Method Anti Slop

---
name: method-anti-slop
description: "Detects and removes generic AI-writing tells from finished scientific prose, keeping passive Methods voice. Use when the user asks to de-slop wording; drafting goes to method-write."
---

# Scientific Anti-Slop Audit

## Skill Value

- **Problem it solves:** Scientific drafts can contain generic AI phrasing, uncited attribution, vague claims, and technical synonym drift.
- **Use when:** Drafting or reviewing any clinical or scientific manuscript text.
- **Output:** A prioritized P0/P1/P2 anti-slop audit with direct rewrites where meaning is unambiguous.
- **Ask only if:** A technical term, citation boundary, or manuscript section cannot be identified from the draft.
- **Do not do:** Do not ban valid passive Methods prose or alter technical meaning for stylistic variety.

Catalog of AI-writing tells adapted for formal scientific prose, drawn from two open-source AI-writing-tell catalogs — `hardikpandya/stop-slop` and `conorbronsdon/avoid-ai-writing` — with the generic rules reconciled against the constraints of clinical manuscript house style (`method-voice`, `method-section-check`). Where a generic rule would break scientific convention, the calibration is stated explicitly below; do not apply the unqualified generic version to manuscript text.

## Calibration against the generic guides

The source catalogs ban passive voice outright, ban em-dashes outright, and treat any adverb or intensifier as a defect. None of that applies unmodified here:

- **Passive voice**: banned in Discussion/narrative prose for authorial actions and interpretation (use "we"), required and correct in Methods for procedural steps. Flag passive only outside Methods, or where it's used in Methods to obscure who made an author-level decision that should be attributed to "we."
- **Em-dashes**: banned generically; here, the em-dash is sanctioned when it embeds a definition ("automation bias — the tendency to..."). Flag any non-definitional em-dash (P1), and flag definitional em-dash pile-up within a paragraph as a density signal (P2) — convert excess uses to parenthetical or appositive form.
- **Intensifiers** ("significant," "substantial," "robust"): banned generically as filler; here they are correct only when grounded: "significant" names a prespecified test reported with its estimate and CI, "substantial" sits beside an effect size and interval, and "robust" points to named sensitivity analyses. The grounding sits in the same sentence or the one before it (in Methods and Results, the same sentence, per `method-section-check`). A nearby statistic is not enough when it contradicts the word (e.g., "a significant reduction (OR 0.91; 95% CI, 0.80-1.03)"). Flag the ungrounded instance, not the grounded one.
- **Synonym rotation**: generically flagged as thesaurus-abuse; in science, precise technical terms (a named outcome, a named test, a named model) should be repeated verbatim, not varied for style. Flag any place a manuscript swaps a defined technical term for a looser synonym — this is a correctness issue here, not just a style one.

Everything else below applies to manuscript prose largely as in the source catalogs.

## Banned constructions (all sections)

**Throat-clearing openers** — state the content directly instead: "Here's the thing," "It is important to note that," "It should be noted that," "The reality is," "It turns out," "The truth is."

**Meta-commentary** — the manuscript should move, not announce its own structure: "In this section, we will...," "As we'll see...," "The rest of this discussion explores...," "Let me walk you through."

**Uncited attribution** — never permitted in scientific writing regardless of AI-tell status: "studies show," "research suggests," "experts believe," "it is well established that" without a citation. Either cite the specific source or state the claim as this study's own finding. A citation supplied to repair this must come from a source that was read and resolves to a PMID or DOI (through `method-write`'s evidence access or `method-citations`); a citation recalled from memory is itself uncited attribution.

**Vague declaratives** — name the specific thing instead of announcing its importance: "The implications are significant" → state which implication; "The reasons are structural" → name the structural reason.

**Formulaic binary contrast** — "not only X but also Y," "This is not just a technical problem, it's a clinical one." State the point directly without the symmetric setup.

**Filler transitions and hedge-padding not covered by the intentional hedges in `method-voice`** — "at the end of the day," "when it comes to," "in a world where," "moving forward." These have no place in formal scientific register.

**Business-jargon substitutes for precise terms** — "leverage," "unpack," "landscape," "deep dive," "game-changer." Use the plain or technical term instead (analyze, examine, context, setting).

**Empty conclusions** — "the future looks bright," "only time will tell," "as we move forward." A Discussion closes with a stated implication or a specific next study, not a generic gesture at the future.

## Severity tiers for the audit pass

- **P0 (credibility killers)** — uncited "studies show"/"experts believe"; unsupported interpretation, retrospective justification, or nonstatistical caveats inside Methods or Results (hand off to `method-section-check`, preserving required or prespecified methodological rationale and factual statistical statements); a calibrated intensifier with no statistic attached; a technical term thesaurus-swapped mid-manuscript.
- **P1 (obvious AI tell)** — throat-clearing openers, meta-commentary, formulaic binary contrasts, vague declaratives, any non-definitional em-dash.
- **P2 (stylistic polish)** — filler transitions, business-jargon substitutes, minor rhythm monotony (three consecutive sentences of near-identical length and structure).

## Audit workflow

1. Read the draft section by section.
2. Mark every instance from the banned-constructions list, tagging it P0/P1/P2 per the tiers above.
3. Fix P0 items first — these are correctness or credibility failures, not style preferences.
4. Fix P1 items by direct rewrite: state the point without the throat-clearing or the symmetric setup.
5. P2 items are optional polish; note them but don't block delivery on P2 alone.
6. Re-read once more for rhythm: are sentence lengths varied per `method-voice`'s architecture guidance? Two consecutive sentences of the same length and shape is fine; three or more in a row reads as machine-generated.

For the combined gate across voice, section discipline, and this audit before a draft reaches an author, use `method-presubmit`.