zero-slop 2.8.6 → 2.8.8

package/README.md

@@ -5,7 +5,7 @@
   <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-2.8.6-72528F">
+  <img alt="version" src="https://img.shields.io/badge/version-2.8.8-72528F">
   <a href="https://hol.org/guard/plugins"><img alt="Listed in the HOL plugin registry" src="https://img.shields.io/badge/HOL%20registry-listed-2C6E8F"></a>
   <a href="https://github.com/hashgraph-online/awesome-ai-plugins#tools--integrations"><img alt="Listed in awesome-ai-plugins" src="https://img.shields.io/badge/awesome--ai--plugins-listed-2C6E8F"></a>
   <a href="https://zero-slop.ai/try/"><img alt="This README scores 11.6 out of 100 on the slop score, inside the human range" src="https://img.shields.io/badge/slop%20score-11.6%2F100-0f7d55"></a>
@@ -14,8 +14,7 @@
 Less slop, more pop in your writing.
 
 Zero Slop finds stock phrasing, mechanical rhythm, vague claims, and canned
-formatting, then gives your AI assistant or agent harness an editing workflow on guardrails. Its
-MIT-licensed local checks run offline; your existing AI assistant or harness does the editing.
+formatting. Local checks run offline; your existing AI assistant does the editing.
 Score a draft at [zero-slop.ai/try](https://zero-slop.ai/try/); benchmark at
 [zero-slop.ai](https://zero-slop.ai).
 
@@ -63,12 +62,12 @@ Writing score: 9.5/100  [clear]
 ## Problem
 
 AI-assisted writing often converges on the same constructions: "It's not X. It's Y."
-"Here's the thing nobody tells you." One emdash is fine. Multiple emdashes, definitely slop. The same effect can come from repetition in the wording or the structure. Mechanical rhythm and overworked formatting can do it too.
+"Here's the thing nobody tells you." Repeated wording, uniform rhythm, and
+overworked formatting can have the same effect.
 
 Zero Slop is an Agent Skill and ships no model. Claude, GPT, or another compatible
-model edits; local tools check names, numbers, quotations, links, code, tables, and
-paths. Your AI compares meaning because matching words cannot catch every changed
-claim. It runs in any harness that reads `SKILL.md`.
+model edits; local tools protect names, numbers, quotations, links, code, tables,
+and paths. Your AI compares meaning that word matching cannot judge.
 
 ## How to install Zero Slop
 
@@ -84,10 +83,7 @@ Or install it with `npx`:
 npx skills add manavmishra/ZeroSlop --global
 ```
 
-Or from the registry, which also installs the scorer as a command. Zero Slop is
-listed in [awesome-ai-plugins](https://github.com/hashgraph-online/awesome-ai-plugins#tools--integrations)
-and carries a public profile in the [HOL plugin registry](https://hol.org/guard/plugins),
-where its trust score and scanner results are published:
+Or install from npm, which also adds the scoring command:
 
