How we work, and where this could be wrong

How topics are chosen

A topic reaches the slate only if it passes four tests. They are applied in order, and the fourth one has already removed two articles that would otherwise have been written.

1. Somebody keeps asking, and nobody answers with a citation

We start from questions practitioners ask each other in public — on nursing and surgical-technologist forums, in professional discussion boards, and in the comment threads under trade articles — and we look for the pattern that defines this subject: a question that recurs for years, attracts confident answers, and never once attracts a citation. One thread we read, on whether suction canisters may be emptied into a toilet, ran for pages; several replies invoked “new OSHA guidelines”, and not one of them named a provision. That thread is the reader this publication is written for: not somebody who wants a guide, but somebody who has been handed a policy they cannot source and needs the sentence so they can take it back to whoever made the claim.

Public discussion decides what we write about. It never decides what we say. Nothing on this site is sourced to a forum post. Demand evidence sets the agenda; primary documents supply every answer.

2. A primary document exists, and we can retrieve it

The question has to be answerable out of something we can obtain and read in full: a codified rule, an agency letter of interpretation, a state administrative code section, a municipal ordinance, a statute as enacted, a published study. Where the honest answer is that no such document exists, that itself is publishable and we publish it — but it has to be established by retrieving the governing document and scanning it, not by failing to find one in a search engine.

3. The existing answer is wrong, uncited, or written by an interested party

We are not trying to write a second version of a page that already exists and is good. The subjects that make the slate are the ones where the first page of results is held by companies paid by the pound to remove the waste stream they are describing, or where the most-cited independent source has quietly gone offline and the numbers from it are still circulating. The single strongest reason this publication exists is that a trade magazine which covered this beat almost every year from 2004 to 2021 was discontinued on 1 November 2025, and its dated archive is now almost entirely unreachable.

4. We can print the sentence that does the work

This is the test that costs us articles. The whole promise of this site is that we quote the operative text; an article whose governing documents are copyrighted standards sold by their publishers would break that promise on its own page. Two planned articles were deferred for exactly this reason — one on whether a building and its plumbing can take a direct-to-drain installation, whose sources are sold design guidelines, a health-care facilities code and a plumbing code section; and one on the handling of suction canisters between cases, whose governing positions are copyrighted association guidelines. They are deferred rather than abandoned: if we obtain lawful access on terms that let us quote enough to be useful, they get written. Until then we would rather publish nothing than publish a summary of a document our reader cannot check.

What we decline outright

• Anything whose honest output is an instruction. Post-exposure management is on the slate, but only as a strictly citational brief: a source-by-source record of what each retrieved guideline says, with no synthesised schedule and no ordered protocol presented as ours. Compressing that material into a short answer could only produce time-critical instruction to a frightened individual, which is medical advice and is not ours to give.
• A regulatory history of any manufacturer’s device problems. We could write it from the public record. We will not publish it here. Published at this domain name — which for six years before us served a medical device company in exactly this category — an article enumerating a competitor’s recall record reads as that company relitigating its market position, whoever actually wrote it. That is not a defensible piece of work on this address, and it is held indefinitely rather than parked for later.
• Any topic we cannot write without naming a winner. The cost-model article runs in both directions or it does not run.

How a claim is traced

Most of the figures that anchor this field circulate without a live citation: a number, an attribution to an agency or a trade publication, and no document. The procedure for those is fixed, and the ledger at Claims publishes the whole trail rather than the verdict.

1. Record the claim as it circulates, in the exact wording used by the pages that repeat it, with the attribution they give it.
2. Collect the citations, then follow them. Each hop is recorded with the document it points at, the HTTP status the server returned, and the date we requested it. A 404, a 403 and a paywall are three different findings and are not collapsed into “unavailable”.
3. Retrieve the document itself. We request it from the issuing body’s own server wherever one exists, strip the markup locally or convert the PDF to layout-preserving text, and read the whole document. No quotation on this site has passed through a summariser. That rule is not decorative: during our own research an automated summariser produced a sentence attributed to an OSHA letter that the letter does not contain. Reproducing that distortion layer is the one failure that would end this publication, because correcting it is why it exists.
4. Scan, and publish the scan. Where the question is whether a document contains something, we run a case-insensitive scan of the retrieved text and publish the term list, the occurrence counts, the character count of the text scanned and the date. That is what turns “the standard contains no percentage” from an assertion into something a reader can repeat.
5. Where the live server refuses us, use an archive capture and say so. Some official code servers return 403 to anything that is not a browser. Where the text came from a web archive capture, the capture date is printed with it, and it is checked against a second official server wherever one exists. A capture-only retrieval never carries our strongest label.
6. Label it, then write the two fields that matter. The label goes on last, from the closed six-label vocabulary published at Editorial standards. Then the row states what the record does support — something the reader can cite instead, with its source — and this would change if, naming the specific document that would move the label. A row without both fields does not publish.

