Docs Cleaner

Documentation goes wrong in two different ways, and they need different work.

It rots. Something changed — a port, a path, a procedure, a decision — and the docs still describe the old world. The field's name for this is documentation rot or version drift: artifacts fall out of sync one at a time until people stop trusting the set.

It sprawls. The same topic accumulates three files, each partly right, none authoritative.

Both trace to one root cause: a fact was written down in more places than it was defined. Everything below follows from fixing that.

Core principle (governs both modes)

Critical evaluation before deletion. Never blindly delete. Analyze each section's unique value before proposing removal. The goal is reduction without information loss.

This governs every deletion-shaped action in this skill, not just consolidation: archiving a document, dropping a derived value, and removing a cross-reference are all deletions and all owe the same analysis first. "It looked redundant" is not that analysis.

Which mode are you in?

They share the Drift Test below. When both apply, run Mode 1 first: consolidating docs that are also factually stale just produces one confidently wrong document.

A note on CLAUDE.md / AGENTS.md: both modes apply to them like any other document — govern them after a change, consolidate them when they duplicate each other. The one job that belongs elsewhere is restructuring a single oversized one — pushing low-frequency detail down into references while keeping the top level lean. That is what daymade-claude-code's claude-md-progressive-disclosurer skill is built for; prefer it when the problem is size and layering rather than truth or duplication. If it is not available, Mode 2 still handles the file — you just do the value analysis by hand.

The Drift Test (the core of both modes)

Before writing any value into a document — and when deciding whether an existing one earns its place — ask these three questions in order, and stop at the first hit:

1. Can this be computed from details already recorded here or in the authoritative source? → It is a derived value. Do not write it. Compute it when someone asks. Examples: a count ("6 sections", "N groups covered"), a total, a status summarizing rows in a table below it, "last updated on" when the log underneath already says.

When you remove one, repair the sentence rather than leaving a hole: "the four checks below" becomes "the checks below". Deleting the number and leaving dangling grammar is a different defect, not a fix.

Exception — generated navigation. A table of contents restating the headings is a derived value by this test, and hand-maintained ones should indeed go. But leave it alone when it is produced by the docs toolchain at build time, or when the platform requires the marker to render navigation: those are computed on demand, which is what question 1 asks for. Check for a generator before deleting one.

2. Not computable, but the same fact is authoritatively defined somewhere else? → It is a copy. Link to that definition. Do not restate the value.

3. Neither — it records what was true at a particular moment? → Now you may write it down. Price history rows, decision log entries, "as of 2026-03 the vendor required X" — these are the legitimate case, and deleting them destroys an audit trail.

Question 3 needs care, because every stale value was true at some moment — that is what made it stale. But note what question 3 does and does not decide.

It decides one thing: this passage is not a drift problem. It is not a derived value and not a copy, so none of the drift remedies apply to it — do not recompute it, do not link it away, do not "update" it to the current value. Whether the passage is worth keeping at all is a content-value question, and it is answered in Mode 2's value analysis with the document's owner, not here.

Three earlier drafts of this section tried to decide keep-vs-delete right here, and each one failed review. The reason is worth stating so nobody rebuilds it: every version needed the agent to know something it does not have — whether the content could be regenerated, at what cost, by whom, or whether the other copy someone believes exists actually does. A delete authorized on an unverifiable belief is the self-certifying shape this skill exists to remove, and it would be the only evidence-free deletion in the file.

So the test is just the classification:

Is this passage describing how things are now? Then a dated old value in it is a question 1 or 2 case wearing a date, and the date does not rescue it.

Or is it recording what was true at a moment — a changelog line, a decision log entry, a price history row, an incident timeline, a dated measurement, a postmortem? Then it is question 3. Leave the value alone.

Apply this to the passage, not the file — where passage means the unit that would be read as one thing: the entry, the table row, the dated note, the section under one heading. A how-to guide can contain one genuinely historical block, and a changelog can carry a current-state summary at the top; a file-level verdict gets both wrong. But do not run it below the passage either — a postmortem's timeline and its root-cause section are one record, and splitting them so the timeline can be deleted as "reproducible from logs" destroys the thing whose parts they are.

