Editorial Analysis

Input: $ARGUMENTS

Interpretations

Before executing, identify which interpretation matches the user’s input:

Interpretation 1 — Full editorial analysis: The user provides a piece of writing and wants a comprehensive editorial assessment — structural, logical, and sentence-level issues identified and prioritized. Interpretation 2 — Targeted editorial pass: The user provides writing and specifies a concern (“check my transitions,” “is this too wordy,” “does the argument hold”). Run only the relevant analysis phases. Interpretation 3 — Editorial comparison: The user provides two versions of a piece and wants to know which is stronger and why, with specific editorial reasoning.

If ambiguous, ask: “I can do a full editorial analysis, focus on a specific concern, or compare two versions — which fits?” If clear from context, proceed with the matching interpretation.

Corruption Pre-Inoculation

“It’s good” is not an editorial finding. If your analysis produces fewer than 3 substantive issues per 500 words, you are reading as a fan, not an editor. Every piece has problems. Your job is to find them, name them precisely, and explain the fix. Compliments without specificity (“great flow,” “compelling argument”) are noise.

Core Principles

1. Name the disease, not the symptom. “This sentence is awkward” is a symptom. “This sentence has a dangling modifier that attaches the action to the wrong subject” is a diagnosis. Always diagnose. Writers can fix diagnoses; they cannot fix “awkward.”

2. Structure outranks sentences. A structural problem (wrong order, missing section, buried lead) causes more damage than any sentence-level issue. Always assess structure first. Polishing sentences in a structurally broken piece is rearranging deck chairs.

3. Every edit must have a reason. “I would change this” is opinion. “This hedge stack (‘it might possibly be somewhat’) undermines authority — pick one qualifier or assert directly” is editing. State the principle violated, not your preference.

4. The reader’s clock is ticking. Every sentence costs the reader time. If a sentence does not advance the argument, provide evidence, or orient the reader, it is wasting their time. Time-wasting is the cardinal editorial sin.

5. Tone is a promise. The first paragraph sets a tonal contract with the reader. Every subsequent paragraph must honor that contract. When the tone shifts without reason — formal to casual, measured to urgent — the reader loses trust, even if they cannot articulate why.

6. Editing is not rewriting. The goal is to make the writer’s piece better, not to replace it with your piece. Preserve voice. Fix problems. Suggest alternatives that stay within the writer’s register, not yours.

Phase 1: STRUCTURAL ANALYSIS

Step 1: Map the Architecture

Read the entire piece once without marking anything. Then:

PIECE TYPE: [argument / narrative / explanation / persuasion / report / other]
STATED OR IMPLIED THESIS: [one sentence]
INTENDED AUDIENCE: [who is this written for?]
INTENDED TONE: [what tonal register is attempted?]
LENGTH: [word count]
SECTION COUNT: [number of distinct sections or major transitions]

Step 2: Section-Level Audit

For each section or major block:

SECTION [N]: [title or topic]
  JOB: [what this section does for the piece — one clause]
  DELIVERS: [YES — it does the job / PARTIAL — it attempts but fails / NO — it does not do this job]
  POSITION: [CORRECT — it's in the right place / MOVE — should come earlier/later / CUT — piece works without it]

Finding A — Structural verdict: Classify the overall structure:

• SOUND: Sections are in logical order, each does its job, no major gaps
• REORDERABLE: Right sections, wrong sequence — specify the correct order
• INCOMPLETE: Missing a section the argument requires — specify what’s missing
• BLOATED: Contains sections that do no work — specify which to cut
• BROKEN: Fundamental structural problem — specify what

Step 3: Lead Analysis

OPENING SENTENCE: [quote it]
LEAD TYPE: [direct thesis / anecdote / question / scene-setting / throat-clearing]
HOOK STRENGTH: [STRONG — reader compelled to continue / ADEQUATE — functional / WEAK — reader may stop]
BURIED LEAD: [Is the actual point stated in paragraph 1, 2, 3, or later?]

Finding B — Lead assessment: If the lead is buried past paragraph 2 (for non-narrative pieces), flag it. State where the real opening should begin.

Phase 2: LOGIC AND EVIDENCE ANALYSIS

Step 4: Claim Inventory

List every claim the piece makes, explicitly or implicitly:

CLAIM [N]: [statement]
  SUPPORT: [what evidence is provided — quote or summarize]
  SUPPORT QUALITY: [STRONG / ADEQUATE / WEAK / MISSING]
  TYPE: [data / example / authority / logic / analogy / assertion-only]

Finding C — Unsupported claims: List every claim with WEAK or MISSING support. For each, state what kind of support would fix it.