The governing sentence of the whole exercise is printed on the ledger itself: a broken chain is a finding about the record, not about the world. A figure we could not trace may well be correct. What we are reporting is that we could not find the evidence, which is a different and much smaller statement, and we hold that line even when the temptation to write the bigger sentence is strong.

How the registers are built and verified

The three registers are the most maintenance-expensive thing here and the most useful: they are the parts of this subject where the answer genuinely varies by jurisdiction, and where every page on the open web either flattens the variation or gets it wrong. Each is a table with one row per jurisdiction, a stable anchor on every row, and per-row provenance.

1. Fix the scope in writing before any row is filled. Each register states what it covers and what it does not. A jurisdiction we chose not to cover is named in a prose scope note as uncovered — which is not the same as silent, and the distinction is stated on the page.
2. Go to the jurisdiction’s own publisher. A state rule comes from that state’s administrative code server; a statute comes from that legislature’s own text; an ordinance comes from the utility’s or the municipality’s own code server. A tracker, a law-firm client alert, an association summary and a vendor page are all secondary and none of them can support a row.
3. Quote, or do not claim. A row carries our Verified or Qualified label only if the retrieved primary text is quoted in the row itself, with its retrieval date. Everything else ships as Not checked, however confident anyone is about the answer. That single rule is what keeps the negative space in these tables honest, and it is why several states a reader might expect to see resolved are published as unread.
4. Check the number as well as the text. Citations move. One state rule in this corpus was renumbered in 2025 and the renumbering was not cosmetic — the new text added a word that narrows what it covers, so anyone quoting the old number is quoting a rule that no longer exists in that form. Every citation is confirmed against the code server’s current numbering, and the superseded number is left in the register, struck through and dated, so that a reader arriving with the old citation lands somewhere true.
5. State the caveat in the row, not in a general disclaimer. An agency web page that contradicts the agency’s own rule, a chapter that has entered rule development, a capture-only retrieval, an ordinance whose current PDF will not load — each is written into the row it affects, next to the quote it qualifies, because a caveat parked at the bottom of a page is a caveat nobody reads.
6. Never delete a row. Superseded rows are struck, dated and left in place with the replacement beneath. Anchors stay stable so a row can be linked into an argument and still resolve later.
7. Re-read on a cadence we can actually keep — annually on 1 March for the two sewer registers, twice a year for the surgical-smoke register, because legislatures move faster than administrative codes. Slower than we would like, and chosen for that reason: a missed review date is a worse signal than no date.

Every register carries a standing line above its first table saying that a reader has to check their own state rule and their own treatment authority’s ordinance, and that the register is a survey of published law rather than advice about any facility. That is not throat-clearing. Two of the three legal layers in this subject are local, and a register is at best a map of where to look.

Where this could be wrong

This section is written before anyone else writes it. Every item is a real weakness in the way this publication is produced, stated in the strongest form we can put it.

One person reads everything

There is no second reader, no fact-checking desk and no clinical reviewer. A misread provision, a transposed subsection letter, a quotation pulled from the wrong paragraph of the right document — none of those would be caught by anyone before publication. The controls that exist are procedural (quotations compared character by character against the retrieved document; every number traced to a numbered source) and procedural controls fail quietly. The mitigation we can actually offer is that everything is quoted and dated, so an error is visible to any reader who opens the citation, and that corrections are published permanently rather than edited away.

The correction process has never been used

No corrections have been issued, because nothing has been published long enough to be corrected. A correction policy that has never been exercised is a promise, not a record. Judge it when there is something at Corrections to judge.

