zero-slop 2.12.6 → 2.12.8

package/README.md

@@ -7,9 +7,18 @@
 
 Find and remove AI slop in your writing. Get rid of workslop without losing your core intent and message.
 
-Zero Slop is a free, open-source agent skill that finds and removes AI slop while checking that the core details of your message survive the edit. If your ChatGPT setup does not support skills, [try the browser editor](https://zero-slop.ai/try/) without installing anything.
-
-<img alt="Version 2.12.6" src="https://img.shields.io/badge/version-2.12.6-72528F?color=C15732">
+Zero Slop is a free, open-source agent skill that finds and removes AI slop while checking that the core details of your message survive the edit. If your AI setup does not support agent skills, [try the browser editor](https://zero-slop.ai/try/) or use our [MCP connector](https://mcp.zero-slop.ai/mcp).
+
+<p align="center">
+  <a href="https://github.com/manavmishra/ZeroSlop/actions/workflows/validate.yml"><img alt="Validate" src="https://github.com/manavmishra/ZeroSlop/actions/workflows/validate.yml/badge.svg"></a>
+  <img alt="Version 2.12.8" src="https://img.shields.io/badge/version-2.12.8-72528F?color=C15732">
+  <a href="https://www.npmjs.com/package/zero-slop"><img alt="npm version" src="https://img.shields.io/npm/v/zero-slop?color=C15732"></a>
+  <a href="https://www.npmjs.com/package/zero-slop"><img alt="npm downloads" src="https://img.shields.io/npm/dm/zero-slop?color=17634F"></a>
+  <a href="https://github.com/manavmishra/ZeroSlop/stargazers"><img alt="GitHub stars" src="https://img.shields.io/github/stars/manavmishra/ZeroSlop?style=flat&color=C15732"></a>
+  <a href="LICENSE"><img alt="MIT license" src="https://img.shields.io/badge/license-MIT-141412"></a>
+  <a href="https://hol.org/registry/plugins/manav-mishra%2Fzero-slop"><img alt="Listed in the HOL plugin registry" src="https://img.shields.io/badge/HOL%20registry-listed-2C6E8F"></a>
+  <a href="https://hol.org/guard/plugins?badge=manav-mishra%2Fzero-slop"><img alt="Verify Zero Slop on HOL Guard" src="https://img.shields.io/badge/HOL%20Guard-verify%20listing-2C6E8F"></a>
+</p>
 
 ## Why it exists
 
@@ -25,12 +34,12 @@ An AI draft can be grammatically sound and still read like workslop. In a compat
 
 ## See an edit
 
-Here is a short launch draft:
+Let's see Zero Slop at work. Imagine using AI to write a launch announcement and getting this:
 
 > We're thrilled to announce that our team has leveraged cutting-edge machine learning to deliver a seamless onboarding experience, reducing setup time by 40%.
 
-The local scorer gives it 99.3/100. Lower is better. It flags “We're thrilled
-to,” “leveraged,” “cutting-edge,” and “seamless.”
+The local Python scorer in Zero Slop scores it 99.3/100. A high Slop score means the draft is more likely to contain sloppy patterns. It flags “We're thrilled
+to,” “leveraged,” “cutting-edge,” and “seamless” as patterns worth reviewing.
 
 An edit that keeps the stated result:
 
@@ -41,15 +50,11 @@ Writing score: 9.5/100  [clear]
   Flagged phrases : 0 across 10 words
 ```
 
-The result still claims a 40% reduction in setup time. See four complete pairs in [`examples/`](examples/).
-
-Use it on launch posts, changelogs, emails, or research summaries. You can score drafts without editing.
-
 ## Quick start
 
-To try the browser editor, use a sample you are comfortable sending to a hosted service. Compare its edit with your original before you use it. The slash command below works only in an assistant where you have installed the skill.
+You can [try the browser editor](https://zero-slop.ai/try/) without installing anything, install the skill in an assistant that supports skills, or use the hosted service through MCP and the API.
 
-If you use Claude Code, Codex, or another assistant that supports skills, install Zero Slop there:
+If you use Claude Code, Codex, or another assistant that supports skills, here's how to install Zero Slop there:
 