Question 3 takes precedence over questions 1 and 2 for these passages, even though it is asked later. A changelog line "2026-01-04: raised the upload limit to 50 MB" does hit question 2 — the upload limit is authoritatively defined in config — and stopping there would replace it with a link and destroy the record. Question 1 does the same damage by a different route: "2026-01-04: added eu-west-3, bringing us to 12 regions" contains a count, and a count is question 1's own example — but recomputing it yields today's region count and rewrites what that day recorded. The stop-at-first-hit rule is about efficiency, not about routing history into a link or into a recomputation.

When you cannot tell whether a passage is history or current-state, ask — do not resolve it by deleting. The asymmetry is the reason: deleting a record you mistook for a current-state restatement destroys the only copy, while keeping a restatement you mistook for a record costs a few lines. If you are running unattended and cannot ask, leave the passage untouched and report it.

Linking only solves case 2. Pointing a link at a derived value is a category error: there is no single authoritative cell to point at, so the link becomes scaffolding that goes stale on its own and creates a false impression that the two places are aligned.

Why this ordering matters: the mature answer to drift in API documentation is "generate it from the source, don't hand-copy it" (OpenAPI/Swagger exist for exactly this reason). Question 1 is that same instinct applied to prose — the surest way to keep a number correct is for it not to be written down twice.

Position gives no exemption

A fact is a fact wherever it sits. All of these are body text for the purposes of the test above, and all of them are where stale values actually survive audits:

frontmatter · YAML/JSON metadata · parenthetical asides · a description column in an index table · checkbox state · introductory framing before the real content · cross-document footnotes · an extra column added to a table · commit-message-shaped notes left in prose

What counts as a fact

Anything that changes over time and has an authoritative definition somewhere: numbers, status, ownership, relationships, classifications, rules, decision sources, dates, counts. Business rules count too — "all quotes route through the finance lead" is defined once and referenced, never restated in each downstream doc.

Mode 1 — Post-change governance

Run these in order. The order is the point: fixing derived docs before establishing what is authoritative just multiplies the work.

1. Scope it from the change, not from the repo. Enumerate the facts this change altered — this port, this path, this procedure name, this default. That list is what bounds the work. The list is bounded; the search for each item on it is not — a fact's quiet copies live precisely in files the change never touched, so each fact gets searched repo-wide in step 5. What is forbidden is the other thing: opening files to look for unrelated problems. An unbounded sweep is how a doc task turns into an unreviewable refactor.

2. Identify the authoritative source (SSOT) for each affected fact. Which file defines this port / path / procedure — as opposed to mentioning it? Update that first. Everything downstream either points at it or is derived from it.

The practical test for "defines" is where a change has to be made for reality to change: the file the deployment actually reads, the schema the code loads, the runbook a human follows step by step. A file that recites the value while explaining something else is mentioning it. When two files both look declarative, prefer, in order: the one a machine consumes over one only humans read; the one nearest the thing it describes; the one other docs already cite. Then say in your report which you picked and on which of those grounds — an arbitrary pick becomes permanent once step 2 points everything else at it, so it should be a stated decision rather than an accident of which file you opened first.

3. Let the implementation win — as evidence, not as an invitation to edit it. Where code, a script, a config file, or an observed run contradicts the docs, the implementation is the evidence and the doc is the claim. Document what is actually true; never write documentation describing what you wish the code did.

If the implementation itself is what is wrong, stop and report it — do not fix code under this skill. You cannot tell "the doc is stale" from "the code has a bug" by reading the disagreement; that judgment needs the intent behind the change, which is the user's. A documentation pass that silently becomes a code change is the same unbounded-scope failure step 1 forbids, with worse consequences.

4. Decide each affected doc's disposition before editing it.

Do not patch a document whose premise is gone — you get something internally consistent and globally wrong, which is harder to spot than an obviously outdated file. A "this section is superseded" note left inline is not a third option. Maintainer bookkeeping does not belong in a reader's path; readers lack the context that makes it meaningful and only get confused. Move it out.

"Archive" means: out of the reader's path, still in history. Concretely: look for a convention the project already has (_archive/, archive/, deprecated/, or whatever it uses) and follow it; if there is none, create _archive/ at the same level as the file. Either way move it with git mv so history follows, then repoint or remove every inbound reference. Do not delete it outright, and do not settle for adding status: deprecated frontmatter while leaving the file where readers still find it — a retired document still in the reader's path is still being read.

