Civilization Bootstrap Adversarial Domain-Semantics Review v0¶
Status: PR5 review contract
This document records the second adversarial review pass for the Civilization Bootstrap seed graph. It focuses on graded SUPPORTED_BY semantics, broad capability concepts, cross-domain support, and relation-local scope interpretation.
Strength semantics¶
PR1 defines RelationStrength only for SUPPORTED_BY. In PR5, WEAK, MODERATE, and STRONG describe the material-help relation of one exact edge under that edge's exact scope.
RELATION STRENGTH != NECESSITY
RELATION STRENGTH != PROBABILITY
RELATION STRENGTH != DIFFICULTY
RELATION STRENGTH != LEARNING PRIORITY
RELATION STRENGTH != HUMAN IMPORTANCE
RELATION STRENGTH != CAPABILITY SCORE
A STRONG SUPPORTED_BY edge remains non-categorical. It must not be silently promoted to REQUIRES by a consumer.
Strength is also not a property of either endpoint in isolation:
Seed v0 contains deliberate counterexamples. microbiology_and_contamination_control supports different source capabilities with different strength values, and pump_systems has different strength values for different supporting capabilities. This prevents interpreting the enum as a global importance label attached to a node.
RelationStrength.rank provides the PR1 ordinal ordering of graded support metadata. PR5 does not define a ranking algorithm, weighted prerequisite score, path cost, recommendation priority, difficulty estimate, or graph-derived human score from those ranks.
Scope locality¶
RelationScope remains relation-local under PR1.
SAME RelationScope.key != GLOBAL SCOPE IDENTITY
SAME RelationScope.key != SHARED POLICY OBJECT
SAME RelationScope.key != AUTOMATIC COMPARABILITY
PR5 reuses human-readable keys such as conceptual_analysis, bounded_execution, and bench_validation as authoring vocabulary. Reuse does not create a global registry. Relation identity remains tied to source, relation kind, target, and its relation-local qualifier.
Consumers must not group unrelated edges into one semantic object merely because their scope keys have equal strings. A future globally governed scope vocabulary would require a separate explicit identity/versioning design.
Broad concepts and scoped claims¶
Some seed concepts intentionally cover a bounded system family rather than one atomic operation. Examples include electric_motor_systems, pump_systems, refrigeration_systems, and public_health_reasoning.
Their breadth does not collapse the PR2/PR3 separation:
A person-scoped assertion remains a CapabilityClaim with an explicit statement and scope. The PR4 ClaimDimensionBinding determines which competence-frame dimensions that exact governed claim contributes to. A broad concept therefore does not automatically imply execution, diagnosis, transfer, independence, or explanation.
Materially different meanings that cannot be bounded honestly by claim scope remain candidates for later concept split/new identity under PR1 semantic-governance rules. PR5 does not claim the v0 decomposition is final.
Cross-domain support¶
Cross-domain edges are useful precisely because the seed is not a school-subject tree. They remain local claims about material support, not global curriculum rules.
For example:
pump_systems SUPPORTED_BY fluid_pressure_and_flow
food_preservation SUPPORTED_BY microbiology_and_contamination_control
microcontroller_sensor_systems SUPPORTED_BY embedded_programming
These edges do not mean:
TARGET MUST BE MASTERED FIRST
TARGET IS MORE IMPORTANT
SOURCE IS HARDER
SUPPORT TRANSITIVELY IMPLIES ALL ANCESTORS
SUBJECT STATE PROPAGATES ALONG THE EDGE
PR5 defines no automatic transitive closure for dependency meaning. A path of SUPPORTED_BY edges is not itself a REQUIRES path.
Review outcome¶
The first ontology review repaired false SPECIALIZES family edges and overly strong categorical dependencies. This second pass found no additional production-data blocker requiring a graph rewrite.
Instead it freezes the interpretation boundary needed for the remaining graded support metadata:
SUPPORTED_BY STRENGTH IS EDGE-LOCAL AND SCOPE-LOCAL
STRONG SUPPORTED_BY != REQUIRES
GRAPH RELATIONS DO NOT DERIVE PERSONAL STATE
RELATION SCOPE KEYS ARE NOT GLOBAL IDENTITIES
Executable regressions protect these boundaries in tests/domains/test_civilization_bootstrap_adversarial_semantics_v0.py.
Non-goals¶
This review does not add a strength-calibration dataset, empirical prerequisite model, difficulty model, learning-order policy, recommendation engine, path optimizer, node-importance metric, global scope registry, or concept-splitting framework.