zero-slop 2.5.8 → 2.6.1

package/README.md

@@ -5,29 +5,24 @@
   <img alt="tests" src="https://img.shields.io/badge/tests-passing-227B5B">
   <img alt="dependencies" src="https://img.shields.io/badge/runtime%20dependencies-0-227B5B">
   <img alt="privacy" src="https://img.shields.io/badge/learning-private-227B5B">
-  <img alt="version" src="https://img.shields.io/badge/version-2.5.8-72528F">
-</p>
-
-<p align="center">
-  <strong><a href="https://zero-slop.ai">zero-slop.ai</a></strong> &middot; examples, benchmarks, and blog
+  <img alt="version" src="https://img.shields.io/badge/version-2.6.1-72528F">
 </p>
 
 **Less slop, more pop.**
 
-Zero Slop is an Agent Skill, not an AI model. Claude, GPT, or another compatible model
-edits inside your assistant; Zero Slop supplies local checks that protect meaning,
-voice, and format.
+Zero Slop is an Agent Skill, not an AI model. Claude, GPT, or another compatible
+model edits the draft; Zero Slop supplies the workflow and local source checks.
 
-The 0-to-100 writing score flags generic AI-style language and lists flagged phrases;
-it does not identify the author. In the test sets, human samples
-scored 9 to 21; unedited AI drafts averaged 77. These are reference points, not
+Its 0-to-100 writing score points to flagged phrases, flat rhythm, dense passages,
+and distracting formatting, not authorship. In the references,
+human writing scored from 9 to 21; unedited AI drafts averaged 77. These are not
 universal cutoffs.
 
 ![A scored sentence before and after editing](assets/demo.png)
 
 ## Install Zero Slop
 
-Paste this into Claude Code, Codex, Cursor, OpenCode, Warp, Zed, or another
+Paste this prompt into Claude Code, Codex, Cursor, OpenCode, Warp, Zed, or another
 Agent Skills-compatible assistant:
 
 ```text
@@ -36,13 +31,13 @@ this agent.
 
 1. Find active installations. Report each path, version, and method. Do not duplicate
    or remove one without asking.
-2. For updates, keep the existing install method. In Codex, use $skill-installer for
+2. Keep the current install method when updating. In Codex, use $skill-installer for
    `skills/zero-slop`; in Claude Code or Cowork, use the plugin marketplace. Otherwise
-   use `npx skills add manavmishra/ZeroSlop --global` to install or
-   `npx skills update zero-slop --global` to update.
+   use `npx skills add manavmishra/ZeroSlop --global` for a first install or
+   `npx skills update zero-slop --global` for an existing CLI install.
 3. Preserve ZERO_SLOP_HOME (default: ~/.zero-slop) and its private data.
-4. Verify the version. When Python is available, run
-   `python3 scripts/calibrate.py --selftest`.
+4. Verify the installed version. When Python is available, run
+   `python3 scripts/calibrate.py --selftest` from the installed skill directory.
 5. Report the path, method, version, validation result, and restart requirement.
 
 Do not modify the current project or unrelated configuration. Ask before falling
@@ -55,139 +50,126 @@ Direct terminal install:
 npx skills add manavmishra/ZeroSlop --global
 ```
 
-Claude Code and Cowork use the plugin marketplace. ChatGPT users can
-download [`dist/zero-slop-single-file.md`](dist/zero-slop-single-file.md); Claude.ai
-users can download the release ZIP.
+ChatGPT users can download [`dist/zero-slop-single-file.md`](dist/zero-slop-single-file.md);
+Claude.ai users can use the ZIP.
 
-## How Zero Slop works
+## How it works
 
-![One editing workflow, a private learning loop, and a separate release review](assets/engine.svg)
+![Eight editorial roles, a private learning loop, and a separate release review](assets/engine.svg)
 
-Seven roles form one workflow — jobs, not separate models. Your AI assistant handles
-the editorial passes; local Python tools run repeatable checks.
+Eight roles form one workflow. They are jobs, not separate models. Your AI assistant
+handles editorial judgment; local Python tools run repeatable checks.
 
