# What Happens During the Revision Coaching Sub-Stage (3→4) in the ARS Pipeline

> Understand revision coaching (3->4) in the ARS pipeline. Learn how this Socratic dialogue transforms editorial decisions into your revision strategy.

- Repository: [Edward Cheng-I Wu/academic-research-skills](https://github.com/Imbad0202/academic-research-skills)
- Tags: internals
- Published: 2026-05-13

---

**The Revision Coaching sub-stage is a human-gated Socratic dialogue between the user and the `eic_agent` that transforms editorial decisions into a user-formulated revision strategy before entering Stage 4.**

The Academic Research Skills (ARS) pipeline, maintained in the `Imbad0202/academic-research-skills` repository, implements a dedicated transition checkpoint between Stage 3 (Review) and Stage 4 (Revise). Known as the **Revision Coaching** sub-stage, this gate ensures that users actively engage with reviewer feedback through structured dialogue rather than passively receiving instructions. According to the architecture documentation in [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md) (lines 27-30), this checkpoint is marked as "🧑 3→4 Revision Coaching" and serves as a decision-heavy pause that requires explicit user participation.

## Understanding the Revision Coaching Gate

Unlike other pipeline stages that produce concrete artifacts, Revision Coaching focuses exclusively on **planning and strategy formulation**. The process activates immediately after the Editorial Board (EIC) issues a **Minor** or **Major** decision on the submitted work.

### Trigger Conditions and Architecture

The Revision Coaching gate appears in the architecture matrix as a human-gated transition (designated by the 🧑 symbol) that bridges the Review and Revise stages. Per [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md) (lines 46-48), the pipeline explicitly checks for non-Accept decisions before engaging this sub-stage. If the EIC returns an **Accept** verdict, the system bypasses coaching entirely and proceeds directly to final integrity checks.

### The Socratic Dialogue Method

The core mechanism is a **Socratic dialogue** managed by the `eic_agent`. This interaction runs for up to **8 rounds** (as documented in [`academic-pipeline/agents/pipeline_orchestrator_agent.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/agents/pipeline_orchestrator_agent.md), lines 10-20) and employs open-ended questions to guide the user's thinking. The protocol specified in [`academic-pipeline/references/two_stage_review_protocol.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/references/two_stage_review_protocol.md) (lines 12-14) mandates that this dialogue produce a **user-formulated revision strategy** rather than a revised document.

## The Four-Step Revision Coaching Workflow

The Revision Coaching sub-stage follows a structured protocol to ensure the user internalizes the feedback before attempting revisions.

### Step 1 – Present the Editorial Decision

The `eic_agent` first displays the complete editorial decision alongside the automatically generated **Revision Roadmap**. This output aggregates the reviewer panel's comments into a prioritized list of required changes.

### Step 2 – Launch Socratic Probing

The agent initiates the dialogue with targeted questions designed to surface the user's understanding and priorities. Typical probes include:

- "What surprised you most about the reviewer comments?"
- "Which three issues would you prioritize if you could only address a few?"
- "How would you restructure the methodology section based on the statistical power concerns raised?"

### Step 3 – User Formulates Strategy

Through iterative responses, the user gradually shapes a concrete **revision strategy**. During this phase, the user may reorder the roadmap, reject specific suggestions with justification, or identify additional improvements not originally flagged by reviewers.

### Step 4 – Hand-off to Stage 4

The dialogue terminates either after exhausting the 8-round limit or when the user explicitly opts out by stating **"just fix it for me"**. The final revision strategy is serialized and passed to the Stage 4 **Revision Coach** agent, which executes the concrete document changes based on this user-approved plan.

## Implementation Details and Code Examples

The Revision Coaching logic resides in the `pipeline_orchestrator_agent` module. Client applications interact with this stage through the ARS SDK's `PipelineOrchestrator` class.

