Research

A proposal that must survive its strongest alternative

The evidence supports the ingredients—not HelmLoop’s necessity or superiority.

Markdown version ↗

HelmLoop

Evolution argument

External reference

Workflows use predefined code paths while agents dynamically choose processes and tools; added autonomy trades predictability, latency and cost. [4]

External reference

Modern runtimes already provide graph routing, persistence, interrupts, replay, handoffs, guardrails and traces. [1][2][3][5][7][8]

Design inference

HelmLoop proposes that authority, evidence freshness, promotion and observed outcomes deserve a portable contract boundary when graph completion alone cannot settle a consequential decision. [9][11]

HelmLoop

Alternative designs

01

Keep controls inside the graph

Best when one runtime, one team and one risk model own the full lifecycle. This is the strongest alternative.

02

Use a durable workflow engine

Best when replay, queues, timers and long-running reliability are the main problem.

03

Use an assurance SOP only

Best when validation and release governance matter but cross-cycle outcome learning does not.

04

Use HelmLoop contracts

Potentially useful when several runtimes share one accountable promotion and outcome policy—still unproven across runtimes.

HelmLoop

Counterarguments and failure modes

01

Over-engineering

L2/L3 governance can cost more than a reversible task is worth.

02

Self-confirming loop

The proposer, validator and learner may share the same model or context.

03

State explosion

Receipts, retries, exceptions and nested instances can become harder to govern than the graph.

04

Verification cost

Fresh independent evidence adds latency and provider expense.

05

Vendor coupling

Checkpoint or tracing semantics can leak through adapters and defeat portability.

06

Category inflation

HelmLoop may simply rename good application architecture. Failing portability tests should collapse the claim.

HelmLoop

Read the evidence at its actual scope

Current fixture results do not certify a live runtime, broad portability, production readiness or business value. The research page keeps these open questions visible.

  • No pair of distinct, fresh, host-attested, source-bound runtime observations exists in this repository; fixtures and runtime aliases are ineligible.
  • No production control plane or live provider reconciliation service is implemented here.
  • Cross-language semantic review is automated for shape and terminology, not independently human-attested.
  • Outcome benefit versus added governance cost has not been measured.

HelmLoop

Primary sources

Only official documentation, standards bodies and an original paper support external claims. Source annotations describe what each source supports—not what HelmLoop has implemented.

  1. 01 / OpenAI

    Agent orchestration ↗

    Manager-owned specialists and handoffs are distinct orchestration choices.

  2. 02 / OpenAI

    Guardrails — OpenAI Agents SDK ↗

    Guardrails attach at specific agent and tool boundaries; coverage is not universal.

  3. 03 / OpenAI

    Tracing — OpenAI Agents SDK ↗

    Traces record model, tool, handoff and guardrail events for debugging and monitoring.

  4. 04 / Anthropic

    Building Effective AI Agents ↗

    Distinguishes workflows from agents and advises adding complexity only when outcomes justify it.

  5. 05 / LangChain

    LangGraph overview ↗

    Describes durable, stateful graph orchestration and human-in-the-loop capabilities.

  6. 06 / LangChain

    LangGraph interrupts ↗

    Interrupts persist state and re-execute nodes, making idempotency material.

  7. 07 / Temporal

    Temporal Workflow ↗

    Event history and deterministic replay separate workflow code from external activities.

  8. 08 / Google

    Template agent workflows — ADK ↗

    Documents sequential, loop, parallel and newer graph workflow structures.

  9. 09 / Model Context Protocol

    MCP Architecture ↗

    Places consent, security policy and authorization decisions with the host boundary.

  10. 10 / Model Context Protocol

    MCP Authorization ↗

    Binds access tokens to intended resources and emphasizes least privilege.

  11. 11 / NIST

    AI Risk Management Framework ↗

    Organizes lifecycle risk work through Govern, Map, Measure and Manage.

  12. 12 / ICLR / arXiv

    ReAct: Synergizing Reasoning and Acting in Language Models ↗

    Shows interleaved reasoning and action; an observation updates a plan but is not a business outcome proof.