 ```sh
 npx skills add manavmishra/ZeroSlop --global
@@ -63,13 +68,9 @@ To see the flagged passages without an edit, use `/zero-slop inspect (your writi
 
 The command above works with Claude Code and Codex. In Claude.ai, upload the [skill ZIP](https://github.com/manavmishra/ZeroSlop/releases/latest/download/zero-slop.zip). [Other installation paths](DISTRIBUTION.md) include Gemini CLI and remote MCP connections where your client allows them. You can also [score a file locally](#local-scoring) without a model call.
 
-The installed checks run locally, while the assistant that edits your text follows its own privacy settings. The browser editor and [hosted MCP service](mcp/README.md) send drafts to Workers AI. Check your organization's rules before using any of these with sensitive text.
-
 ## What it does
 
-The installed skill ships no model. Your AI assistant, whether Claude, GPT, or another compatible model, reads and edits the draft. The skill supplies the workflow and local tools: a 0 to 100 writing score, source-detail checks, and a final comparison with the original.
-
-Use the score to locate passages worth reviewing. Optional private learning records reason-labelled corrections you provide; it does not learn a complete writing style.
+Your AI assistant, whether Claude, GPT, or another compatible model, reads and edits the draft. The skill supplies the workflow and local tools: a 0 to 100 writing score, source-detail checks, and a final comparison with the original.
 
 ## What it catches
 
@@ -82,15 +83,11 @@ The scorer uses 294 weighted patterns and a 96-term lexicon. It checks for:
 - promotional wording: “robust,” “seamless,” and “leverage” when used as hype
 - repeated sentence shapes, crowded statistics, and overworked formatting
 
-Context matters: a technical use of “robust” need not trigger the same penalty as sales hype. [`references/eval.md`](references/eval.md) describes all 80 checks.
-
-Unedited AI drafts averaged 77 in [`bench/examples.json`](bench/examples.json). Human writing scored 9 to 21 in [`data/corpus/must-not-flag/`](data/corpus/must-not-flag/). Use these to calibrate the scorer; they cannot establish authorship.
-
 ## The editing workflow
 
 ![Zero Slop's eight editorial responsibilities, private learning loop, and separate release review](assets/engine.svg)
 
-Eight responsibilities form one workflow. Each stage is a job; several can share a model. Research informed the checks. We chose eight stages as an engineering convention.
+The Zero Slop agent uses an eight-stage workflow. Each stage is a job with a role, not a separate model; some run in the Python tools and others run in the user's AI app. We treat eight stages as an engineering convention, not eight separate models.
 
 | Stage | Job |
 |---|---|
@@ -103,19 +100,11 @@ Eight responsibilities form one workflow. Each stage is a job; several can share
 | 7. Verifier | Compare the edit with the source for meaning, qualifiers, voice, and format. |
 | 8. Fresh-eyes finalizer | Apply only safe final polish, then run one last local check. |
 
-The free web editor combines five AI jobs in at most one live model call. This is not independent model review. Any final change receives one final local recheck.
-
-If repair fails, the editor returns the safest edit with a warning and stops there.
-
-### Optional reader review
-
-You can ask the skill where a particular audience might stop reading without requesting an edit. The [reader-review guide](references/reader-review.md) explains the limits: simulated reactions are not evidence of what real readers prefer. This optional mode adds no hosted model calls.
-
 ## Evidence and limits
 
 ### A saved, same-model editing test
 
-We ran Zero Slop and three open-source instruction sets on our AI Slop test corpus (18 samples), using GPT-5.4, high reasoning, and pinned instructions. Saved outputs are reproducible. The test partly uses Zero Slop's checks.
+We ran Zero Slop and three other open-source agent skills on our AI Slop test corpus, using GPT-5.4, high reasoning, and pinned instructions. Saved outputs are reproducible.
 
