Opsle Research

What if we stopped using intelligence for work that doesn’t require intelligence?

Opsle researches how much work in agentic AI actually requires model cognition and how much can be safely moved into deterministic systems.

Operating rule

DOES THIS OPERATION REQUIRE COGNITION?
                 │
            ┌────┴────┐
            │         │
           YES        NO
            │         │
          MODEL    SOFTWARE
            │         │
            └────┬────┘
                 │
              MEASURE

When the answer is unclear: experiment → measure correctness → measure efficiency → keep, change, or reject.

Research lifecycle

Discover → isolate → formalize → specify → falsify → benchmark → publish → optionally integrate later.

Important mechanisms should remain understandable, falsifiable, benchmarkable, reusable, removable, and traceable outside a large product.

Start here

• PROGRAM_STATUS.md — generated current and historical repository dashboard.
• program/registry.json — authoritative machine-readable research ledger (Tasks owns workload management).
• program/PRIORITY.md — generated NOW / NEXT / THEN / LATER / PARKED view.
• program/THEORY_MAP.md — canonical conceptual topology and Gearbox boundary.
• program/theory-registry.json — machine-readable concept classifications and dispositions.
• program/experiments.json — canonical experiment registry.
• experiments/exp-001/ — frozen provider-free EXP-001 benchmark, launch preregistration, and one-block coordinator with zero subject/model runs.
• program/evidence/exp-001-preregistration/ — provider-free allocation, adapter, interoperability, and Visible Value verification.
• program/evidence/exp-001-block-coordinator/ — secret-backed deterministic coordinator qualification with fixture-only authorization and zero experiment results.
• program/LIFECYCLE.md — evidence gates for lifecycle promotion.
• program/OPERATING_RULES.md — mandatory execution rules.
• CONCEPTS.md — concept overview and subordinate concepts.
• METHODOLOGY.md — evidence and experiment rules.
• MATURITY.md — legacy bootstrap maturity labels.
• ARCHITECTURE.md — cross-repository system map.
• OPEN_QUESTIONS.md — what remains unanswered.
• OPSLE_SITE_PLAN.md — future opsle.com content architecture.

Validate ledger integrity with python3 tools/validate_program.py. Regenerate the dashboard and priority view with python3 tools/render_program_status.py.

Product relationship

Opsle Tasks is the authoritative current workload and task-management system and integrated reference implementation in opsle/tasks. Its current default combines the control plane and local execution in one Incus instance, with optional SSH targets and a separate release broker. It does not depend on this research repository at runtime. Exact architecture and Notebook recovery evidence are recorded in program/registry.json.

Integrity

Hypotheses are not facts. Feasibility in one application does not prove general superiority. Negative and failed experiments remain part of the record. No private chain-of-thought is collected; experiments use observable artifacts only.

License

This research control plane is Apache-2.0. The separately released Gearbox repository preserves the predecessor's AGPL-3.0-only license and provenance.