The following Python pseudo-code demonstrates how to trigger and manage the coaching phase:

```python
from ars.pipeline import PipelineOrchestrator

# Initialize pipeline after Stage 3 completion

pipeline = PipelineOrchestrator(session_id="sess_123")

# Check decision type per academic-paper-reviewer/SKILL.md

if pipeline.decision in {"Minor", "Major"}:
    # Launch the Socratic coaching session

    revision_plan = pipeline.orchestrator.launch_revision_coaching(
        editorial_decision=pipeline.decision,
        revision_roadmap=pipeline.revision_roadmap,
        max_rounds=8  # Hard limit as per protocol

    )
    
    # Store the user-formulated strategy

    pipeline.set_revision_strategy(revision_plan)
    pipeline.advance_to_stage(4)  # Proceed to Revise

else:
    # Accept decision bypasses coaching

    pipeline.advance_to_stage(4.5)  # Jump to final integrity

```

For command-line interfaces, the coaching workflow wraps the same orchestration logic:

```bash
$ ars run --stage 3 --decision Minor
... (review reports displayed) ...

$ ars revision-coaching --max-rounds 8
> EIC: What surprised you most about the reviewer comments?
> You: The persistent concern about statistical power in Experiment 2.
> EIC: If you could change only three things, which would they be?
> You: 1) Add power analysis, 2) Clarify methodology section, 3) Strengthen discussion limitations.
Revision strategy saved. Proceeding to Stage 4 (Revise)…

```

## Key Source Files and References

- **[`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md)** (lines 27-30, 46-48): Defines the "🧑 3→4 Revision Coaching" gate location and round limits.
- **[`academic-pipeline/references/two_stage_review_protocol.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/references/two_stage_review_protocol.md)** (lines 12-14): Formal specification of the transition protocol and expected outputs.
- **[`academic-pipeline/agents/pipeline_orchestrator_agent.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/agents/pipeline_orchestrator_agent.md)** (lines 10-20): Implementation details for the Socratic dialogue flow and termination conditions.
- **[`academic-paper-reviewer/SKILL.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-paper-reviewer/SKILL.md)**: Describes activation conditions for Revision Coaching based on editorial decisions.

## Summary

- The Revision Coaching sub-stage acts as a **human-gated checkpoint** between Stage 3 (Review) and Stage 4 (Revise) that activates only on **Minor or Major** editorial decisions.
- It runs a **Socratic dialogue** via the `eic_agent` for up to **8 rounds**, using probing questions to help users formulate their own revision priorities.
- Users can exit early by stating **"just fix it for me"**, which skips the dialogue and passes a default roadmap to Stage 4.
- The output is a **user-formulated revision strategy**, not a revised document, ensuring that subsequent automated revisions align with the user's intent.
- Source references in [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md) and [`academic-pipeline/agents/pipeline_orchestrator_agent.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/agents/pipeline_orchestrator_agent.md) define the protocol limits and interaction patterns.

## Frequently Asked Questions

### What triggers the Revision Coaching sub-stage in the ARS pipeline?

The Revision Coaching gate activates exclusively when the `eic_agent` returns a **Minor** or **Major** decision. **Accept** decisions bypass this stage entirely and proceed directly to final integrity verification, as specified in the [`academic-paper-reviewer/SKILL.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-paper-reviewer/SKILL.md) logic.

### How many rounds of dialogue are allowed during Revision Coaching?

The protocol permits a maximum of **8 rounds** of Socratic dialogue. This limit is hard-coded in the `pipeline_orchestrator_agent` implementation (lines 10-20) to prevent infinite loops while ensuring sufficient depth for strategy formulation.

### What is the final output of the Revision Coaching stage?

The sub-stage produces a **user-formulated revision strategy** document that prioritizes and contextualizes the required changes. According to [`academic-pipeline/references/two_stage_review_protocol.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/academic-pipeline/references/two_stage_review_protocol.md), this strategy serves as the instruction set for the Stage 4 Revision Coach agent, not as the revised manuscript itself.

### Can I skip the Revision Coaching dialogue if I already know what changes to make?

Yes. The protocol allows users to terminate the coaching session early by stating **"just fix it for me"**. This command bypasses the remaining rounds of Socratic questioning and passes the auto-generated Revision Roadmap directly to Stage 4, as documented in [`docs/ARCHITECTURE.md`](https://github.com/Imbad0202/academic-research-skills/blob/main/docs/ARCHITECTURE.md) (lines 46-48).