 ```sh
 npx zero-slop install          # add --harness codex|cursor|opencode|zed
@@ -95,8 +91,7 @@ npx zero-slop score draft.md   # score without installing anything
 ```
 
 ChatGPT users can download [`dist/zero-slop-single-file.md`](dist/zero-slop-single-file.md).
-Claude.ai users can upload the [latest release ZIP](https://github.com/manavmishra/ZeroSlop/releases/latest/download/zero-slop.zip).
-`npx skills update zero-slop --global` updates a skills CLI installation later.
+Claude.ai users can upload the [release ZIP](https://github.com/manavmishra/ZeroSlop/releases/latest/download/zero-slop.zip).
 
 ## How to use Zero Slop
 
@@ -135,9 +130,8 @@ above.
 9. Synonym cycling: the agent, the assistant, the tool, all one thing.
 10. Marketing riders: "robust" and "leverage" score only beside a marketing trigger, so a runbook stays quiet.
 
-A reading pass covers document-wide problems: repeated shapes, crowded statistics,
-and paragraphs that shuffle without loss. [`references/eval.md`](references/eval.md)
-has all 80 checks.
+A reading pass covers repeated shapes, crowded statistics, and paragraphs that
+shuffle without loss. [`references/eval.md`](references/eval.md) has all 80 checks.
 
 Human writing scored 9 to 21 in [`data/corpus/must-not-flag/`](data/corpus/must-not-flag/);
 unedited AI drafts averaged 77 across [`bench/examples.json`](bench/examples.json).
@@ -147,7 +141,13 @@ unedited AI drafts averaged 77 across [`bench/examples.json`](bench/examples.jso
 
 ![Eight editorial roles, a private learning loop, and a separate release review](assets/engine.svg)
 
-Eight roles form one workflow. Each is a job rather than a service, run as its own pass so nothing grades its own output. The research supports the checks, not the number eight, which is an engineering choice.
+Eight responsibilities form one workflow. They are jobs, not separate models. An
+installed assistant can isolate them.
+The free web editor combines
+the five AI responsibilities into one model response, then runs the scorer and source
+checks locally. It makes at most one live model call and never describes that response
+as independent review. The research supports the checks, not the number eight, which
+is an engineering choice.
 
 | Role | Who does it | What happens |
 |---|---|---|
@@ -155,10 +155,15 @@ Eight roles form one workflow. Each is a job rather than a service, run as its o
 | 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 any version that changes names, numbers, quotations, links, code, tables, paths, or 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 text with source for facts, meaning, qualifiers, voice, format, structure. |
-| 8. Fresh-eyes finalizer | Fresh AI pass | Reads the verified text as a first-time reader, applying only safe polish. Any final polish restarts the final checks; the same text must return unchanged before release. |
+| 5. Copy desk | Your AI assistant | Corrects grammar, spelling, usage, and consistency in the actual deliverable. |
+| 6. Read-aloud editor | Your AI assistant | Fixes stumbles, repetition, weak transitions, and awkward flow. |
+| 7. Verifier | Local tools and your AI assistant | Compares text with source for facts, meaning, qualifiers, voice, format, structure. In the web editor, source safety is checked locally. |
+| 8. Fresh-eyes finalizer | Your AI assistant | Reads the finished text as a first-time reader and applies only safe polish. Any change receives one final local recheck. |
+
+Missed editorial targets do not erase a useful edit or start an open-ended loop. Zero
+Slop permits one targeted repair and one local recheck, then returns the safest
+source-preserving edit with a plain warning. On the website, a failed model call falls
+back to a conservative local edit instead of retrying through more providers.
 
 Studies find
 [predictable wording](https://arxiv.org/abs/2301.11305) and
@@ -193,7 +198,7 @@ The saved replay ran Zero Slop, [avoid-ai-writing](https://github.com/conorbrons
 [no-ai-slop](https://github.com/petergyang/no-ai-slop) and
 [humanizer](https://github.com/blader/humanizer) over 18 drafts with GPT-5.4, high
 reasoning, and pinned instructions. Zero Slop's outputs came from v2.5.9; later
-releases rescore those frozen outputs rather than pretend to regenerate them.
+releases only rescore those frozen outputs.
 
 | Method | Mean writing score ↓ | Passed Zero Slop's local gates | Source check passed | Average length change |
 |---|---:|---:|---:|---:|
@@ -209,16 +214,15 @@ releases rescore those frozen outputs rather than pretend to regenerate them.
 
 ![Method-hidden editorial preference on 18 drafts](assets/bench-incumbent-hidden.png)
 
-Cross-checks the tools didn't build: the AIStoryHub checker's clean
-rates, and a method-hidden quality ranking.
+Two cross-checks: AIStoryHub clean rates and a method-hidden quality ranking.
 
 ![External checker clean rates per method](assets/bench-external-checker.png)
 
 ![Method-hidden quality ranking, lower is better](assets/bench-blind-quality.png)
 
-This is a small LLM-reviewed regression study. It measures neither field accuracy nor a
-universal ranking. Drafts, hashes, and limits are in [`bench/README.md`](bench/README.md). The separate method-hidden two-way replay used
-Zero Slop v2.6.0 and is preserved in
+This small LLM-reviewed regression study measures neither field accuracy nor a
+universal ranking. Drafts, hashes, and limits are in [`bench/README.md`](bench/README.md).
+The separate two-way replay used Zero Slop v2.6.0 and is preserved in
 [`bench/incumbent-blind-replay/`](bench/incumbent-blind-replay/).
 
 For the 38-item editorial panel, the current scorer matched the prior 84.2% result.
@@ -228,13 +232,11 @@ proof of general accuracy.
 
 ### Speed
 
-One busy Apple silicon Mac. Meter: 1,000 documents in 1.9929 seconds (501.8 per
-second), 15,201 words in 0.3223 seconds, worst stress case 2.2932 seconds.
-Reading pass: 0.7823 seconds for the same 1,000 (1278.3 per
-second), 0.1092 for the same large document, linear to 96,000 words. Learning
-pass, 8,000 words: 0.1592 seconds. Across 24 interleaved runs against 2.7.7 we measured
-0.00% higher median throughput—effectively no difference. Editing
-time is excluded.
+One Apple silicon Mac: the meter scored 1,000 documents in 1.9929 seconds
+(501.8 per second), 15,201 words in 0.3223 seconds, and the worst stress case in 2.2932
+seconds. The reading pass handled the same 1,000 in 0.7823 seconds (1278.3/second);
+learning on 8,000 words took 0.1592 seconds. Across 24 interleaved runs against 2.7.7,
+median throughput was 0.26% lower, effectively no difference. Editing is excluded.
 
 ### Reading-pass accuracy
 
@@ -282,7 +284,7 @@ passes, private learning, portfolio analysis, and release tests.
 
 ![Documented capabilities at pinned repository versions](assets/competitor-capabilities.png)
 
-The chart records which features each project documents. It says nothing about writing quality and is not a claim about which tool writes better. Reproduce by using these tests:
+The chart says nothing about writing quality. Reproduce it with:
 