Ask before archiving anything you did not create, for the same reason step 3 of Phase 4 needs agreement. Running unattended and cannot ask: leave the document untouched and report it as needing retirement. Do not patch it instead — a document whose premise died gets more misleading, not less, when you update its facts.

5. Find every copy of each changed fact — before you change it. This is the step that gets skipped, and it is where the classic failure lives: the obvious occurrence gets updated and three quiet ones do not. See Verifying below for the exact commands.

The highest-risk shape is renumbering or inserting an item into a sequence, because you are not changing one fact but a whole family of mutually-referencing ones: body text saying "see step 7", cross-references in sibling files, and — the nastiest — prose that baked the count into a sentence ("the four checks below"). That last one contains none of the numbers you are searching for, so a search by step number will never surface it.

There is no mechanical sweep for this, and you should not pretend otherwise. Two obvious scopings were measured and both fail:

• Search count words repo-wide → 224,330 lines / ~550 MB on a real documentation repo (17.7 MB scoped to a single project directory). Nobody runs that, and an instruction that guarantees its own bypass is worse than no instruction.
• Search count words only in files that mention the subject → bounded, but wrong in the direction that matters. Prose that bakes a count into a sentence is exactly the prose that refers to the subject by alias, translation, pronoun, or bare link — "the four checks below", 「预检有四个检查」 — and never by its title. Measured: a file containing a genuine stale count matched zero searches for the subject's name or filename. The filter is anti-correlated with the target.

So do this instead, and report what it does not cover:

1. Read the file you renumbered, in full. Most stale self-references live in it.

2. Build the file list, then search count words inside it. Nothing earlier in Mode 1 produced this list — do not reach for the inbound-reference search in Updating references, which belongs to Mode 2 and answers a different question ("what pointed at this file", not "what discusses this subject"). Produce it here:

   ROOT=$(git rev-parse --show-toplevel 2>/dev/null) || { echo "not in a git repo — pass the docs root explicitly" >&2; exit 1; }
   rg -l -F "<the subject's name>" "$ROOT" --no-ignore --hidden -g '!.git'
   

   That gives you files naming the subject outright; step 3 covers the ones that do not. It is a small set, so name the files directly rather than piping a list into xargs (an empty list makes GNU xargs run the search with no path at all, which searches the current directory instead — a wrong scope that looks like output):

   rg -n -i -H \
     -e '\b(one|two|three|four|five|six|seven|eight|nine|ten|both) [a-z]+\b' \
     -e '[一二三四五六七八九十两](个|条|步|项|点)' -e '[0-9]+(个|条|步|项|点)' \
     -e '\b[0-9]+ [a-z]+\b' \
     path/to/one.md path/to/two.md
   

   Two details, both measured: -H because rg omits the filename when given exactly one file, and "which file" is the entire answer here. And the pattern deliberately takes any following word rather than a list of nouns — a list will not contain the noun your project uses ("four gates", "three lanes") and the miss is silent. Over a handful of files the extra noise is cheap; the missed line is not.

3. List the subject's aliases yourself and search those, then the count words in whatever that turns up. This is the step no tool does for you.

Then say what you checked: "read the renumbered file; searched count words in the N files that reference it by name; prose referring to it by other names was not mechanically covered." That sentence is worth more than a green check mark that quietly meant one of the two failures above.

6. Clean derived values you meet along the way — but only in files you were already editing. Removing a stale count in a file you have open is hygiene, not scope creep. Opening unrelated files to hunt for more is scope creep. That boundary is what keeps this a reviewable change.

7. Report what you did in the structure under What to report.

Mode 2 — Consolidation

Consolidate redundant documentation while preserving all valuable content. The Core principle above governs every deletion in this mode.