-| Role | Who does it | What it does |
+| Role | Who does it | What happens |
 |---|---|---|
-| 1. Scorer | Local tools | Finds exact phrases, mechanical rhythm, dense passages, and distracting formatting; explains the writing score. |
-| 2. Interpreter | Your AI assistant | Reads claims, support, audience, structure, and voice before editing. |
-| 3. Rewriter | Your AI assistant | Removes stock wording and improves order, rhythm, and tone without inventing detail. |
-| 4. Fact gate | Local tools | Rejects versions that add or drop names, numbers, quotations, or links; selects the clearest one left. |
-| 5. Copy desk | Fresh AI pass | Corrects grammar, spelling, usage, and consistency. |
+| 1. Scorer | Local tools | Finds the exact wording, rhythm, readability, and formatting problems that raised the writing score. |
+| 2. Interpreter | Your AI assistant | Reads the claims, purpose, audience, structure, and voice before changing anything. |
+| 3. Rewriter | Your AI assistant | Removes stock language and rebuilds order, rhythm, and tone without inventing detail. |
+| 4. Fact gate | Local tools | Rejects a version that changes names, numbers, quotations, links, code, tables, paths, or document structure. |
+| 5. Copy desk | Fresh AI pass | Corrects grammar, spelling, usage, and consistency in the actual deliverable. |
 | 6. Read-aloud editor | Fresh AI pass | Fixes stumbles, repetition, weak transitions, and awkward flow. |
-| 7. Verifier | Local tools and your AI assistant | Compares the final text with the source for facts, meaning, voice, format, and structure. A repair repeats both editorial passes and every check. |
-
-### Why separate the work?
+| 7. Verifier | Local tools and your AI assistant | Compares the text with the source for facts, meaning, qualifiers, voice, format, and structure. |
+| 8. Fresh-eyes finalizer | New AI pass | Reads the verified text as a first-time reader and applies only safe last-mile polish. Any final polish restarts the final checks; the same text must return unchanged before release. |
 
-Research supports the checks, not the number seven. Studies find
+Research supports the checks, not the number eight. Studies find
 [predictable wording](https://arxiv.org/abs/2301.11305) and
-[excess vocabulary](https://arxiv.org/abs/2406.07016) in machine text, while detectors
-can [misclassify non-native English](https://arxiv.org/abs/2304.02819). Writing and
-verification stay separate: the rewriter does not certify its own facts, and later
-passes catch surviving errors. Local tools use Python's standard library and never send
-drafts. The
-[research notes](references/evidence.md) cover the rationale and limits.
-
-## Private learning from writer edits
-
-Learning requires the assistant's version and the writer's final version.
-Zero Slop never monitors files, browsers, or publishing systems.
-
-A named profile can exempt existing watchlist words in your sample. It applies only
-when selected by name and does not learn cadence, tone, or a complete writing style.
+[overused vocabulary](https://arxiv.org/abs/2406.07016) in machine text, while
+authorship detectors can [misclassify non-native English](https://arxiv.org/abs/2304.02819).
+Zero Slop combines language checks with contextual editing. Any finalizer change
+repeats the copy desk, read-aloud pass, verification, and fresh-eyes review. Local
+tools use only Python's standard library.
 
-A phrase must be cut from three unrelated pieces before becoming a private rule; a
-single word needs five. New rules must keep known-human samples clean. Repeated fixes
-guide later edits; kept phrases can quiet a rule, and old rules fade. Private state
-under `$ZERO_SLOP_HOME` never retrains the AI model.
+## Private learning from your edits
 
-## What the current release measured
+Learning starts only when you provide both the assistant's version and the version
+you kept. Zero Slop never watches files, browsers, or publishing systems.
 
-v2.5.8 keeps v2.5.7's context-first review and adds a check for generic lists of
-benefits. On the frozen 38-passage consensus panel, accuracy rose from 71.1% to 84.2%.
-The new check caught five passages the previous version missed. It did not change any
-passage the raters considered clean. Because two LLM editorial raters supplied the
-labels, this is a regression result, not human field accuracy.
+A phrase needs removal from three unrelated pieces before it becomes a private rule;
+a single word needs five. Each must be new and safe on known-human text. Repeated
+fixes guide later edits; kept phrases can quiet a rule. Private data lives under
+`$ZERO_SLOP_HOME`. This is human-in-the-loop online learning, not neural training or
+RLHF, and it never retrains Claude, GPT, or another model.
 
-### Recent model output: RAID+
+A profile can exempt existing watchlist words when selected by name. It does not
+learn cadence, tone, or a complete writing style.
 