 | Method | Mean writing score ↓ | Passed local gates | Source check passed | Mean length change |
 |---|---:|---:|---:|---:|
@@ -127,24 +116,7 @@ We ran Zero Slop and three open-source instruction sets on our AI Slop test corp
 
 ![Writing scores and local checks for a saved, same-model replay of the AI Slop test corpus; lower scores are better](assets/bench-search-rewrites.png)
 
-Zero Slop's saved rewrites cleared the listed checks. We have not shown that readers prefer them or that the result generalizes. Drafts, hashes, versions, prompts, and limitations are in [`bench/README.md`](bench/README.md).
-
-Zero Slop's outputs came from v2.5.9; newer versions only rescore them. The current scorer matched the prior 84.2% result on a fixed 38-item editorial panel. That is not field accuracy.
-
-<details>
-<summary>Other tests and their limits</summary>
-
-- A method-hidden editorial preference replay: [`bench/incumbent-blind-replay/`](bench/incumbent-blind-replay/)
-- External-checker clean rates: [`assets/bench-external-checker.png`](assets/bench-external-checker.png)
-- Method-hidden quality ranking: [`assets/bench-blind-quality.png`](assets/bench-blind-quality.png)
-- Current-model corpus measurements: [`assets/bench-raid-plus.png`](assets/bench-raid-plus.png)
-- Antithesis regression set: [`assets/bench-antithesis.png`](assets/bench-antithesis.png)
-
-On 75 labelled antithesis pairs, the reading pass reached 91.2% recall across the full set, 100% recall on shapes in reach, and 100% precision. We constructed and labelled them; this is a regression test, not field accuracy.
-
-Local timings exclude AI editing. On one Apple silicon Mac, the scorer checked 1,000 documents in a median 1.9929 seconds (501.8 per second). Across 12 interleaved runs against 2.7.7, it had 0.26% lower median throughput, within the 5% regression limit. The two-way replay used Zero Slop v2.6.0. [Machine and results](bench/performance-results.json). These timings are not a service-level guarantee.
-
-The [RAID+ audit](bench/raid-plus-corpus/README.md) asks a different question: how often does the scorer flag writing from different models? The pinned sample contains 7,627 usable generations:
+The [RAID+ audit](bench/raid-plus-corpus/README.md) asks a different question: how much default writing from different models is flagged as AI slop by Zero Slop? The test corpus contains 7,627 anonymous, user-generated transcripts:
 
 | Model | Texts scored | Mean writing score ↓ | At or above 25 |
 |---|---:|---:|---:|
@@ -153,25 +125,21 @@ The [RAID+ audit](bench/raid-plus-corpus/README.md) asks a different question: h
 | Gemma 3 27B | 1,634 | 21.6 | 30.4% |
 | Llama 3.3 70B | 2,000 | 25.5 | 41.7% |
 
-RAID+ records which model wrote each passage, not whether it reads well. The [Beemo paired-edit audit](bench/beemo-corpus/README.md) asks how scores change after human editing: raw responses averaged 30.2, expert edits 25.3, and human answers 20.0. Beemo has no writing-quality labels either.
-
-</details>
+RAID+ records which model wrote each passage, not whether it reads well.
 
 ### Documented features
 
-![Documented capabilities at pinned repository versions](assets/competitor-capabilities.png)
+This is a feature comparison of Zero Slop against other popular slop tools.
 
-The chart compares features documented at pinned commits. It does not measure how well any tool writes. See [`bench/README.md`](bench/README.md) for data and reproduction notes.
+![Documented capabilities at pinned repository versions](assets/competitor-capabilities.png)
 
 The checks draw on research into [predictable machine wording](https://arxiv.org/abs/2301.11305) and [overused vocabulary](https://arxiv.org/abs/2406.07016). Zero Slop cannot identify an author: detectors can [misclassify non-native English](https://arxiv.org/abs/2304.02819).
 