Everything here is a snapshot, and law is not

A row labelled Verified is verified as of the retrieval date printed in it, and not as of the moment you are reading it. Administrative codes are amended without announcement; chapters enter rule development; a legislature can change an answer in a spring session that our March review will not see until the following year. At least one register row rests on text retrieved through an archive capture from an earlier date, with the residual risk of a later amendment stated in the row itself. Read the dates. They are printed on every row for exactly this reason.

A keyword scan proves a term is absent, not that a duty is absent

Our Silent label means that the governing text we retrieved does not contain the terms we searched for. A rule can impose an obligation without using the word we scanned for, through a definition somewhere else in the code, through a general prohibition, or through an incorporation by reference we did not follow. A scan is only as good as its term list, which is why we publish the term list, the counts and the character count of the text scanned — so that a reader who knows the term we missed can send it to us.

Our reading of an ambiguous provision is a reading, not a holding

Nobody here is a lawyer. Where a provision can be read two ways and no agency or court has settled it, we publish the ambiguity as an ambiguity and label our reading as a reading. But the choice of which ambiguities to raise is itself a judgment, and a provision we found clear may be one a regulator reads differently.

The demand evidence behind our topic choices is partial

The public discussion we read to choose topics came largely from archived records rather than live pages, because several of the relevant platforms block automated access. Vote and reply counts in those archives are snapshots that drift. Several communities we wanted to hear from — surgical technologists, sterile processing, infection control, biomedical engineering — could not be reached at all, and the absence of a question from a community we could not reach is not evidence that the question is not being asked there. This affects what we write about. It does not touch what any page says, because no page is sourced to a forum.

Numbers we compute are ours, and inherit every input’s weakness

Where this site models anything — cost per case, cost per ton, the arithmetic of a business case — the output is our arithmetic on other people’s inputs. The best available benchmark data in this field comes from a self-selected cohort of award applicants rather than a census of hospitals, its publisher has flagged one recent collection year as possibly artifactual, and several of its own published tables contain errors we found while reading them. We use it because it is the best there is, we label every figure with its collection year and its cohort, and we put every assumption in a table a reader can override. A model built on the best available data in a field with poor data is still a model built on poor data, and its output should be treated as a structure for the reader’s own arithmetic rather than as a number.

We could not obtain one document that this whole subject rests on

A November 2014 trade article is where several of the figures still quoted in this field originated, including the cost-per-pound figure that anchors most business cases now in circulation. Its publication has been discontinued and the article no longer resolves. We have not obtained a copy. That has two consequences we would rather state than have discovered: we cannot confirm from the original what its figures were measuring or what cohort they described, so we rebuild from current primary inputs instead of correcting it line by line; and we cannot rule out that it contained a qualification everyone dropped. If you have a copy, we would like to see it — and to be explicit about why, the only use we would make of it is to know what ground it covered. Not a word of it would be reused.

Where this site is published has shaped what is on it

This domain served a medical device company’s website from 2007 to about 2013, and the company’s only product was in the category this publication covers. That history constrains us in ways a reader is entitled to know about. One article we could write from the public record is held indefinitely for that reason alone. Two articles on the slate do carry material adverse to identifiable product categories; neither is published yet, and both will run under a rule that no brand of fluid-management system, canister, solidifier or liner is named anywhere in them. Where a manufacturer is named at all it will be because the identity of the party is itself the fact being reported, and the agency’s own position that it does not approve or endorse products will be quoted alongside it. In our assessment those constraints make for better work. But they are constraints, they are not neutral, and a reader weighing this publication should weigh them. The full history of the domain, with its sources, is at The Domain Record.

Being right about a citation is not the same as being useful

The method here is good at one thing: establishing what a document says and where a number came from. It is structurally poor at everything a practitioner also needs — what a surveyor will do in your building, what your state’s inspectors actually enforce, what your treatment authority will say when you telephone them. We can tell you what the rule says. We cannot tell you what will happen, and a page here should never be read as though we could.

If any of the above is worse than we have described it, that is the correction we most want to receive. Send it, with the document, to [email protected].

Signed

Zane Hitchcox, publisher, sole editor and sole author of The Operative Sentence. 6 August 2026. Not a clinician; has never worked in an operating room. No financial relationship with any manufacturer, distributor, waste contractor or trade body. [email protected]