core-labs/core
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions
core/docs/decisions/ADR-0123-symbolic-logic-shape-remap.md

4.1 KiB
Raw Permalink Blame History

ADR-0123 — symbolic_logic Lane-Shape Remap (ADR-0109 Amendment)

Status: Accepted Date: 2026-05-22 Author: CORE agents + reviewers Amends: ADR-0109 Depends on: ADR-0109, ADR-0122

Context

Under ADR-0109, threshold verification rules are lane-shape-aware, mapping specific lane IDs to defined metric shapes. ADR-0109 mapped the symbolic_logic lane to symbolic_logic_shape, which internally dispatched to the accuracy-shape threshold checker (_check_accuracy_shape).

However, as documented in the deferred promotion of systems_software (ADR-0122), the symbolic_logic lane output payload actually contains inference-style metrics (all_pass_rate, replay_determinism, and overall_pass) rather than accuracy-shape metrics (accuracy or passed/total). This mismatch caused the gate to fail closed with the error message: lane 'symbolic_logic' missing accuracy (and no passed/total fallback).

This amendment corrects the mapping to unblock the systems_software promotion.

Decision

Re-map the symbolic_logic lane to inference_shape within the LANE_SHAPE_REGISTRY in core/capability/expert_demo.py:

LANE_SHAPE_REGISTRY["symbolic_logic"] = "inference_shape"

Rationale

The symbolic_logic lane validates propositional inference closure. Because it checks deterministic symbolic inferences rather than approximate/statistical accuracy scalars, it naturally produces the standard inference-style metrics:

  • all_pass_rate
  • replay_determinism
  • overall_pass

Mapping it to inference_shape aligns the contract verification logic with the lane's actual runtime evidence schema.

Additionally, because no other lane in the registry maps to symbolic_logic_shape, this shape checker is retired and removed from SHAPE_CHECKERS to preserve code cleanliness and avoid dead-shape drift.

Invariants

inference_shape_thresholds_applied

The existing inference_shape thresholds defined in ADR-0109 now govern symbolic_logic verification:

  • all_pass_rate >= 0.95
  • replay_determinism == 1.0
  • overall_pass == True

Acceptance Evidence

Accepted when:

  • The registry mapping in core/capability/expert_demo.py is amended to map symbolic_logic to inference_shape.
  • Unused symbolic_logic_shape is removed from SHAPE_CHECKERS.
  • A unit test TestSymbolicLogicShapeGate in tests/test_lane_shape_thresholds.py asserts that:
    • resolve_lane_shape("symbolic_logic") == "inference_shape"
    • symbolic_logic results carrying compliant inference-style metrics pass the gate successfully.

Consequences

  • The contract blocker for systems_software is resolved: the lane gate now evaluates symbolic_logic outputs under the correct metric schema.
  • The deferral in ADR-0122 can be resolved in a subsequent promotion attempt (likely ADR-0124) when a signed promotion claim is recorded.

Out of Scope

  • Actually signing the systems_software claim or updating the audit_passed_claims in reviewers.yaml (reserved for a follow-up promotion ADR).