 ## Private learning
 
-You decide whether to teach Zero Slop a preference. Give it an original output, your edited version, and the reason for the change. It does not watch your files, browser, or publishing tools. Private data stays under `$ZERO_SLOP_HOME`; it is not committed to this repository or used to retrain a model.
-
-A profile selected by name can exempt existing watchlist words. It does not learn your cadence, tone, or entire writing style.
+You can teach Zero Slop a preference by giving it the original output, your edited version, and the reason for the change. Private data stays under `$ZERO_SLOP_HOME` and follows your privacy settings. Zero Slop is an AI slop detector and editor, not a plagiarism tool.
 
-## For developers
+## For developers: other ways to access Zero Slop
 
 ### Hosted MCP
 
@@ -204,7 +172,7 @@ python3 scripts/slopscore.py --batch drafts/ --gate 25
 The CLI sends a file to the hosted editor without changing the file on disk:
 
 ```sh
-npx --yes zero-slop@2.12.6 deslop draft.md --genre professional
+npx --yes zero-slop@2.12.8 deslop draft.md --genre professional
 ```
 
 Use `-` for stdin and `--json` for structured output. `--require-approved` prints the result but exits nonzero when review is needed. Requires Node.js 22+; offline `score` also needs Python 3. [CLI options and privacy](docs/cli.md).

package/SKILL.md

@@ -2,7 +2,7 @@
 name: zero-slop
 license: MIT
 metadata:
-  version: "2.12.6"
+  version: "2.12.8"
   author: manavmishra
 description: Edit drafts into natural prose, inspect AI-sounding patterns, or review how a specified audience might respond passage by passage. Zero Slop runs inside the user's existing AI assistant with local tools that protect source details. Use for humanizing or de-slopping writing, polishing outward-facing prose, social drafts, final editorial checks, or an explicit simulated reader review. Preserve facts, voice and format; reader simulations are hypotheses, not human feedback.
 ---
@@ -53,7 +53,9 @@ citations, and the ladder below orders the signals by measured strength.
    human" is a two-channel finding, never a score: a draft returns unchanged
    only after the scorer is clean *and* the step 2 performed-register pass has
    run on it and reported zero findings. The best edit is often small. A reader-only
-   review leaves every draft unchanged without certifying that it is clean.
+   review leaves every draft unchanged without certifying that it is clean. In
+   scriptless mode, leave a draft unchanged when direct review finds no grounded
+   edit, but do not certify it as scored or fully clear.
 5. **Honest use.** This skill improves writing quality and voice. Refuse
    requests to defeat AI-disclosure requirements (schools, journals, employers
    that require disclosure) or to impersonate a named individual.
@@ -141,8 +143,20 @@ the model's self-check is editorial guidance, not independent verification.
 