Phase 1: Discovery

1. Identify all documentation files covering the topic, and state the boundary you searched within (a directory, a suite, one project) before you search — the same bounded-scope rule Mode 1 step 1 uses. Two agents given "the docs" and no boundary consolidate different file sets. A topic-keyword sweep inside that boundary is the usual method — note -F, since topic terms like C++, .NET or v1.0.0 are literals and as patterns they match almost everything: rg -l -i -F -e '<topic term>' -e '<synonym>' <boundary> --no-ignore --hidden -g '!.git' This list scopes the entire consolidation, so a garbage list means a garbage analysis.
2. Count total lines across files — wc -l over the list from step 1
3. Map content overlap: for each heading in each file, note which other files cover the same ground. The artifact is a table of heading → files-that-also-cover-it; sections appearing in more than one file are the consolidation candidates. This step decides overlap only — it does not decide keep-or-delete. A unique section can still be dropped in Phase 2 for a reason that has nothing to do with duplication (self-evident, superseded, a byproduct), so read this as "nothing here licenses deleting a unique section as redundant", not as "unique means it survives".

Phase 2: Value analysis

For each document, build a section-by-section table:

Note what the Delete example is not: a dated test run, a meeting note, an incident timeline. Those look like the easiest deletions in any document and they are the one class this skill will not let you delete quietly — see the rule three paragraphs down. The example above is a Delete because a named file holds the same content, which is the only Delete you can justify by pointing at something.

• Keep — unique, essential, frequently referenced

• Condense — valuable but verbose

• Delete — duplicate, one-time, self-evident, outdated

• Archive — a whole document whose premise is dead (a cancelled system's design notes, a retired service's postmortem). It is not stale current-state to be corrected and not duplicate content to be dropped. Without this category, "superseded" and "it is a record" pull the same document in opposite directions and whichever you read last wins.

  A premise-dead section inside a live document is the same problem one level down, and Archive does not accept it — it is defined over whole documents. Treat it as a Delete whose reason is the dead premise, and name it in the Phase 3 plan rather than deciding it yourself: "the Kafka migration section, for a migration that was cancelled" is exactly the kind of call the owner may answer with "actually we still reference that".

  An Archive verdict removes that document from this consolidation. It does not go on to Phase 3 or Phase 4 — there is nothing to merge and nothing to condense. Handle it with Mode 1's archive procedure — the paragraph under Mode 1 — Post-change governance that begins "Archive" means: out of the reader's path, still in history — say in your report that you did, and carry on consolidating the rest. Do not put it in the Phase 3 line-count plan either: it is leaving intact, so counting its lines as "reduction" would claim a saving you did not make.

Run the Drift Test over every row as you go, and let it override the categories in both directions:

• A section that only restates values computable from elsewhere is a Delete regardless of how well written it is (question 1).
• A section recording what was true at a moment — a changelog, a decision log, an incident timeline, a dated measurement, a postmortem — is question 3, and question 3 says only that drift remedies do not apply to it. Its keep-or-delete verdict is a value judgment you make here, on this table, like any other row. The one thing you may not do is delete it as a "one-time record" without saying so out loud: it is the class the Drift Test warns is an audit trail, so it goes to the owner as a named proposed deletion with your reason, never as a quiet Delete row.

Detailed criteria: references/value_analysis_template.md.

Phase 3: Consolidation plan

Propose the target structure with real numbers, and present it for agreement — this is the plan Phase 4 asks you to get signed off before anything is deleted:

Before: 726 lines (3 files, high redundancy)
After:  ~100 lines (1 file + a pointer from the project's main doc)
Reduction: 86%
Value preserved: pending — established by the preservation check after execution

The line counts and the reduction are measurements you can take now. "Value preserved" is not: at plan time it would be the deleting party asserting its own deletion was harmless, which is the self-certifying shape this skill exists to remove.

It is filled in after Phase 4 step 7, and it is not a percentage — a percentage of what would have to be invented. It is a count against a stated denominator: "all 34 items on the step-1 inventory survived", or "31 of 34" naming the three and where each went.

The denominator is the Keep and Condense items only. The step-1 inventory deliberately also covers Delete sections, and those items are supposed to be absent from the consolidated file — counting them here would make "all survived" unreachable by construction, and a target nobody can hit is one people stop aiming at. They are accounted for on their own line: "and 6 items from Delete sections, each named with where its copy survives or why it is going". Two numbers, two questions, neither one hiding in the other. Write the denominator, not just the verdict — "every item I checked survived" is true of one item as easily as thirty-four, so it reports your diligence rather than the document's fate. Anything short of all is a finding to report, never a number to round up, and never a reason to undo your own condensation so the report reads better.