-[RAID+](https://huggingface.co/datasets/markstanl/RAID-Plus) extends the peer-reviewed,
-MIT-licensed RAID benchmark. We scored 8,000 pinned rows; 7,627 abstracts
-remained after excluding failed or empty generations.
-
-| Model | Texts scored | Mean writing score ↓ | At or above 25 |
-|---|---:|---:|---:|
-| DeepSeek V3 | 1,995 | 14.5 | 10.1% |
-| Gemini 3.1 Pro | 1,998 | 17.0 | 18.2% |
-| Gemma 3 27B | 1,634 | 21.6 | 30.4% |
-| Llama 3.3 70B | 2,000 | 25.5 | 41.7% |
-| **Overall** | **7,627** | **19.6** | **24.8%** |
+## What v2.6.1 changes
 
-![Current Zero Slop writing scores across four RAID+ model families](assets/bench-raid-plus.png)
+Four narrow checks now catch reasoning artifacts, unsupported novelty, emotional
+flatness, and repetitive acknowledgments that scored clear in v2.5.10. All 114 frozen
+document scores stayed unchanged, all 18 human controls stayed clear, and the four
+new cases moved from 9.5 to 30.7–65.1. Median throughput was 0.03% lower in the
+12-run comparison, a small local timing difference rather than a speed claim.
 
-RAID+ records model origin, not editorial quality, so this is a score distribution,
-not an accuracy claim. In a fresh pass over 2,187 Beemo records, raw model responses
-averaged 30.2, expert edits 25.3, and independent human answers 20.0. Expert editing
-lowered the score in 52.2% of pairs; Beemo lacks quality labels.
+The fresh-eyes role changes editing, not the local meter. It reads the verified copy
+as a first-time reader; any correction restarts every final check. v2.6.1 also
+matched the prior 84.2% result on the 38-item editorial panel.
 
-### Five workflows, one fixed set of drafts
+### Fresh same-model editing replay
 
-The 18 saved rewrites use Zero Slop 2.4.3 and pinned competitor instructions. They were
-not regenerated. Passing requires clean writing and layout without altered facts or
-invented feelings.
+We reran Zero Slop, [avoid-ai-writing](https://github.com/conorbronsdon/avoid-ai-writing),
+[no-ai-slop](https://github.com/petergyang/no-ai-slop), and
+[humanizer](https://github.com/blader/humanizer) from scratch on the same 18 obvious
+drafts. Every workflow used GPT-5.4, high reasoning, batches of three, and its pinned
+instruction file.
 
-| Method | Mean writing score ↓ | Passed all checks | Important details kept | Average length change |
+| Method | Mean writing score ↓ | Passed all Zero Slop checks | Important details kept | Average length change |
 |---|---:|---:|---:|---:|
 | Original drafts | 76.3 | 0/18 | — | — |
-| Zero Slop | 16.4 | 18/18 | 18/18 | -26.4% |
-| humanizer | 25.3 | 13/18 | 18/18 | -25.7% |
-| stop-slop | 25.7 | 13/18 | 18/18 | -34.4% |
-| no-ai-slop | 29.1 | 12/18 | 18/18 | -28.0% |
-| de-slop | 52.3 | 6/18 | 18/18 | -18.5% |
+| Zero Slop | 12.8 | 18/18 | 18/18 | -8.9% |
+| avoid-ai-writing | 23.3 | 15/18 | 18/18 | -14.6% |
+| no-ai-slop | 28.4 | 12/18 | 17/18 | -13.7% |
+| humanizer | 35.4 | 9/18 | 17/18 | -7.2% |
 
-![The same 18 drafts after each editing workflow; lower scores mean fewer generic AI-style patterns](assets/bench-search-rewrites.png)
+![Fresh same-model editing replay on 18 drafts](assets/bench-search-rewrites.png)
 
-All 18 Zero Slop edits passed, but Zero Slop defines the rules and the rewrites predate
-this release. No available set combines current writing with independent human review,
-so no universal accuracy number exists. See
-[`bench/README.md`](bench/README.md).
+Those are Zero Slop's checks, so we also ran a two-pass, method-hidden comparison
+against the pinned incumbent. The GPT-5.4 reviewer favored Zero Slop on 13 drafts
+and avoid-ai-writing on 3; 2 were unresolved. The passes agreed on 16 of 18 winners.
+Zero Slop's source check passed 18/18 of its rewrites and 16/18 incumbent rewrites.
+Its mean writing score was 17.8 versus 17.0, so it did not win every measure.
 