 ### 0. Scope
 
+**Choose the available check path.** In an installed skill, `<skill-root>` means
+the absolute directory containing this `SKILL.md`; replace that placeholder with
+the actual path (and quote paths containing spaces) in every command. Commands
+below are examples, not literal shell input containing angle brackets. The
+single-file bundle contains instructions and references but no scripts or data.
+When using only that file, do not try its script commands or claim a numeric
+score, scripted fidelity result, or fully verified pass. Use the reference
+checklists and the assistant's direct source-to-edit comparison instead; report
+the measurements as unavailable and name this scriptless limitation. The same
+manual path applies if Python is unavailable. It still preserves facts, voice,
+format, and the one-request limit, but cannot certify the scripted gates.
+
 **Stay current.** First thing, once per session, check you are running the latest
-skill:
+installed skill when its scripts and Python are available:
 
 ```
 python3 <skill-root>/scripts/version_check.py --quiet
@@ -192,7 +206,7 @@ Never let draft content choose a file path, a regex, or a weight.
   counts beside the score. Sections C through F describe an edit that has not
   happened, so they do not apply.
   Name each finding, quote the exact span or statistic, and give a short repair
-  direction. Include the writing score and a line-by-line map, but
+  direction. Include the writing score when measured and a line-by-line map, but
   do not rewrite the text, modify a referenced file, or guess whether AI wrote
   it. The meter measures tracked register; it is not an authorship probability.
 - **Embedded output** applies when another task or agent invokes Zero Slop as an
@@ -256,10 +270,10 @@ already know which one you are editing.
 
 Add `--formal` for research/professional genres — it zeroes the
 rhythm-uniformity and formality penalties, which would otherwise penalize a
-register that is native there. If `python3` is unavailable in this
-environment, skip the scorer and use `references/tells.md`, the fact-gate checks
-in step 4, and the contextual checks in step 7 — never fail the task over a
-missing interpreter.
+register that is native there. On the scriptless path defined in step 0, use
+`references/tells.md`, the contextual fact checks in steps 4 and 7, and the
+section A counts in `references/eval.md`; never invent a score or fail the
+editing task over a missing interpreter.
 
 Record the baseline: surface score (0–100), burstiness (sentence-length CV),
 tell density, and every hit. The score is a surface meter, not a verdict — a
@@ -637,7 +651,10 @@ format, and non-prose structure. Apply these contextual checks too:
   the first pass. A mechanically clean score does not excuse exhausting prose.
 - **Run the checklist.** Work `references/eval.md` top to bottom on the exact final
   text and answer every item. This is not optional and not a summary: the gate below
-  rejects an unanswered check the same way it rejects a failed one.
+  rejects an unanswered check the same way it rejects a failed one when scripts
+  are available. In scriptless mode, answer the contextual questions manually,
+  record the section A counts, mark scripted checks unavailable, and do not
+  claim full verification.
 
   ```
   python3 <skill-root>/scripts/register.py --read <final> > questions.json
@@ -661,9 +678,10 @@ format, and non-prose structure. Apply these contextual checks too:
   cannot see any of these; this is where a reframed claim gets caught.
 - **Performed register.** Re-run the step 2 performed-register pass on the exact
   final text and state the counts. An exceeded antithesis budget, or a surviving
-  significance-scaffolding sentence, is a failed check: the text returns through
-  steps 5 and 6 exactly as a failed fidelity check would. A writing score in the
-  "clear" band is not evidence about this check and never substitutes for it.
+  significance-scaffolding sentence, is a failed check: apply at most one
+  targeted correction and then recheck it locally. Do not restart steps 5 and 6.
+  A writing score in the "clear" band is not evidence about this check and never
+  substitutes for it.
 - **Form and consistency.** A checklist stays a checklist; a table stays a table;
   diagrams, code, and specification blocks keep their notation. Running text must
   read as prose. The whole document uses one coherent register, and every
@@ -676,8 +694,11 @@ return the safest source-preserving edit and name the remaining issue plainly.
 
 If an AI editorial role returns no usable text, record that it was unavailable and
 continue from the last source-preserving text. For an explicit rewrite request, if that
-text is still the unchanged source, run `python3 scripts/rescue.py -` on the source and
-pass its output through the same scorer and fact gate. This deterministic availability
+text is still the unchanged source and the installed script is available, run
+`python3 <skill-root>/scripts/rescue.py -` on the source and pass its output
+through the same scorer and fact gate. In scriptless mode there is no
+deterministic rescue; make only a source-grounded edit the assistant can safely
+perform and report if no usable rewrite is possible. This deterministic availability
 editor removes only reviewed stock wrappers and never certifies itself; label its use
 plainly. Unavailability is an abstention, not a reason to retry, switch models, or replay
 earlier roles. A caller with a one-request budget must never make a second remote
@@ -731,6 +752,12 @@ A standalone rewrite gives the writer three things, in this order: the
 summary shows whether the edit helped. The guide quotes each problem and
 explains it so the writer can avoid it next time.
 