Phase 4: Execution

Deleting someone's documentation is irreversible from the reader's side even when git can undo it. Present the Phase 3 plan and get agreement before step 3. Phase 3 says "propose" for a reason; the Core Principle governs how carefully you analyze, not whether you are authorized. If you are running unattended and cannot ask, do steps 1–2, stop, and report what you would delete.

1. Extract the preservation evidence first — before you write a single line of the consolidated document. The order is what makes the check mean anything: a claim list written after the rewrite is a list of what you happened to carry over, so it always matches. Written before, it is a specification the rewrite has to satisfy.

   Be honest about what enforces that: nothing in the finished artifact distinguishes a list written before from one derived after, so this is a convention, not a gate. Give it the only teeth available — emit the list before you start step 2, into the report or a scratch file. A list that exists in writing before the rewrite can be checked against it; one that appears afterwards cannot be told from a transcription of the result.

   The unit is one entry per load-bearing item, not per section and not per category. How finely to cut is a judgment call and two careful agents will land on different counts — that is tolerable. What is not tolerable is which clauses go missing when someone cuts coarsely, so two of them get their own entry no matter how you are counting:

   • The consequence clause. "Roll back before retrying" is an instruction; "retrying on a half-applied migration corrupts the events table and there is nothing that repairs it automatically" is the reason anyone obeys it. Coarse lists keep the instruction and drop the consequence, which is the half that makes it survive a skim.
   • The symptom → cause mapping. "A 401 from the deploy step means the token lapsed, not that the account is wrong" cannot be re-derived from "the token expires every 90 days". It is the troubleshooting content, and it is one line long, so it disappears first.

   Those two are a floor, not the boundary of what matters. They are named because they are the clauses coarse lists drop most reliably — not because a clause outside them is safe. Three that fall through and are worth checking for by hand: ordering constraints ("apply the migration before restarting the workers" — no consequence stated, so the first rule does not reach it), scope qualifiers ("this applies to the staging cluster only"), and thresholds buried in prose ("backs off up to 5 attempts"). Each is one clause long, each changes what the reader does, and none announces itself.

   Go through every section — including the ones marked Delete — and write down, from the source:

   • Keep → one distinctive verbatim line per item in it (each gotcha, each constraint, each snippet — a section holding three gotchas owes three lines, not one)
   • Condense → each load-bearing claim: every constraint, number, condition, warning
   • Delete-as-duplicate → the items in it, and for each, where the surviving copy is. This is the step most likely to be skipped and the one that catches the worst failure: Phase 1 maps overlap by heading, so a section can be marked duplicate on its title while holding one gotcha no other file has. Skip the inventory and that gotcha is invisible to every later check — it was never on the list, so step 7 reports "every item survived" and means it.
   • Delete for any other reason → the items in it and why each one is going. A deletion you can name is a decision; one you cannot is an accident you have not noticed yet.
   • All of the above → every URL, verbatim

   What this list can and cannot do. It catches recall failures — you meant to carry something over and did not. It cannot catch judgment failures: if you decided at Phase 2 that something did not matter, it never enters the list, and step 7 will happily pass without it. That gap is real and nothing in this procedure closes it. The Phase 3 plan goes to a human, but it carries target structure and line counts — not this list — so it cannot catch a misjudged item either; it is a check on the shape of the result, not on what went into it. Do not report step 7 passing as evidence that nothing valuable was lost — report it as what it is, a check that everything you identified was carried over. If you want the judgment reviewed too, the only thing that does it is showing the item list to someone who did not write it.

   One sanity check before continuing: does the item count look like the source material? A three-page runbook that yields four items means you skimmed it — and a thin list is exactly the list that will later pass.

2. Create the consolidated document, carrying every item on that list

3. Delete the redundant source files

4. Update every reference to them (CLAUDE.md / the project's main doc, README, imports)

5. Leave a pointer to the new location in the main doc or README — create one if none existed. Consolidation moves content to a place nobody has a path to yet; without this step the work is discoverable only by whoever did it.

6. Verify no broken links (see Updating references below — grepping the bare filename is not enough)

7. Before calling it finished, run the preservation check below, using the list from step 1. It is a gate on this phase, not an optional appendix.

