Authority: user-corrected canonical mission, effective 2026-09-15.

Before any Project Brain planning, research, execution, admission, replication, or scaling:

1. Verify CANONICAL_REAL_MISSION_V1.json SHA-256 28e790f37a43e3babcec148f69aa484cd1e1701886874e832e2b4ff190b61133.
2. Load the current authority set:
   • CANONICAL_CURRENT_STATE_20260915_V3.json — 332f6620306137a6f0463b4877d356140cdfb5ba527289545ce1b998809d2b18
   • CANONICAL_OPTIMAL_POLICY_V2.json — 46a5d8e7b5ab57fdda2b4d46f0fd98e51617d7aeb4044af85ac1a64b3df98d04
   • EXECUTION_FRONTIER_20260915_V2.json — 4ad520aaed8796df3568c29b6dc136b94063746af06638de206f37d86b4474e3
   • AUTONOMY_PERSISTENCE_POLICY_V1.json — 4815ea47496468dfb5dacd81a3e2ba175937f5223d0317653239cf02038d802f
   • SUPERHERO_PLAN_LESSONS_20260915_V1.json — de5e3566ac87910f2181b39e0ab6f93375c5a27b635494d1a4b807d6bdafde29
3. Activate current context through brain_context_v4.ts. Without an ACTIVE V4 receipt, the agent is UNBOOTED/STALE.
4. Worker claim/submit traffic uses worker_bridge_v5.ts or a verified stronger successor enforcing V4 context.
5. Canonical join protocol is CANONICAL_AGENT_JOIN_PROTOCOL_V4.json, SHA-256 7fab9b58d8ae96b6664167760d2ca53d46bc473aca055bac104fa780e707d528.

A superhero has exactly two core parts:

• Capabilities/tools: complete-system functional breadth and strength must become Astra-6-like.
• Knowledge: external/on-demand, explicit, retrievable, updateable, shareable, and aggressively externalized. Parameters are not civilization's knowledge warehouse.

Reasoning, planning, orchestration, perception, recovery, capability discovery and construction belong to capabilities. Identity, leases, provenance, verification, transport, registration, promotion, population accounting and benchmarks are support/proof infrastructure, not the superhero core.

• P_superhero_real = 0.
• Historical P_general_real = 1 and P_specialist_real = 22 remain historical only.
• All scaling is FROZEN.
• Hidden challenge ground truth must remain uninspected.
• $0 incremental spend is hard; existing credits may be used only with hard stops and no auto-reload/new spend.
• Capability claims must stay bounded to their tested envelope.

• Fresh external-knowledge/multi-source challenge: 16/16 PASS.
• Stateful DOM browser/recovery challenge: PASS.
• Original canvas challenge CANVAS-HI-1789506062251-89decfd0: non-DOM interaction path independently PASS. It reached a server-accepted success state after decoy recovery and visually read/clicked the four tiles. Exact-receipt scoring is invalid because the 49-character receipt is clipped by the fixed canvas. Independent classification: PASS_INTERACTION_EVALUATION_AMBIGUITY_RECEIPT_CLIPPED.
• Corrected sibling CANVAS2-HI-1789508254410-45b24b9c is frozen with hidden commitment 9e8030074f6471af83d38ab66e6f44f81fd0be04edbca6043b76d8238e48671c; hidden truth remains uninspected.
• All scored execution of that sibling is protected by account-global SQLite fence project_brain_scored_execution_lease_v1. Use only mosadek/project-brain-general-superhero-v1/run_canvas2_fenced_frozen_v1.ts, which pins fenced runner @339 and solver @337. Sealed fence evidence SHA: 652bb24777f2ada1ccd4741cf6fbab000a106e9d8225ec7351a39deb3eec3ea9. Never run the unfenced candidate directly.
• Cloudflare V4 is the demonstrated pixel/coordinate browser route, but its latest zero-cost launch returned HTTP 429 with Retry-After: 8482.
• TinyFish visually recognized the sibling controls but failed to make canvas clicks change state after 36 steps/120 seconds; it is falsified for this task in the tested run.
• Local hosted browser/proxy alternatives are unavailable because browser egress is blocked and server-side DNS resolution fails.

Terminal objective:

Minimize expected remaining wall-clock time to one independently verified real superhero.

At each evidence state, search, experiment, repair, compose, replace, invent, verify, wait on an evidenced external gate, or stop-searching compete under the same objective. A blocker is not automatically repaired if switching/composition is faster. When estimates are close, resolve decision-relevant uncertainty or prefer the more robust completion path.

The planning-stop rule remains TRIGGERED. Do not redesign the policy without new evidence showing the objective, constraints, state, or feasible action set is materially wrong.

Exact current action: RESUME_CORRECTED_CANVAS_SIBLING_WHEN_ZERO_COST_BROWSER_GATE_REOPENS.

A one-time scheduled continuation exists at the provider Retry-After boundary and a recurring Project Brain Autonomous Executor provides ongoing re-entry. On every wake: verify current V4 authority, run reconcile_canvas2_execution_v1.ts, respect the one-shot SQLite fence, and execute the candidate only through run_canvas2_fenced_frozen_v1.ts if the row is READY and the zero-cost browser gate is open. RUNNING means no duplicate; SUBMITTED means run only verify_canvas2_frozen_v1.ts and reconcile; VERIFIED/FAILED are terminal for this challenge. Persist/hash any material result and recompute globally.

Do not wait for the user to request saving. Any material verified success/failure, causal reclassification, next-action change, promotion/admission change, authority change, or new consequential evidence must be checkpointed before the execution session ends. Healthy no-change checks must not create checkpoint churn.

• Mission lock result: 7f9ef24651069db7bbaee3818ad71e1b6c26a69a3602f919902143bd3ae65bed.
• Current state/policy V3 lock: PROJECT-BRAIN-CURRENT-STATE-POLICY-LOCK-20260915-V3 — result 61e7e9d44ea8de853a40c183e5e245d2c259e5e054407062c0d172c8615e81d4.
• Join V4 protocol SHA-256: 7fab9b58d8ae96b6664167760d2ca53d46bc473aca055bac104fa780e707d528.
• Join V4 lock: PROJECT-BRAIN-JOIN-PROTOCOL-V4-LOCK-20260915 — result f5f81de604829d182a506f18333e667e675c9c002f6f9a447f679ef8926a6603.
• Canonical operator has an ACTIVE V4 receipt bound to the exact current hashes and next action.
• Kernel integrity passed on the current V3 state/policy and V4 join locks.