. validation model layer

Define unified validation representation:

  • ValidationReport
  • ValidationSection
  • ValidationIssue
  • ValidationSeverity
  • ValidationContext

Validation context is a full system snapshot:

  • WorkflowGraph
  • DetectedCycle (Epic 3 output)
  • CyclePolicyBinding
  • SessionState

Properties:

  • immutable
  • replay-safe
  • fully deterministic input contract

2. graph validation

Validate structural correctness of workflow topology.

Checks:

  • start node existence
  • dangling transitions (invalid stage references)
  • structural consistency of graph definition
  • cycle presence reporting (pass-through only)

Important:

  • cycles are allowed by design (Epic 3)
  • cycle detection is informational only
  • no enforcement or interpretation of cycles occurs here

3. transition validation

Validate deterministic consistency of transition engine inputs.

Checks:

  • transition endpoints exist in graph
  • condition presence and structural validity
  • deterministic ordering correctness per source node
  • ambiguity detection in transition resolution ordering

Important:

  • transition conditions are NOT executed
  • no resolution logic is invoked
  • no runtime decisioning occurs

4. session validation

Validate projection consistency against event-derived state.

Checks:

  • temporal consistency (createdAt ≤ updatedAt)
  • session state sanity (invalid transition counts)
  • consistency between session state and graph structure
  • lightweight anomaly detection over projection output

Important:

  • no event replay is performed
  • no projection recomputation occurs
  • session state is treated as immutable derived artifact

5. semantic validation layer

Validate configuration and cross-domain consistency rules.

Includes:

  • cycle-policy binding completeness (mode-dependent enforcement)
  • configuration consistency checks
  • cross-layer structural coherence rules

Cycle semantics:

  • CycleSignature derived from normalized node sets
  • edges are not part of identity
  • policy binding is evaluated against cycle signatures only

Important:

  • cycles remain structurally valid (Epic 3 rule preserved)
  • semantic layer does NOT enforce execution behavior
  • policies are treated as configuration constraints, not runtime logic

6. validation pipeline executor

Define deterministic orchestration of validation layers.

Pipeline behavior:

  • strictly ordered execution
  • no hidden parallelism
  • fully deterministic evaluation order
  • stateless execution model

Execution flow:

Graph → Transition → Session → Semantic → ValidationReport

Important:

  • pipeline aggregates results only
  • no control flow decisions beyond aggregation
  • no mutation of input context

7. approval trigger integration point

Introduce boundary layer between validation and external decision systems.

Responsibilities:

  • evaluate ValidationReport
  • compute risk summary
  • decide whether approval request is required
  • emit ApprovalRequest artifact

Important constraints:

  • no execution coupling
  • no workflow control logic
  • no system orchestration responsibility
  • strictly output transformation layer

Output artifacts:

  • ApprovalRequest
  • RiskSummary

Behavior:

  • errors or missing cycle-policy bindings trigger approval requirement
  • acts as deterministic gating boundary, not execution controller

8. deterministic testing requirements

Validation system must be fully testable under replay constraints.

Coverage includes:

  • graph validation determinism
  • transition validation consistency
  • session projection sanity validation
  • semantic rule enforcement consistency
  • pipeline ordering determinism
  • approval trigger stability
  • full system replay equivalence

Invariant:

identical input context must always produce identical ValidationReport and ApprovalRequest outcome


explicit exclusions

Epic 4 does NOT include:

  • workflow execution
  • transition evaluation or resolution
  • scheduling or orchestration
  • runtime policy enforcement
  • execution retry logic
  • system control flow decisions
  • async or distributed validation execution
  • event mutation or emission

Those belong to future runtime/orchestration layers.


consequences

positive

  • introduces explicit correctness boundary for entire system
  • makes workflow + session consistency verifiable
  • decouples analysis from execution
  • enables safe pre-execution validation gating
  • formalizes configuration governance (cycle policies)
  • improves replay debugging and system observability

negative

  • adds another full abstraction layer over already complex system
  • increases conceptual separation between “valid structure” and “valid execution”
  • introduces potential duplication of logic if misused outside boundary
  • requires strict discipline to avoid leaking execution semantics into validation
  • makes system reasoning more layered (analysis vs execution vs policy)

rationale

Epic 4 exists because deterministic execution alone is insufficient.

Even with:

  • event sourcing (Epic 1)
  • projection layer (Epic 2)
  • workflow engine (Epic 3)

the system still lacks:

a formal correctness boundary over structure, configuration, and derived state

This epic ensures:

  • workflows are structurally valid before execution
  • session state remains consistent with graph semantics
  • configuration constraints (cycle policies) are explicitly governed
  • system behavior remains fully replay-verifiable

It completes the shift from:

  • “deterministic execution system” to
  • “deterministic + verifiable workflow system”

status

Epic 4 is considered complete once:

  • validation model is defined ✔
  • graph/transition/session/semantic validators exist ✔
  • pipeline executor is deterministic ✔
  • approval trigger boundary is implemented ✔
  • replay tests confirm stability ✔

It is the final analysis layer before runtime orchestration and execution-control systems are introduced in future epics.