 ```sh
 python3 tests/test_all.py

package/SKILL.md

@@ -2,7 +2,7 @@
 name: zero-slop
 license: MIT
 metadata:
-  version: "2.8.6"
+  version: "2.8.8"
   author: manavmishra
 description: Turn drafts into sharp, natural prose or inspect them without rewriting. Zero Slop runs inside the user's existing AI assistant; Claude, GPT, or another compatible model reads and edits in context while local tools point to exact phrases and protect the source. Use when the user asks to humanize or de-slop writing, inspect AI-sounding patterns, fix text that reads like ChatGPT, polish outward-facing prose, draft social or LinkedIn content, or apply a final quality check to prose the agent generated. The workflow preserves facts, voice, and format and learns privately from repeated, reason-labelled human edits.
 ---
@@ -75,10 +75,12 @@ citations, and the ladder below orders the signals by measured strength.
 
 ## Eight roles, one pipeline
 
-Run the rewrite workflow as eight ordered roles. They are separate jobs, not eight
-models or services. The same Claude, GPT, or other compatible model in the user's AI
-assistant may perform every editorial role, but each must be a separate pass. Keep
-local and AI responsibilities distinct:
+Run the rewrite workflow as eight ordered responsibilities. They are editorial jobs,
+not eight models or services. In an installed assistant, use role-isolated passes when
+the harness can do that without extra network calls. When a service has a one-request
+budget, combine the AI responsibilities into one structured editorial response and
+run the local checks before and after it. Name that consolidation honestly; one model
+response is not independent review. Keep local and AI responsibilities distinct:
 
