ARS compatibility for Hermes research pipeline

Purpose

This note adapts the academic-research-skills reference workflow to the current Hermes operating limits and the company cross-department pipeline.

What changed in this session

The user corrected two important research-workflow errors:

1. Phase discipline matters. Do not jump from early scanning to deep synthesis or master reporting before the topic and scope are explicitly locked.
2. Counter-hypotheses or open findings are not the same as research hypotheses. Hypotheses must be derived from the original research need, not reverse-engineered from preliminary findings.

Hard rules for future research runs

1. Do not skip phase order

For academic or policy research, preserve this order:

1. Phase 1: clarify the original research need and decision use case
2. Phase 2: first-round scanning and exploratory collection
3. Phase 3: lock the topic, scope, and actual research questions
4. Phase 4: literature review and theory-grounded synthesis
5. Phase 5: deep analysis, conclusions, or master report

If Phase 3 is not approved, do not run Phase 4 or Phase 5.

2. Hypothesis discipline

Do not label these as hypotheses:

• open findings
• preliminary tensions
• reviewer objections
• implementation caveats

A valid hypothesis in this workflow must:

• originate from the original research need
• state a testable relationship or expectation
• be falsifiable
• be distinguishable from later findings and reviewer commentary

3. Reviewer-panel adaptation under Hermes limits

The ARS reference repo assumes a broader reviewer panel. This Hermes environment currently supports at most 3 concurrent subagents and no nested delegation. Therefore the panel must be staged.

Recommended pattern:

Batch A

• Editor or scope reviewer
• methodology reviewer
• domain reviewer

Batch B

• perspective reviewer
• devil's advocate reviewer

Final synthesis

• one synthesis pass merges Batch A and Batch B
• user checkpoint follows before advancing phase

Do not promise or imply 5 simultaneous reviewers in this environment.

4. Cross-department governance stays intact

Even when borrowing ARS logic, preserve the company handoff rule:

• R&D drafts
• Grill-QA critiques
• CEO approves

Do not create new handoff subtasks just to mimic ARS stages. Adapt the internal reasoning and review method, not the board topology.

Practical implication for current skill evolution

When upgrading the Hermes research stack, prioritize:

1. phase checkpoints
2. hypothesis formulation discipline
3. staged reviewer batching for 3-agent concurrency
4. explicit synthesis gates before master reporting

Source repo inspected

Reference repo cloned in this session: /opt/ai-os/products/ceo/academic-research-skills

Most relevant source files reviewed:

• docs/SETUP.md
• docs/ARCHITECTURE.md
• deep-research/SKILL.md
• academic-paper-reviewer/SKILL.md
• academic-pipeline/SKILL.md