   If you skipped step 1 and the sources were tracked in git, they are still recoverable — but mind which ref: before the deletion is committed the content is at git show HEAD:<path>; once it is committed that path no longer exists in HEAD and you need git show HEAD~1:<path> (or the commit before the deletion). The path is repo-root-relative, so from a subdirectory use git show HEAD:./<path>. Both wrong forms fail loudly with fatal: path ... does not exist, so you will know. What you must never do is take the expected strings from the consolidated file — that measures it against itself.

Verifying

Two properties make a documentation check trustworthy: it must match literally (facts contain characters a regex engine reads as wildcards — searched as a pattern, v1.0.0 also matches v1X0X0, and [2026-08-13] aborts as an invalid character range), and it must match contiguously (a fact spanning two lines is one fact, not two).

ripgrep gives you both: -F matches literally, -U lets a match span lines.

Always pass an explicit path. This is not a style preference: with no path argument and a non-interactive stdin — which is what an agent's shell has — rg searches stdin instead of your files. Which of two symptoms you get depends on that stdin, and both are measured:

• stdin is closed or an exhausted pipe → exit 1, no output. Byte-identical to "the string does not appear anywhere", so the mistake reads as a clean pass. In a directory where four files contained the string, the pathless form reported nothing.
• stdin is inherited and never reaches EOF → the command hangs until something kills it. If a search inexplicably hangs, check for a missing path argument before concluding the filesystem is slow.

Two flags belong on every search that walks a directory, for the reason spelled out under What makes this check honest: rg skips VCS-ignored files and hidden ones by default, and .github/ — full of documented commands — is hidden. A search without them is looking at a strict subset of the docs (measured: 4 files instead of 6). Searches that name their files explicitly do not need them — rg reads a path you hand it regardless of ignore or hidden rules — which is why the per-item checks later in this file omit them.

Set the search root once, and make it fail loudly rather than quietly shrinking:

ROOT=$(git rev-parse --show-toplevel 2>/dev/null) || { echo "not in a git repo — set ROOT to the docs root by hand" >&2; exit 1; }

Do not write || echo . here. A silent fallback to . scopes the search to whatever directory you happen to be in and still exits 0 — so "repo-wide" quietly becomes "this subtree", and a fact living one level up reports as absent (measured). If you are not in a repo, set ROOT to the docs root by hand; the point is that you chose it.

# Every copy of a fact, before you change it. Path and both flags are load-bearing.
rg -U -F "the exact fact string" "$ROOT" --no-ignore --hidden -g '!.git'

# Per-file counts — the artifact you paste into your report.
rg -U -F --count-matches "the exact fact string" "$ROOT" --no-ignore --hidden -g '!.git'

# Multi-line facts: the newline has to be a real newline. In bash/zsh that means $'...';
# "\n" inside double quotes is a literal backslash-n and matches nothing (measured).
rg -U -F $'first line of the fact\nsecond line of it' "$ROOT" --no-ignore --hidden -g '!.git'

-g '!.git' is not optional once --no-ignore is on: without it the search reports hits in .git/COMMIT_EDITMSG and .git/logs/HEAD — the commit messages that describe your change — as surviving copies. Those can never be "cleaned", so a check that counts them trains you to wave off count mismatches, which disarms it for the real ones. --no-ignore also restores node_modules/ and other vendored trees; exclude those the same way when they appear.

Write down the expected counts before you search, not after. A count you decide once you have seen the output cannot fail. State which files should still contain the value and which should not, then run:

rg -U -F --count-matches "the old value" "$ROOT" --no-ignore --hidden -g '!.git'
rg -U -F --count-matches "the new value" "$ROOT" --no-ignore --hidden -g '!.git'

Compare each result against your list as set equality, in both directions: a file that appears and should not is a leftover, and a file on your list that does not appear is equally a failure — --count-matches omits zero-match files from its output entirely, so a document that was supposed to receive the new value and silently did not looks exactly like a clean run.

What makes this check honest:

• The old value does not always go to zero. A changelog entry, a decision log, or an "as of <date> this was X" sentence records history and legitimately keeps the old value (Drift Test question 3). Deleting those so a count reaches zero destroys an audit trail in order to satisfy a check. Name the historical files in advance; anything outside that list is a leftover.
• --no-ignore --hidden is on purpose, and they are not the same flag. By default rg skips both VCS-ignored files and hidden ones. --no-ignore only restores the first group; without --hidden the search still never opens .github/, which is where contributing guides, issue templates, and workflows full of documented commands live (measured: adding --hidden took a sample search from 4 files to 6). A convergence gate that cannot see those directories goes green over a surviving stale copy. Include them and judge the extra hits rather than assuming they do not matter.