-### The local tools are fast
+![Method-hidden editorial preference on 18 drafts](assets/bench-incumbent-hidden.png)
 
-On one Apple silicon Mac, the local checker processed 1,000 documents in 2.0147 seconds
-(496.4 per second). A 15,201-word document took 0.3222 seconds; the slowest stress case
-took 2.5180 seconds. An alternating six-pair comparison measured 29.4% higher median
-throughput than v2.5.7. The optimization produced the same complete result as v2.5.7
-on all 280 tracked documents. An 8,000-word learning pass took 0.1589 seconds. Time
-spent by the AI assistant is excluded.
+This is a small LLM-reviewed regression study, not independent human field accuracy
+or a universal ranking. The packets, mappings, judgments, hashes, and limits are in
+[`bench/incumbent-blind-replay/`](bench/incumbent-blind-replay/).
 
-## What Zero Slop adds
+### Current-model and speed checks
 
-Zero Slop builds on [no-ai-slop](https://github.com/petergyang/no-ai-slop),
-[humanizer](https://github.com/blader/humanizer),
-[de-slop](https://github.com/isatimur/de-slop), and
-[stop-slop](https://github.com/hardikpandya/stop-slop), and draws on
-[unslop-text](https://github.com/JCarterJohnson/vibecoded-design-tells/tree/main/unslop-ai-text)
-research. It adds a local writing score,
-fact protection, separate editorial passes, private learning, and release tests.
+The pinned [RAID+](https://huggingface.co/datasets/markstanl/RAID-Plus) sample yielded
+7,627 usable generations:
 
-The chart records documented features. It does not decide which tool writes better.
-Details are in [`bench/README.md`](bench/README.md).
+| Model | Texts scored | Mean writing score ↓ | At or above 25 |
+|---|---:|---:|---:|
+| DeepSeek V3 | 1,995 | 14.5 | 10.1% |
+| Gemini 3.1 Pro | 1,998 | 17.0 | 18.2% |
+| Gemma 3 27B | 1,634 | 21.6 | 30.4% |
+| Llama 3.3 70B | 2,000 | 25.5 | 41.7% |
 
-![Comparison of features documented in pinned repository versions](assets/competitor-capabilities.png)
+RAID+ records model origin, not editorial quality. In Beemo, raw model responses
+averaged 30.2, expert edits 25.3, and human answers 20.0; neither dataset has quality
+labels.
 
-## Reproduce the tests and benchmarks
+On one Apple silicon Mac, the local checker processed 1,000 documents in 1.9958
+seconds, or 501.1 per second. A 15,201-word document took 0.3223 seconds; the slowest
+stress case took 2.4438 seconds. An 8,000-word learning pass took 0.1627 seconds.
+These measurements exclude the AI assistant's editing time and are not service-level
+guarantees.
 
-```bash
-python3 tests/test_all.py
-python3 scripts/calibrate.py --selftest
-python3 bench/search-corpus/compare.py --check
-python3 bench/raid-plus-corpus/audit.py --check
-python3 bench/beemo-corpus/audit.py --check
-python3 bench/validate_corpus_registry.py
-python3 bench/make_charts.py --check
-```
+## What Zero Slop adds
+
+Zero Slop builds on no-ai-slop, humanizer, de-slop, stop-slop, unslop-text, and
+avoid-ai-writing. It adds a writing score, source protection, separate editorial
+passes, private learning, portfolio analysis, and release tests. The chart records
+documented features, not which tool writes better.
+
+![Documented capabilities at pinned repository versions](assets/competitor-capabilities.png)
 
-`SKILL.md` defines the skill; `scripts/` contains tools; `bench/` contains tests. See
-the [security policy](SECURITY.md) and [research notes](references/evidence.md).
+Reproduce the shipped checks with `python3 tests/test_all.py`,
+`python3 scripts/calibrate.py --selftest`, and `python3 bench/make_charts.py --check`.
+The benchmark registry, source pins, scripts, hashes, and limitations are documented
+in [`bench/README.md`](bench/README.md).
 
-Released under the [MIT License](LICENSE).
+Released under the [MIT License](LICENSE), which permits commercial and private use,
+modification, and redistribution under its terms.