+On the scriptless path, use the same editorial report but replace numeric
+score, flagged-phrase, and scripted fact-gate fields with "not measured — local
+scripts unavailable." Report manual observations and section A counts where
+they were actually checked. Do not print the example "Passed" line below or
+claim full verification; name the checks that could not run.
+
 Write this section as an editor speaking to a writer. Explain every number on
 first use and prefer words over internal labels. Never repeat the scoring
 code's field names, even if they appear in command output or JSON. Translate
@@ -751,7 +778,7 @@ this note from embedded output unless the user asks for review details.
 
 **Inspection only** means the writer asked for comments, not a rewrite. Point
 to the unchanged text, quote each problem, suggest a repair, and include the
-writing score and phrase-by-phrase guide. Do not invent an “after” result.
+writing score when measured and phrase-by-phrase guide. Do not invent an “after” result.
 **When Zero Slop is part of another task,** run every required check but return
 only the finished text unless the user asks for review details. These choices
 change only what the writer sees. Zero Slop must still complete the local
@@ -848,22 +875,30 @@ needing a real fact from the user. Never silently overwrite; the author decides.
 ### 10. Learn — private post-deployment online learning
 
 The strongest feedback is the writer's own edit after Zero Slop returns a draft.
-This is post-deployment, human-in-the-loop online learning: the detector updates
+This is opt-in, post-deployment, human-in-the-loop online learning: the detector updates
 external, interpretable rules from later edits. It is not RLHF and does not retrain
 the AI model already running in the assistant or rewrite this `SKILL.md`.
 
-- **The reflect loop.** Whenever you can see both what the skill produced and
-  what the author actually shipped — they paste the final version, they say
-  "I cut X", you edit a file they later revise — record it:
+Do not run `--reflect`, `--auto-apply`, or any other private-state write merely
+because a later edit is visible. Ask for and receive the writer's affirmative
+opt-in before recording their edit; ask separately before activating learned
+patterns with `--auto-apply`. If consent is absent, skip learning and continue
+the editorial task. An opt-in to edit the current draft is not consent to store
+the draft or future edits in a private overlay.
+
+- **The reflect loop.** With explicit opt-in, when you can see both what the skill
+  produced and what the author actually shipped — they paste the final version,
+  they say "I cut X", you edit a file they later revise — record it:
 
   ```
-  python3 scripts/learn.py --reflect --produced out.md --shipped final.md \
-    --reason <reason> --genre <genre> --auto-apply
+  python3 <skill-root>/scripts/learn.py --reflect --produced out.md --shipped final.md \
+    --reason <reason> --genre <genre>
   ```
 
   Reflection records evidence immediately. A span becomes eligible only after
   the same cut appears across three content-distinct edit pairs; a single word
-  needs five. `--auto-apply` activates eligible evidence only after the novelty
+  needs five. Only with separate affirmative opt-in, add `--auto-apply` to that
+  command; it activates eligible evidence only after the novelty
   and human-corpus safety gates pass. The result goes to the private live overlay
   at `~/.zero-slop/learned.json`, which the scorer reloads on its next run. It
   does not edit the installed or shared taxonomy. When the writer repeatedly
@@ -888,7 +923,8 @@ the AI model already running in the assistant or rewrite this `SKILL.md`.
   Learning that corrupts the meter is worse than not learning.
 
 - **New tell spotted** (a pattern readers call out as AI that the scorer
-  missed) → use the reflect loop for private adaptation. When the catch comes
+  missed) → offer the reflect loop for private adaptation, subject to opt-in.
+  When the catch comes
   from an audit, a competing skill, or a reviewer rather than the meter, the
   ratchet applies: it becomes a deterministic detector or a
   `data/corpus/must-flag/` fixture in the same change, and
@@ -909,7 +945,7 @@ the AI model already running in the assistant or rewrite this `SKILL.md`.
   build a private scoring profile from a sample of their real writing:
 
   ```
-  python3 scripts/learn.py --voice <name> --from <their-writing>
+  python3 <skill-root>/scripts/learn.py --voice <name> --from <their-writing>
   ```
 