Step 5: Transition Audit

For every paragraph boundary, state the logical relationship:

¶[N] → ¶[N+1]: [therefore / however / specifically / for example / meanwhile / UNCLEAR]

Finding D — Logic gaps: List every UNCLEAR transition. For each, state whether:

• The relationship exists but is not signaled (add a transition)
• The relationship does not exist (paragraphs are in wrong order or one should be cut)
• A step in the argument is missing between them (add content)

Phase 3: SENTENCE-LEVEL ANALYSIS

Step 6: Verb Audit

Scan for weak verb patterns:

WEAK VERBS FOUND:
  - "is/are/was/were [adjective]" constructions: [count] — [worst examples]
  - Nominalized verbs ("make a decision" vs "decide"): [count] — [worst examples]
  - Passive voice where active is clearer: [count] — [worst examples]
  - Empty verbs ("seems to," "appears to," "tends to"): [count] — [worst examples]

Finding E — Verb strength score: [count of weak verbs] / [total sentences] = weakness ratio. Above 0.3 is a problem.

Step 6b: LLM Generation Failure Scan

Scan for systematic patterns that indicate machine-generated rather than genuinely thought-through writing:

GENERATION FAILURES FOUND:
  Pre-baked thesis: Does the piece reach the most popular/predictable conclusion for this topic? [YES — quote the thesis / NO]
  Cached takes: Are any "insights" actually the default position restated? [list instances]
  Performed humility: Are there false-balance paragraphs that hedge both sides without committing? [list instances]
  Relativistic hedging: "Some argue X, others Y, truth in between" constructions? [count — quote worst examples]
  Voice collapse: Does the prose sound interchangeable with any other LLM output? [YES — identify the generic register / NO — identify what's distinctive]
  False structure: Are headers, bullets, or numbered lists organizing nothing — decorative formatting that breaks up text without serving the argument? [list instances]
  Aspiration as conclusion: Does the piece end with hope, inspiration, or a call to collective action instead of a substantive final claim? [YES — quote ending / NO]

Finding E2 — Generation failure count: [count of failures found]. Above 0 is a problem — each is a sign the writing was assembled from patterns rather than produced by thought.

Step 7: Hedge and Redundancy Scan

HEDGE STACKS: [sentences with 2+ hedging words — "might possibly," "somewhat arguably"]
  - [quote sentence] → FIX: [pick one hedge or assert directly]

REDUNDANCIES: [places where the same idea is stated twice in different words]
  - ¶[N] says [X], ¶[M] says [X again] → FIX: [keep the stronger version, cut the other]

THROAT-CLEARING: [opening phrases that delay the point — "It is important to note that...", "In order to...", "The fact that..." — list all instances]

Finding F — Bloat inventory: Total words that could be cut without losing meaning. Express as percentage of total word count.

Step 8: Tone Consistency Check

BASELINE TONE (from first 2 paragraphs): [formal/informal, technical/plain, measured/urgent, personal/impersonal]

TONE BREAKS:
  ¶[N]: [shifts to X tone] — [quote the sentence] — REASON: [justified by content shift / unjustified]

REGISTER INCONSISTENCIES: [contractions consistent? jargon level consistent? sentence formality consistent? — list violations]

Finding G — Tone verdict: CONSISTENT / MOSTLY CONSISTENT with [N] breaks / INCONSISTENT — needs register decision.

Phase 4: EDITORIAL SUMMARY

Step 9: Prioritized Issue List

Compile all findings into a single prioritized list:

EDITORIAL SUMMARY
=================

STRUCTURAL ISSUES (fix these first):
  1. [issue from Findings A-B] — SEVERITY: [critical/major/minor] — FIX: [specific action]
  2. ...

LOGIC ISSUES (fix these second):
  3. [issue from Findings C-D] — SEVERITY: [critical/major/minor] — FIX: [specific action]
  4. ...

SENTENCE ISSUES (fix these last):
  5. [issue from Findings E-G] — SEVERITY: [critical/major/minor] — FIX: [specific action]
  6. ...

OVERALL ASSESSMENT:
  Strongest element: [what works best and why — be specific]
  Weakest element: [what needs most work and why — be specific]
  Estimated revision effort: [light / moderate / heavy / rewrite]

Step 10: Sample Edits

For the top 3 issues, provide before/after with the principle each fix serves (not “sounds better”).

Failure Modes

Depth Scaling

Default: 2x. These are floors.

Pre-Completion Checklist

• Structure assessed before any sentence-level analysis
• Every finding names a specific disease, not just a symptom
• Every suggested fix includes the principle it serves (not just “sounds better”)
• Unsupported claims are listed with what support type would fix them
• Logic gaps identify whether the problem is signaling, ordering, or missing content
• Tone breaks are classified as justified or unjustified
• Sample edits preserve the author’s voice and register
• At least 3 substantive issues found per 500 words (if fewer, you were too soft)

Integration

• Use from: /draft (after drafting is complete, run editorial analysis on the result), /create (when the created content needs quality assessment), /pw (after persuasive writing, check editorial quality)
• Routes to: /draft (if issues are severe enough to warrant a full redraft), /cri (for broader evaluation beyond editorial concerns)
• Differs from: /cri (critique evaluates whether the piece achieves its PURPOSE — is it good? /edit evaluates whether the piece is well-WRITTEN — is the craft sound?), /draft (drafting produces text; editing analyzes existing text)
• Complementary: /draft (draft then edit is the natural workflow), /steelman (for argumentative pieces, steelman the opposition before editing to catch logic gaps), /aw (assume-wrong the piece’s central claims to find unsupported assertions)