Calibrate before you trust a zero — whichever tool you used. A search returning nothing has two causes: the fact is gone, or the command is wrong. They look identical. Before reporting "no remaining copies", run the same command shape against a string you know is present. If that also returns nothing, the instrument is broken, not the docs. This is the direct antidote to the pathless-rg failure above, and it costs one line.

Without ripgrep

git grep -nF "fact" -- . works in any git repo and is one implementation across platforms (unlike grep, whose recursion and symlink behavior differ between BSD, GNU, and drop-in replacements). It has these limits, all measured, all of which produce a confident wrong answer rather than an error:

It searches tracked files only. The consolidated document you just wrote in Mode 2 is new and untracked — so a preservation check run through git grep reports its content missing and you "restore" what was never lost. git add it first, or read that file directly.

-F with a multi-line string silently degrades to matching the lines independently. A file that kept only the first line of a two-line fact still reports a hit — so a lossy move gets certified as "the content is still there", the one error a preservation check exists to catch. True of both grep and git grep.

With neither tool available, a contiguous check runs anywhere Python does:

python3 -c 'import sys, pathlib
# .rstrip("\n") matters: your editor adds a trailing newline to needle.txt, and a fact
# sitting at a file'"'"'s end has none — without the strip, that copy is silently missed.
needle = pathlib.Path(sys.argv[1]).read_text(encoding="utf-8").rstrip("\n")
for f in sys.argv[2:]:
    n = pathlib.Path(f).read_text(encoding="utf-8").count(needle)
    if n: print(f, n)' needle.txt doc1.md doc2.md

Updating references to a file you moved or deleted

Searching the bare filename finds some inbound links and misses the rest, then reports "verified". A file is referenced in more shapes than one:

BASE=old-doc            # filename without extension
rg -n -F -e "$BASE.md" -e "[[$BASE" -e "($BASE" -e "/$BASE" "$ROOT" --no-ignore --hidden -g '!.git'

That covers Markdown links, wiki-style [[old-doc]], bare paths, and imports. What it will not do for you:

• Relative paths break when a file changes directory depth. ../guide.md from one level is ../../guide.md from another. Grepping the filename finds the line; whether the path still resolves is a separate question, per link.

• "Verify no broken links" has two directions, and Phase 4 only owes you one of them. Inbound — "does anything still point at what I removed?" — is what consolidation can break, and the search above answers it. Do that one; it is the step's actual obligation.

  Outbound — "does every link in these files still resolve?" — is a general docs-QA job, not something to hand-roll here. Use a real link checker:

  lychee --offline --root-dir "$PWD" docs/   # local paths only, no network
  lychee docs/                               # also checks external URLs (slow, fails for unrelated reasons)
  

  --root-dir is not optional if the docs use root-relative links (/docs/x.md — the normal style under mkdocs, Docusaurus, GitHub Pages): without it lychee reports every one of them as an error, which is the same false-flagging the paragraph below says to avoid. Anchors (x.md#section) are not checked unless you add --include-fragments.

  0 Total is not a pass — calibrate before you believe it. Measured (lychee 0.24.2): a path that does not exist fails loudly with exit 1, but a directory that exists and holds no Markdown prints 🔍 0 Total … ✅ 0 OK 🚫 0 Errors and exits 0 — identical to a clean run. Point it one directory too high or too deep and you get a green check that looked at nothing. Read the Total count: if it is 0, either confirm the files genuinely contain no links, or you searched the wrong place. (Files whose links are all in inline code or fenced blocks legitimately total 0 — lychee skips both, correctly.)

  lychee (brew install lychee / cargo install lychee) parses Markdown properly, which a shell pipeline does not: angle-bracket links [x](<my file.md>), titled links [x](y.md "Title"), nested parens, reference-style [x][ref], and links inside fenced code blocks all break a regex-and-test -e approach, in both directions — silently missing real breakage and falsely flagging valid links. If no link checker is available, say the outbound check was not run rather than substituting a hand-rolled one.

