report-review

name	report-review
description	Review a SOLAR LINE episode or summary report for quality, accuracy, and readability. Checks data integrity, Japanese text quality, and orbital mechanics correctness.
argument-hint	["episode number or \"summary\" or \"all\" or \"final\""]

Report Quality Review

Two-stage systematic review for SOLAR LINE reports.

How to Use

Specify what to review with $ARGUMENTS:

• 1 through 5 — draft review of a specific episode report
• summary — draft review of all summary pages
• all — draft review of everything
• final — final review of deployed site (isolated, no codebase access)

Stage 1: Draft Review (with codebase access)

Reviewer Persona

IMPORTANT: All reviews must be conducted by agents with separate context (Task agents or Codex), NOT the current session. Set this system prompt for the reviewer:

あなたはSF作品と軌道力学に関心がある読者ですが、どちらにも専門的な知識はありません。 「SOLAR LINE」というアニメについて事前知識は一切ありません。 レポートを初めて読む一般読者として、以下の観点でレビューしてください：

• 専門用語の説明は十分か？初見の読者が理解できるか？
• 図表やチャートは分析をサポートしているか？何を示しているか明確か？
• 分析の論理展開は追えるか？飛躍はないか？
• 前提知識なしで内容を楽しめるか？ 率直にフィードバックしてください。

Implementation

Use a Task agent (Sonnet) for the draft review:

Task(subagent_type="general-purpose", model="sonnet", prompt=`
[Insert reviewer persona above as system context]

Review the report at reports/data/episodes/ep0X.md (or summary/*.md).
Read the report data, check against the checklist below, and provide
specific, actionable feedback.
`)

For physics-specific review, use Codex (gpt-5.2 reasoning model):

/nice-friend Review the physics calculations in EP0X.
Are the orbital mechanics claims well-supported?
Are the ΔV, transfer time, and mass calculations consistent?

Automated Checks (run first)

cd ts && npm test

Manual Review Checklist

1. Data Integrity

• All transferId references are consistent (transfers ↔ explorations)
• All evidenceQuoteIds reference existing dialogueQuotes
• computedDeltaV is not null where a calculation was performed
• Verdict is appropriate (not "indeterminate" for reference calculations)
• Source citations have valid URLs (not plain text)

2. Japanese Text Quality

• All UI text and analysis prose is in Japanese
• Character names use correct forms: きりたん, ケイ (ケストレルAI)
• Dialogue quotes match the きりたん「…」(MM:SS) format
• Timestamps are accurate (cross-reference with video)
• No untranslated English in body text (technical terms in English are OK)
• Terminology: "brachistochrone" preferred over ブラキストクローネ

3. Orbital Mechanics Accuracy

• ΔV calculations use correct formulas (vis-viva, brachistochrone)
• Units are consistent (km, km/s, seconds, AU)
• Planetary distances match ephemeris data for the given epoch
• Transfer times are physically plausible for the given ΔV and distance
• Cross-episode values are consistent (ship mass, Isp, thrust)

4. Report Structure

• Video cards at the top (YouTube + Niconico)
• Dialogue citations woven naturally into analysis text
• Most plausible scenario presented first
• Implausible scenarios collapsed by default (<details>)
• Explorations nested under parent transfers
• Orbital diagrams have animation config
• Scale legend and timeline annotations on complex diagrams

5. Narrative Plausibility

• Transit durations match the show's pacing
• Distances/conditions consistent with on-screen depiction
• Crew experience matches depicted reactions (SF G-tolerance assumed)
• Cold sleep NOT assumed unless explicitly depicted
• Ambiguity documented where the show is unclear

Stage 2: Final Review (deployed site, no codebase)

Purpose: Evaluate the reader experience of the deployed site without developer context.

Implementation

Launch a Task agent in isolation (no file access), using only WebFetch:

Task(subagent_type="general-purpose", model="sonnet", prompt=`
[Insert reviewer persona above]

You are reviewing the deployed website at:
https://sksat.github.io/solar-line/

Using WebFetch, visit the following pages and evaluate the reader experience:
1. Top page (index.html) — is the project purpose clear?
2. Episode 1 report — can you follow the analysis as a newcomer?
3. Cross-episode summary — does it provide useful synthesis?
4. Navigation — can you find what you need? Are links working?

Focus on:
- Layout and readability on the rendered page
- Navigation flow between pages
- Broken links or missing content
- Whether the analysis is accessible to the target audience
- Overall impression as a first-time visitor

Report your findings with specific page URLs and quotes.
`)

Final Review Checklist

• Top page communicates project purpose clearly
• Episode pages load correctly with all diagrams/charts
• Navigation dropdown menus work
• Cross-episode links function
• KaTeX formulas render properly
• Tables are readable
• Orbital diagrams have visible animations
• Source citation links are clickable
• GitHub repo link visible in footer
• Mobile-friendly layout (responsive design)

Efficiency Notes

• Run automated tests first — they catch most data integrity issues
• Use Sonnet subagents for draft review (separate context required)
• Use gpt-5.2 (reasoning model) for physics consultations via nice-friend
• Focus final review on reader experience, not data accuracy
• Don't fix issues during review — create separate tasks for fixes
• Record review findings in ideas/ for future sessions