PR8 progression-authority adversarial review v1¶

Status¶

This supplement records the first adversarial authority review of PR8 — Progression Frontier, Prerequisite Evidence Gap and Exploration Projection v1.

The review attacks ways an advisory projection could be overread or forged into a recommendation, prerequisite truth, whole-person capability statement, readiness decision, ranking, or self-confirming development path.

Central boundary¶

COULD BE CONSIDERED != SHOULD DO
PROGRESSION FRONTIER != RECOMMENDATION

Blocker 1 — structurally valid frontier could be forged independently of its stored effective inputs¶

Initial PR8 strict serialization guaranteed schema validity and deterministic roundtrip, but a caller could deserialize a structurally valid ProgressionFrontier whose stored focuses/seed bindings/prerequisite bindings/exploration inputs looked plausible while candidates, gaps, witnesses, policy, deriver, or rationale had been changed.

That made this implication invalid:

DESERIALIZED FRONTIER
    ->
ACTUALLY PRODUCED BY deterministic_progression_frontier_v1

Repair¶

PR8 now exposes validate_progression_frontier_v1(...).

The verifier:

reconstructs an exact ProgressionFrontierRequest from the frontier's stored effective inputs;
re-runs derive_progression_frontier_v1(...) against the supplied capability catalog, frame catalog, epistemic snapshot, and state snapshot;
requires exact equality between the re-derived frontier and the supplied frontier;
contains lower-layer validation failures inside the InvalidProgressionFrontier verification error domain.

Therefore:

STRUCTURALLY VALID FRONTIER != VERIFIED DERIVATION
DESERIALIZED FRONTIER != VERIFIED DERIVATION
SERIALIZED FRONTIER != VERIFIED DERIVATION

VERIFIED PR8 FRONTIER
    =
EXACT RE-DERIVATION FROM STORED EFFECTIVE INPUTS
AND SUPPLIED SOURCE SNAPSHOTS

The verifier is deterministic source-backed validation, not a signature, content-address, issuer authentication service, acceptance workflow, or publication authority.

Relation-direction inversion attack¶

Frozen PR1 dependency direction remains:

source = capability being described
 target = dependency/supporting capability

For a selected supported seed S, PR8 may surface candidate C only when the direct accepted relation is oriented:

C -> S

for one of the bounded adjacency kinds.

A dependency edge is not traversed in the opposite direction merely because the graph is connected.

DEPENDENCY EDGE != UNDIRECTED PROGRESSION EDGE
RELATION DIRECTION != PRESENTATION DIRECTION

Executable regression uses the real Civilization Bootstrap relation:

low_voltage_power_distribution
    REQUIRES
basic_electricity

and verifies that supported basic_electricity can surface low_voltage_power_distribution, not vice versa.

`SUPPORTED_BY -> REQUIRES` laundering attack¶

PrerequisiteCheckBinding is only consumed while inspecting an actual accepted REQUIRES edge of an actual frontier candidate.

A caller cannot provide the endpoints/scope of a SUPPORTED_BY relation and thereby manufacture a PrerequisiteEvidenceGap.

SUPPORTED_BY != REQUIRES
STRONG SUPPORTED_BY != REQUIRES
PREREQUISITE BINDING != RELATION-KIND OVERRIDE

A binding that does not correspond to a real REQUIRES edge of an actual frontier candidate fails closed as unused/invalid input.

Whole-capability aggregation attack¶

A state may contain, for the same capability concept:

conceptual_knowledge = SUPPORTED
calculation          = UNKNOWN

Selecting conceptual_knowledge as a frontier seed does not make calculation supported. Attempting to seed from the UNKNOWN dimension fails closed even though another dimension in the same state is supported.

ONE SUPPORTED DIMENSION != WHOLE-CAPABILITY SUPPORT
SUPPORTED DIMENSION A != SUPPORTED DIMENSION B
FRONTIER SEED BINDING != CAPABILITY AGGREGATION

Prerequisite-binding laundering boundary¶

PR8 deliberately does not infer a global mapping from RelationScope to competence dimensions.

A PrerequisiteCheckBinding explicitly states which exact frame and dimension keys a caller wants the deterministic projection to inspect for one exact REQUIRES relation.

