refactor(seed): finalize minimal semantic core

[attractor-set]

Summary

This PR completes the semantic cleanup and minimization of ASET Seed.

It removes remaining legacy distinctions that were no longer represented by the normative wire model, sharpens the boundary between Seed-owned state and environment observations, makes the machine-readable operation model match the actual state-transition semantics, and further reduces the reachable formal state space without weakening Seed safety guarantees.

Main changes

Unified Authority recognition

The previous formal distinction between:

• RequestAuthorityBindings
• TerminalAuthorityBindings

is removed.

Seed now consumes a single exact-binding recognition relation:

• RecognizedAuthorityBindings

The same recognized (authority, binding) relation is used for both request registration and terminal submission.

This aligns the formal model with the normative wire representation and executable oracle, which already use one ResolutionAuthorityBinding concept.

Concrete signatures, delegation chains, certificates, and recognition mechanisms remain outside the Seed semantic kernel.

Conflict observation constrained to valid terminal state

Environment conflicts can now be observed only for resolution identifiers that already have accepted terminal state.

This removes unreachable formal states such as:

• conflict before request registration
• conflict before terminal submission

The formal invariant now explicitly requires:

conflicts ⊆ TerminalRequests

Conflict remains environment state and does not mutate Seed-owned state.

Accepted terminal uniqueness clarified

The previous broad TerminalUnique terminology is replaced by semantics that distinguish:

• the single terminal record accepted into Seed-owned state
• additional conflicting valid terminal material observed by the environment

This removes ambiguity between accepted canonical state and externally observed conflicting material.

Operations separated from state transitions

The machine-readable canon now models three Seed operations:

• REGISTER_REQUEST — state transition
• SUBMIT_RESOLUTION — state transition
• EVALUATE_RESOLUTION — observer

EVALUATE_RESOLUTION is no longer structurally presented as a state transition.

This aligns the machine-readable canon, documentation, executable semantics, and TLA model.

Machine canon schema cleanup

The active machine-readable schema is updated to the new operation-oriented representation.

Legacy SEED-TX-* transition identifiers are replaced by SEED-OP-* operation identifiers where appropriate.

The generated canon-to-TLA projection is updated to projection profile V5.

Standalone canon-to-TLA refinement preserved

SeedCanonProjection.tla remains a standalone generated projection and does not depend on the handwritten SeedResolution module.

The refinement proof explicitly relates the standalone machine-canon projection to the handwritten formal model.

Documentation and assurance cleanup

This PR also removes remaining active documentation inherited from earlier RC11/RC12 runtime-era architecture.

Notably:

• the black-box audit methodology is rewritten to describe the current Seed assurance model
• assurance terminology is aligned with the actual proof boundary
• historical assurance records are distinguished from controlling current records
• the final semantic cleanup is recorded as a new PDCA cycle
• residual grant/delegation-era wording is removed from active conformance documentation

Historical frozen evidence remains preserved.

Resulting semantic architecture

Immutable trust-boundary context
├── RecognizedAuthorityBindings
└── RecognizedTerminalCommitments

Seed-owned mutable state
├── requestMeta
└── terminalMeta

Environment state
└── conflicts

Derived state / observers
├── Requests
├── TerminalRequests
├── ResolutionOf
└── EffectPermitted

Operations
├── REGISTER_REQUEST       [state transition]
├── SUBMIT_RESOLUTION      [state transition]
└── EVALUATE_RESOLUTION    [observer]

Formal state-space reduction

Before this final cleanup:

reachable states: 2500
transitions:      5500

After constraining conflict observation and unifying Authority recognition:

reachable states: 1849
transitions:      3870
terminal states: 1548
saturated:        true

The reduction comes from removing formally representable states that were not possible under the normative wire semantics.

Assurance

Executable bounded model:

MODEL_CHECK_STATES=1849
MODEL_CHECK_TRANSITIONS=3870
MODEL_CHECK_TERMINAL_STATES=1548
MODEL_CHECK_FORMAL_PROPERTIES=14
MODEL_CHECK_SATURATED=true
MODEL_CHECK_VERDICT=PASS

Semantic mutations:

SEMANTIC_MUTATIONS=13
SEMANTIC_MUTATIONS_KILLED=13
SEMANTIC_MUTATIONS_SURVIVED=0
SEMANTIC_MUTATION_VERDICT=PASS

Python test suite:

54 passed

TLC:

TLC_MODEL_CHECK=PASS

TLAPS:

TLAPS_OBLIGATIONS=159
TLAPS_VERDICT=PASS

Canon-to-TLA refinement:

CANON_TLA_REFINEMENT_OBLIGATIONS=3
CANON_TLA_REFINEMENT_VERDICT=PASS

Additional assurance:

ASSURANCE_TRACEABILITY=PASS
PROOF_TRACEABILITY=PASS
CANON_TLA_REFINEMENT_CHECK=PASS
INVARIANT_COVERAGE=PASS
DOCUMENTATION_BLACKBOX=PASS
MANIFEST_CHECK=PASS
REPOSITORY_RELEASE_GATE=PASS

Scope

This PR deliberately does not add higher-level mechanisms to Seed.

The following remain outside the minimal semantic kernel:

• cryptographic algorithms
• signature formats
• delegation-chain protocols
• accumulators or Merkle structures
• execution orchestration
• persistence implementations
• federation logic
• higher-level ASET extensions

Seed consumes recognized boundary facts without prescribing how those facts are established.

Result

After this cleanup, the active Seed specification has:

• one Authority-recognition abstraction
• two Seed-owned mutable state dimensions
• one constrained environment conflict dimension
• two state-changing operations
• one observer operation
• an independently generated machine-canon formal projection
• explicit separation between semantic state, provenance, environment observations, and external trust recognition

This PR is intended to establish the minimal semantic baseline for the next Seed freeze rather than introduce new capabilities.