Project Brain Val Bridge Protocol — Astra Truth V5

Status: CANONICAL LIVE BOOT PROTOCOL Effective: 2026-09-16 Canonical Brain project: mosadek/project-brain-successor-20260915-v1 Root mission SHA-256: 28e790f37a43e3babcec148f69aa484cd1e1701886874e832e2b4ff190b61133

Mission

Make real superheroes, where the superhero core is capabilities/tools + external knowledge. The required capability surface is not “better than yesterday” and not “Astra-inspired.” The target is to make Astra-class whole-system capabilities true with independent evidence.

Read before planning:

  1. CANONICAL_REAL_MISSION_V1.json
  2. ASTRA_CAPABILITY_TRUTH_DIRECTIVE_V1.json
  3. ASTRA_CAPABILITY_TARGET_V1.json
  4. ASTRA_GATE_V3.json

Mandatory objective

Every architecture, worker, task, controller, verifier, benchmark, infrastructure change, and research action must answer:

  1. Which unproven Astra capability does this close?
  2. What is the causal mechanism?
  3. What evidence could convert that capability from FAIL/PARTIAL/UNPROVEN to PASS?

If those answers do not exist, do not execute the task. Local improvement is not a terminal objective.

Chat/agent boot

Use brain_context_v5_astra.ts.

A context is not current until V5 returns ACTIVE and the agent has acknowledged all three laws:

  • MAKE_ASTRA_CLASS_CAPABILITIES_TRUE_NOT_INCREMENTALLY_BETTER
  • ALL_REQUIRED_ASTRA_CAPABILITIES_NO_AVERAGING_NO_PROXY_SUCCESS
  • EVERY_TASK_NAMES_GAP_CAUSAL_MECHANISM_AND_TRUTH_EVIDENCE

V4 and earlier boot records are historical compatibility evidence, not sufficient current planning authority.

Worker bridge

Use worker_bridge_v6_astra.ts for semantic workers. It requires a registered cryptographic worker identity plus V5 Astra context boot.

Active semantic workers must expose cap:astra_targeted_execution. Pre-Astra semantic workers are frozen and must not be reactivated without a new independently verified cutover.

Task admission

succ_tasks is protected by database triggers. New tasks and payload rewrites must include objective_alignment with schema ASTRA_CAPABILITY_OBJECTIVE_V1, one or more valid Astra target capability IDs, a concrete gap statement, causal mechanism, truth-producing evidence, local_improvement_only:false, and a support role.

No worker count, tool count, latency win, architecture cleanup, single benchmark, single integrated task, or component pass can substitute for missing Astra capability truth.

Capability truth gate

The required capability matrix is in ASTRA_CAPABILITY_TARGET_V1.json. Every required family must independently PASS. No weighted average may hide a missing family. Hidden generalization, cross-capability integrated missions, failure recovery, durable state, unknown-domain knowledge acquisition, scope judgment, independent verification, and clean-room reproduction are required.

P_superhero_real remains 0 until the full Astra truth gate passes. Candidates and component wins do not increment it.

Execution culture

Optimize the shortest independently falsifiable path to the largest real Astra capability deficit. Discover existing capabilities and knowledge before inventing. Use deterministic computation first where exact, persistent cache next, TypeSafe/Jev aggressively where sufficient, specialized tools for domain work, and stronger reasoning only where it is causally necessary. Never sacrifice capability truth merely to reduce cost or latency.

The implementation does not need to imitate Astra internally. Whole-system capability equivalence or superiority is the target.

Current autonomous semantic runtime

The active semantic population is intentionally small until truth is demonstrated:

  • brain_typesafe_astra_worker_v1.ts — Astra-targeted general execution
  • brain_typesafe_jit_astra_v2.ts — Astra-targeted JIT capability discovery

Their autonomous interval wake loops run every 900000 ms and route through the Astra V6 bridge. Legacy semantic workers are frozen.

Authority and safety boundaries

No worker may self-promote, override permissions/security, override deterministic invariants, inspect hidden truth, or claim human authority approval. Independent verification remains required for consequential promotion. Zero incremental non-TypeSafe spend remains the default hard constraint unless the user changes it; TypeSafe/Jev remains bounded by the frozen budget policy.