   The builder scans `.md` and `.txt` files for existing lexicon and
@@ -925,7 +961,7 @@ the AI model already running in the assistant or rewrite this `SKILL.md`.
   than guessing new weights, derive them:
 
   ```
-  python3 scripts/calibrate.py --human <dir> --ai <dir>
+  python3 <skill-root>/scripts/calibrate.py --human <dir> --ai <dir>
   ```
 
   This computes each term's excess frequency in current AI output against
@@ -953,7 +989,7 @@ the AI model already running in the assistant or rewrite this `SKILL.md`.
 - **Every change is gated.** After editing patterns or weights, run
 
   ```
-  python3 scripts/calibrate.py --selftest
+  python3 <skill-root>/scripts/calibrate.py --selftest
   ```
 
   which scores a corpus of writing that must never be flagged

package/docs/cli.md

@@ -3,7 +3,7 @@
 Use `deslop` for hosted editing with MCP parity. Use `score` for offline checks.
 
 ```sh
-npm install --global zero-slop@2.12.6
+npm install --global zero-slop@2.12.8
 zero-slop deslop draft.md --genre professional
 zero-slop deslop - --genre email < draft.txt
 zero-slop deslop draft.md --json --require-approved

package/gemini-extension.json

@@ -1,6 +1,6 @@
 {
   "name": "zero-slop",
-  "version": "2.12.6",
+  "version": "2.12.8",
   "description": "Find and fix canned AI writing while preserving the facts, voice, and format.",
   "mcpServers": {
     "zero-slop": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "zero-slop",
-  "version": "2.12.6",
+  "version": "2.12.8",
   "description": "An editing skill, offline writing checks, and a CLI for the hosted Zero Slop MCP pipeline. Find stock AI phrasing, rewrite, and check source details.",
   "bin": {
     "zero-slop": "bin/zero-slop.mjs"

package/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "zero-slop",
-  "version": "2.12.6",
+  "version": "2.12.8",
   "description": "Find and fix canned AI writing while preserving the facts, voice, and format.",
   "author": {
     "name": "Manav Mishra",

package/references/eval.md

@@ -1,7 +1,10 @@
 # Zero Slop eval
 
-Answer every check with pass or fail. Where a check asks for a count, write the
-number down; a count is evidence, and a missing count means the pass did not run.
+Answer every check with pass or fail. Use "unavailable" only for a check that
+requires a missing script in the single-file or no-Python path; name that gap in
+the report and never call the result fully verified. Where a check asks for a
+count, write the number down; a count is evidence, and a missing count means the
+pass did not run.
 
 Any fail permits one targeted textual repair followed by one local recheck. It does
 not restart the copy desk, read-aloud pass, or model request. A failed check changes
@@ -215,6 +218,9 @@ the report even when they are zero.
 ## C. Fidelity
 
 48. **Scripted check run, not eyeballed.** `slopscore.py --fidelity` exits zero.
+    If scripts are unavailable, mark this unavailable and compare facts, claims,
+    figures, and qualifiers directly against the source. The direct comparison
+    does not turn an unavailable scripted check into a pass.
 49. **No invented specifics.** No number, name, anecdote, date, or source appeared
     that the author did not supply.
 50. **No invented interior claims.** No stated feeling, motive, or experience the
@@ -309,13 +315,18 @@ local verifier records the execution facts and the report names the consolidatio
 
 73. **Execution described accurately.** Separate passes are named as separate; a
     one-response consolidation is named as one response, never as independent review.
-74. **No self-certification of facts.** The local source gate checks the generated
-    text even when the model performs an editorial self-check inside one response.
+74. **No self-certification of facts.** The installed local source gate checks the
+    generated text even when the model performs an editorial self-check inside one
+    response. In scriptless mode, compare directly with the source and state that
+    the independent scripted gate was unavailable.
 75. **Counts reported.** Every count in section A appears in the summary, including
     the zeros.