Preservation check (Mode 2)

The old form of this check was a list of things to tick off — "essential procedures preserved", "gotchas preserved". The person who deleted them ticks the boxes, so it can never fail. Replace self-assessment with evidence: name each specific item and show where it now lives. The categories further down are a checklist of kinds to sweep for, not the unit the evidence is counted in.

Use the list you built in Phase 4 step 1 — one entry per load-bearing item, written before the rewrite. (Missed that step? Phase 4 step 7 says how to recover the sources from git and which ref to use.) What you must never do is take the expected strings from the consolidated file: that only proves the file matches itself.

The method depends on the disposition, and using the wrong one produces a false deletion report. A section marked Keep should survive word for word; a section marked Condense was supposed to be rewritten, so a verbatim search for its old sentence correctly returns nothing — that is the condensation working, not content loss.

Which of those two rows applies is decided before the rewrite, never at check time. If you cannot point at a plan or inventory line saying this section was being merged, the row above applies and a miss is a miss. This is not bookkeeping: letting the agent whose search just failed pick which row it is graded by converts a hard verbatim gate into a soft one, and it converts it only in the cases where it just caught something. The failure it lets through is not merging — it is a plain Keep section quietly reworded during the rewrite, its consequence clause dropped, relabelled "merged" so the softer method returns a hit on the weakened sentence. | Condense | Per saved claim (each constraint, number, condition, warning): quote the line in the consolidated file that carries it. A claim you cannot point to was dropped, whatever the prose length says. | | Delete | Per saved item: where the surviving copy is, or — if there is none — the reason it is going. "It looked redundant" is not a reason; naming what was in it is. |

The categories below are a recall prompt, not the unit — they exist so you notice a class you forgot, and each one still resolves to per-item evidence from the table above. Six ticked categories is not six pieces of evidence.

A hit proves the string is in the file; it does not prove it is in the right place. A staging-only warning that landed under a "Production" heading, or a constraint moved out from under the conditional that scoped it, passes this check while now saying something false. When an item's meaning depends on where it sits, read the surrounding lines rather than stopping at the match — the same reason a lossy multi-line move can be certified as "still there".

Anything you cannot produce a hit for was not preserved — it was deleted. That may be the right call, but it belongs in the report as a deletion, not inside a checkmark.

A zero-deletion outcome is a legitimate result. If the value analysis says every section earns its place, report "nothing to consolidate" and stop. The reduction percentage in Phase 3 is a measurement, never a target — an agent that deletes something in order to have a number to report has inverted the whole point.

What to report

State each of these explicitly. Silence on any of them reads as "nothing to say", which is rarely true:

1. Inconsistencies found — what disagreed with what
2. Authoritative source used — which file you treated as defining each fact, and on which ground you picked it when more than one looked declarative
3. Files changed and why — one line each
4. Derived values and redundant references removed — where each one was, and why it qualified as derived rather than as a fact worth keeping
5. Search evidence — the commands you ran and their counts, so every claim above is checkable rather than asserted. This is the same artifact listed under Output artifacts; both lists mean one report, not two.
6. The corrected procedure, end to end, if a procedure changed
7. What is still unresolved — untouched exceptions, known-stale areas out of scope, risks you could not close

In Mode 2 this report also carries the value-analysis table and the before/after metrics.

Point 4 is the one that gets dropped, and it is the one that compounds: naming a removed derived value teaches the reader why it should not come back.

Anti-patterns

Output artifacts

A successful pass produces:

1. The corrected documents — with a single authoritative location per fact
2. Value analysis (Mode 2) — section-by-section justification
3. Before/after metrics (Mode 2) — lines reduced, value preserved
4. Search evidence — the commands run and their counts, so the claims are checkable
5. Updated references — a pointer to the new location exists in the main doc or README, and every pointer to moved content resolves

Next Step

After the docs are consistent, the natural follow-ons:

Docs are now consistent — [N] files updated, [M] derived values removed.

Options:
A) Convert the recurring parts into a project rule — if the same drift keeps
   recurring, the fix is a rule in the project's main doc, not another cleanup pass
B) Render or publish the result — if these docs have a build/publish step
C) Nothing further — the docs are the deliverable