The binding can be checked for structural and snapshot consistency, but PR8 v1 does not authenticate the semantic wisdom of that caller-selected mapping.

RELATION SCOPE != COMPETENCE DIMENSION
PREREQUISITE BINDING != GLOBAL SEMANTIC MAPPING
PREREQUISITE BINDING != AUTHORITY
REQUESTER-SUPPLIED DIMENSION SET != UNIVERSAL PREREQUISITE TRUTH

This is why a resulting gap means only:

under this explicit request-local binding, the selected governed state lacks represented support for one or more named dimensions.

It does not mean the person lacks the prerequisite capability.

Gap-as-prohibition/readiness attack¶

PR8 exposes no fields or helpers for:

ready
readiness
permitted
prohibited
can_attempt
must_not_attempt

A prerequisite evidence gap remains attached advisory information and does not remove the candidate.

PREREQUISITE EVIDENCE GAP != MISSING CAPABILITY
PREREQUISITE EVIDENCE GAP != PROHIBITION
PREREQUISITE EVIDENCE GAP != ACCESS CONTROL
NO GAP != READY
NO GAP != SAFE
NO GAP != PERMITTED

Hidden latest-state attack¶

Only exact state IDs named by FrontierSeedBinding or state-backed PrerequisiteCheckBinding are selected.

Unselected older/newer/stale states are inert even when present in the same PersonalCapabilityStateSet.

LATEST STATE != AUTOMATIC FRONTIER INPUT
NEWEST STATE != AUTHORITY
UNSELECTED STATE != FRONTIER INPUT

PR8 v1 performs no recency weighting or newest-wins selection.

Witness-count ranking attack¶

Multiple direct adjacency witnesses are preserved only as inspectable reasons why a candidate is present.

PR8 exposes no witness-count score, priority, rank, recommendation strength, distance, or utility value.

MULTIPLE WITNESSES != HIGHER PRIORITY
WITNESS COUNT != RECOMMENDATION STRENGTH
RELATION STRENGTH != FRONTIER PRIORITY
CANONICAL ORDER != RANKING

Exploration self-confirmation attack¶

ExplorationInput is explicit request input. PR8 does not accept a previous frontier, history, achievement record, or Legend as derivation input and does not automatically promote prior exploration into later focus or adjacency seeds.

An exploration-only request therefore produces exploration opportunities and no frontier candidates.

A caller may explicitly choose to submit the same concept again in a later request, but that is a new attributed input, not hidden feedback from PR8.

EXPLORATION OPPORTUNITY != FUTURE SEED
EXPLORATION OPPORTUNITY != SUBJECT INTEREST
PREVIOUS FRONTIER != AUTOMATIC NEXT-FRONTIER INPUT
EXPLICIT RE-SUBMISSION != HIDDEN SELF-CONFIRMATION

Model/requester non-authority¶

ProgressionRequesterRef(kind=MODEL, ...) records who supplied request-local focus/exploration assumptions. It does not change those assumptions into goals, interests, identity, or authority.

MODEL-SUPPLIED FOCUS != SUBJECT GOAL
MODEL-SUPPLIED EXPLORATION != SUBJECT INTEREST
REQUESTER != AUTHORITY
DERIVER != AUTHORITY

Frozen verification and persistence limits¶

POLICY REF != AUTHENTICATED POLICY CONTENT
DERIVER REF != AUTHENTICATED GLOBAL IDENTITY
FRONTIER ID != CONTENT HASH
SNAPSHOT-LOCAL UNIQUENESS != GLOBAL UNIQUENESS
STRICT JSON != SOURCE AUTHENTICATION
VERIFIED DERIVATION != GOVERNANCE ACCEPTANCE
SERIALIZED FRONTIER != PUBLISHED FRONTIER

A future durable store may preserve source snapshot digests, signatures, policy content, or acceptance decisions under separate contracts. PR8 does not imply those guarantees.

Non-goals of this review¶

This review does not add ranking, scoring, path search, semantic embeddings, LLM runtime, global scope-to-dimension mappings, recommendation authority, permission/safety inference, persistence, sync, signatures, content-addressed snapshots, Player Window presentation policy, or a generic recommender framework.