 1. **Scorer — local tools.** Point to exact phrases and problems with rhythm,
    readability, formatting, and register; explain the writing score.
@@ -90,25 +92,28 @@ local and AI responsibilities distinct:
    quotations, or links; among the rest, select the version that best clears the
    measured checks. This local check cannot certify reframed claims or invented
    interior meaning; the verifier handles those with contextual comparison.
-5. **Copy desk — a fresh AI pass.** Correct grammar, spelling, punctuation, usage,
+5. **Copy desk — the AI assistant.** Correct grammar, spelling, punctuation, usage,
    diction, and consistency in the selected text.
-6. **Read-aloud editor — a fresh AI pass.** Read the complete copy-edited text aloud
+6. **Read-aloud editor — the AI assistant.** Read the complete copy-edited text aloud
    and directly fix stumbles, repetition, weak transitions, and awkward flow.
 7. **Verifier — local tools plus the AI assistant.** Check the exact final text
    against the source for the writing score, facts, meaning, qualifiers, voice,
-   format, and structure. This is a release gate, not advice: a warning blocks
-   delivery. Any repair returns through roles 5 and 6 before role 7 runs again.
-8. **Fresh-eyes finalizer — a new AI pass.** Read the verified text as a first-time
-   reader, apply only safe final polish, and approve it without changes. A role 8
-   edit restarts roles 5 through 8; the finalizer never bypasses verification.
+   format, and structure. A warning prevents an unqualified approval; it never erases
+   a source-safe edit or starts an open-ended loop. Apply at most one targeted repair,
+   then rerun the local checks on the exact changed text.
+8. **Fresh-eyes finalizer — the AI assistant.** Read the verified text as a first-time
+   reader and apply only safe final polish. If it changes the text, rerun the local
+   score and fact checks once. Deliver the safest edit with a plain warning if a
+   remaining concern would require another model request or a guess.
 
 This is an engineering separation of responsibilities, not a claim that research has
 proved eight to be the uniquely correct number. Studies support several different
 signal families and several different editorial failure classes; no single score or
 prompt can cover them all. The local roles provide repeatable measurements. The AI
-roles supply contextual judgment and editing. A generating role never certifies its
-own output. Role 7 verifies; role 8 confirms that the same verified text reads cleanly
-to someone seeing it for the first time.
+roles supply contextual judgment and editing. A generating role does not certify its
+own factual safety. Role 7 supplies the local release checks; role 8 confirms that the
+result reads cleanly to someone seeing it for the first time. In a one-request service,
+the model's self-check is editorial guidance, not independent verification.
 
 ## Detailed workflow
 
@@ -632,25 +637,29 @@ format, and non-prose structure. Apply these contextual checks too:
   read as prose. The whole document uses one coherent register, and every
   cross-reference resolves exactly.
 
-If verification requires any textual repair, send the repaired text through the copy
-desk and final read-aloud pass again, then repeat every final check. Continue until the
-same text clears the copy desk, final read-aloud pass, semantic and format review,
-scorer, and fidelity gate.
+If verification finds a concrete textual defect, apply one targeted repair. Run the
+local score, fact, format, and structure checks once more on that exact text. Do not
+restart the complete editorial sequence. If the second check still finds a problem,
+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. 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 request. Report any role that did not complete.
+
 A required repair may raise the writing score from the previous draft as long as it
 stays below the release limit. Source meaning, stated emotion, and factual accuracy
 outrank a smaller number. Never discard a necessary semantic repair merely because an
 unsafe version scored lower; rerun every check on the repaired text instead.
-Limit this repair loop to three rounds. If a problem still cannot be resolved
-without guessing, return the best source-preserving version that completed both
-editorial passes and state the unresolved issue and failed check plainly. Outside
-that explicit three-round fallback, nothing reaches the user until the exact text
-being returned has cleared every final check. A fallback still must have completed the
-copy desk and read-aloud pass; never describe it as fully verified.
-
-Initial gate failure → rewrite and recheck (max 3 passes). Final verification
-repair → copy desk, read-aloud pass, and every check again (max 3 rounds). If the
-initial gate still fails after three passes, keep the best version and flag it:
-"needs a real claim/detail, not better words."
+There is at most one targeted repair. If a problem still cannot be resolved without
+guessing, return the safest source-preserving edit and state the unresolved issue,
+unavailable role, or missed target plainly. Never replace an explicit rewrite request
+with the unchanged source merely because a quality target was missed. A quality target
+controls the confidence label; it does not decide whether the writer receives the work.
+A fallback must never be described as fully verified when a check did not complete.
+
+Initial gate failure → keep the best safe edit and flag the missed target. Final
+verification defect → one targeted repair and one local recheck. Then deliver.
 