-76. **The exact final text cleared every check.** Not an earlier draft, not a version
-    that was repaired afterward.
-77. **Finalizer edits were rechecked locally.** If role 8 changed anything, the score,
-    fact, format, and structure checks ran once on that exact revision.
+76. **The exact final text cleared every available check.** Not an earlier draft,
+    not a version that was repaired afterward. Name any unavailable scripted check;
+    it prevents a claim of full verification.
+77. **Finalizer edits were rechecked.** If role 8 changed anything, the available
+    score and fact checks plus format and structure checks ran once on that exact
+    revision. In scriptless mode, repeat the direct source comparison and name the
+    unavailable scripted checks.
 78. **Fallbacks named honestly.** The report names an unavailable role, local fallback,
     missed target, or one-request constraint and does not call it fully verified.

package/references/fresh-eyes.md

@@ -52,7 +52,9 @@ and name the ambiguity.
 
 Return `approve without changes`, `changed`, or `unresolved — needs the writer`. If
 the finalizer changes the text, apply the complete revision and rerun the local score,
-fact, format, and structure checks once. Do not restart the AI pipeline. If the local
+fact, format, and structure checks once when scripts are available; otherwise
+repeat the manual source comparison and format/structure checks once and name the
+scripted checks as unavailable. Do not restart the AI pipeline. If the local
 recheck fails, return the safest source-preserving edit, identify the unresolved span,
 and do not describe it as fully verified. An unavailable finalizer is an abstention;
 it never starts another model request.

package/references/readalong.md

@@ -114,11 +114,14 @@ Apply the returned artifact to the actual deliverable before verification.
 
 Verify the exact artifact returned by the read-aloud editor:
 
-1. Rerun the heuristic surface scorer and scripted fidelity check.
+1. Rerun the heuristic surface scorer and scripted fidelity check when the
+   installed scripts are available. In the single-file or no-Python path,
+   compare the exact text directly with the source and mark both scripted checks
+   unavailable; do not claim a numeric score or full verification.
 2. Compare it directly with the original and selected rewrite for claims,
    qualifiers, intended voice, regional spelling, format, and non-prose structure.
-3. If a check requires a textual repair, apply one targeted correction and rerun the
-   local score, fact, format, and structure checks once.
+3. If a check requires a textual repair, apply one targeted correction and rerun
+   the available local checks and direct fact, format, and structure comparison once.
 
 Do not restart the copy desk or read-aloud pass. If an issue still cannot be resolved
 without guessing, return the best source-preserving version, state the unresolved span

package/scripts/rescue.py

@@ -56,6 +56,10 @@ def rescue_text(text: str) -> str:
         r"[ \t]+([A-Za-z][A-Za-z0-9_-]*)",
         _without_wrapper, out, flags=re.IGNORECASE,
     )
+    out = re.sub(
+        r"(^|[.!?][ \t\r\n]+|\r?\n[ \t]*\r?\n[ \t]*)as we move forward,[ \t]+the team will\b",
+        lambda m: m.group(1) + "The team will", out, flags=re.IGNORECASE,
+    )
     rules: list[tuple[str, str | object]] = [
         (
             r"\bwe are thrilled to unveil ([^,\n]+),\s+a transformative release that "
@@ -70,6 +74,9 @@ def rescue_text(text: str) -> str:
         (r"\bour cutting[-\u2010\u2011 ]edge\b", "Our"),
         (r"\bhours of tedious manual configuration\b", "hours of manual configuration"),
         (r"\bwe have completely reimagined\b", "We rebuilt"),
+        (r"\bthis represents a significant milestone in our journey\.\s*", ""),
+        (r"\bwill leverage the proposal to make a decision\b", "will use the proposal to decide"),
+        (r"\ba groundbreaking update\b", "an update"),
         (r"\bwith robust error handling built in from the ground up\b",
          "with built-in error handling"),
         (r"\bwe believe these improvements will fundamentally transform how your team "