 ### 8. Fresh-eyes finalizer — approve the reader's copy
 
@@ -671,14 +680,12 @@ roles actually stayed separate, whether any role certified its own output, and
 whether the counts were reported. A generating role cannot answer those about
 itself, which is why they sit with the finalizer.
 
-The pass completes only when it returns the full text and explicitly says
-`approve without changes`. If it changes anything, apply that complete revision,
-then rerun the copy desk, read-aloud editor, verifier, and fresh-eyes finalizer in
-that order. A role 8 edit therefore restarts roles 5 through 8. Limit the loop to
-three rounds. The same exact text must clear roles 5, 6, and 7 and then receive a
-no-change approval from role 8. If safe approval is impossible without guessing,
-return the best source-preserving version that completed every pass and name the
-unresolved span; do not call it fully verified.
+The pass returns the full text plus one status: `approve without changes`, `changed`,
+or `unresolved — needs the writer`. If it changes anything, apply the complete revision
+and rerun the local score, fact, format, and structure checks once. Do not restart the
+model pipeline. If safe approval is impossible without guessing, return the safest
+source-preserving edit, name the unresolved span or unavailable pass, and do not call
+it fully verified.
 
 ### 9. Report in plain language
 
@@ -951,8 +958,8 @@ the AI model already running in the assistant or rewrite this `SKILL.md`.
   the signs of human writing to preserve.
 - `references/readalong.md` — the mandatory, separate read-aloud pass that
   fixes flow, cohesion, and stumbles directly in the deliverable.
-- `references/fresh-eyes.md` — the separate first-time-reader finalizer that
-  approves the verified text without changes or restarts every final pass.
+- `references/fresh-eyes.md` — the first-time-reader finalizer and its bounded
+  one-recheck rule.
 - `references/copy-desk.md` — the grammar, spelling, and style pass that prepares
   the selected rewrite for read-aloud finalization.
 - `references/eval.md` — the pass/fail checklist for roles 7 and 8. It carries the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "zero-slop",
-  "version": "2.8.6",
+  "version": "2.8.8",
   "description": "An Agent Skill that scores AI-sounding prose locally, guides the host AI through an editorial rewrite, and checks changed source details. The local tools run offline with zero dependencies.",
   "bin": {
     "zero-slop": "bin/zero-slop.mjs"

package/references/copy-desk.md

@@ -76,14 +76,11 @@ artifact returned by the read-aloud editor:
 2. Compare it directly with both the original and selected rewrite for preserved
    claims, qualifiers, intended voice, regional spelling, format, and non-prose
    structure. The script cannot detect every semantic or stylistic change.
-3. If any check or comparison requires a repair, apply it, send the repaired
-   text through the copy desk and read-aloud pass again, and repeat all
-   final checks.
+3. If a check requires a repair, apply one targeted correction and rerun the
+   local score, fact, format, and structure checks once.
 
 After verification, send that exact text through the finalizer in
-`references/fresh-eyes.md`. Stop only when the same artifact clears the copy desk,
-read-aloud pass, semantic and format review, scorer, fidelity check, and receives a
-no-change fresh-eyes approval. Limit this repair loop to
-three rounds. If an issue still cannot be resolved without guessing, return the best
-source-preserving version that completed both editorial passes, flag the unresolved
-span and failed check plainly, and do not describe the fallback as fully verified.
+`references/fresh-eyes.md`. Do not restart the model pipeline. If an issue still
+cannot be resolved without guessing after the one local recheck, return the best
+source-preserving version, flag the unresolved span and failed check plainly, and do
+not describe the fallback as fully verified.

package/references/eval.md

@@ -3,8 +3,11 @@
 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.
 
-Any fail sends the text back through the copy desk and read-aloud pass, after which
-every check here runs again on the new text. Limit that loop to three rounds.
+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
+the confidence label. It never suppresses the edited text when that edit preserves
+the source. After the recheck,
+return the safest source-preserving edit and name every unresolved check plainly.
 
 **This file exists because the meter cannot see most of what is on it.** The scorer
 reads the lexically anchored subset: listed phrases, sentence-length variance,
@@ -298,16 +301,18 @@ the report even when they are zero.
 
 ## F. Process integrity
 
-Role 8 answers this section. It is the check no single-agent eval can make.
+Role 8 answers this section when it runs separately. In a one-request service, the
+local verifier records the execution facts and the report names the consolidation.
 
-73. **Roles stayed separate.** The copy desk, read-aloud pass, verification, and
-    fresh-eyes review each ran as a distinct pass.
-74. **No self-certification.** No role graded text it generated.
+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.
 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. **Role 8 approved without changes.** If it changed anything, roles 5 through 8
-    ran again on the revision.
-78. **Fallbacks named honestly.** If the three-round limit was reached, the report
-    says which check failed and does not describe the result as fully verified.
+77. **Finalizer edits were rechecked locally.** If role 8 changed anything, the score,
+    fact, format, and structure checks ran once on that exact revision.
+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

@@ -48,17 +48,11 @@ and name the ambiguity.
 > followed by exactly one status: `approve without changes`, `changed — rerun final
 > checks`, or `unresolved — needs the writer`.
 
-## Closed finalization loop
-
-Approval is valid only when the finalizer returns `approve without changes`. If it
-changes even one character, apply the complete revision and rerun, in order:
-
-1. the copy desk;
-2. the read-aloud editor;
-3. the verifier, including local and contextual checks; and
-4. this fresh-eyes finalizer.
-
-Repeat for at most three rounds. The same exact text must clear all four roles, with
-the fresh-eyes finalizer approving it unchanged. If the remaining issue cannot be
-fixed without guessing, return the best source-preserving text that completed every
-pass, identify the unresolved span, and do not describe it as fully verified.
+## Bounded finalization
+
+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
+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

@@ -110,22 +110,19 @@ use this brief:
 
 Apply the returned artifact to the actual deliverable before verification.
 
-## Finalization loop
+## Bounded finalization
 
 Verify the exact artifact returned by the read-aloud editor:
 
 1. Rerun the heuristic surface scorer and scripted fidelity check.
 2. Compare it directly with the original and selected rewrite for claims,
    qualifiers, intended voice, regional spelling, format, and non-prose structure.
-3. If any check requires a textual repair, apply it, run the copy desk again, run
-   this read-aloud pass again, and repeat every final check.
-
-Stop only when the same artifact has cleared the copy desk, final read-aloud pass,
-semantic and format review, scorer, fidelity check, and the separate fresh-eyes
-finalizer in `fresh-eyes.md`. Limit this repair loop to
-three rounds. If an issue still cannot be resolved without guessing, return the best
-source-preserving version that completed both editorial passes, state the unresolved
-span and failed check plainly, and do not describe the fallback as fully verified.
+3. If a check requires a textual repair, apply one targeted correction and rerun the
+   local score, fact, format, and structure checks 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
+and failed check plainly, and do not describe the fallback as fully verified.
 
 ## Why it is separate