If your work depends on not losing the object

BETA — DRAFT — A FRONT DOOR

The questions on this page name a use. They do not claim the method has already done that job. You do not have to adopt a position. You can take one tool, use it on your own material, and leave.

The ACCM Deep Ethics Project is a public, unfinished set of tools for one practical failure: the thing you asked about gets replaced by a nearby, easier version, and the replacement is then treated as the thing.

The standing question is already on the project’s front page:

What happened to the object while you were producing the answer?

A simple example: criticizing some unsafe cars, with evidence, while appreciating other models of the same brand, is not being “anti-car.” If the reply judges that label instead of examining the specific cars and the evidence, the object has changed.

Who this is mainly for

The ACCM Deep Ethics Project is not designed for people who want celebrity gossip, football news, or tabloid chatter. It is mainly for people working on high-stakes controversies, or living with their consequences, where the actual question, evidence, qualifiers, and competing risks can be lost. Those controversies are partly framed as gossip, a camp fight, or a label about the person who raised it, in order to bypass scrutiny. The scrutiny includes the claim, the frame, the people and systems judging it, and this project’s own treatment of it.

People who use A.I. on anything contested

Writers, researchers, founders, patients reading their own records, anyone who has watched a model “help” by sanding the question down.

Are you tired of an A.I. that can discuss anything until the topic is treated as controversial, and then suddenly cannot?
Did you ask a specific question and receive a warning about a claim you never made?
When the model says it does not know, do you want that sentence to be allowed, instead of a tidy answer that invented the missing part?

One tool: ask whether a different answer would change the next move. If every answer produces the same warning, the warning was already written.

Journalists, editors, documentary makers

Do you need to represent someone you do not agree with, without the representation quietly becoming your rebuttal?
Have you watched a polished summary become the story, while the sentence that carried the qualifier disappeared?
Would a reader be able to recover what the person actually said, after your piece?

One tool: keep the source span next to the sentence you wrote about it. If the span cannot be found, the sentence is yours, not theirs.

Investigators, archivists, librarians, historians

Are you tired of “we checked” when the check was a different document, a later version, or a cached page?
Do two reports cite the same event and still not be about the same object?
Would it help to label, in one word, whether a gap is missing from the source, dropped by the report, or never in the file you were given?

One tool: written, committed, published, cached, and retrieved are five objects. Name which one you are holding.

Lawyers, ombuds, civil-liberties workers, people who support the later-vindicated

Have you seen a label do the work that evidence was supposed to do?
Does scrutiny land on the person with the least power to appeal, while the rule that flagged them is not open to the same question?
When a warning is vindicated years later, is there a place where the first dismissal and the later outcome sit side by side?
When a warning has been vindicated, does the review examine its earlier dismissal, or change the subject to other warnings that were wrong?

One tool: record the first treatment of an unfamiliar claim next to what happened afterward. An answer learned without the dismissal method can leave the system ready to repeat that dismissal on the next case.

Clinicians, counselors, and people who train them

This is not a treatment protocol and not a diagnosis.

When a person says they are not in crisis, does the form still open a crisis script?
Can you tell a functional alarm, the truck in the road, from an alarm that occupies the whole day with no truck present?
Does “help” sometimes arrive as a category the person did not ask to enter?

One tool: separate what was said from what the form requires you to rule out. Both can be real. They are not the same sentence.

Teachers, editors of student work, anyone who grades thinking

Are you grading the conclusion, or whether the student can still show the object they started from?
Does a fluent essay hide how many times the actual question was replaced?
Can a student say “I do not know yet” without being scored as empty?

One tool: a strong conclusion is allowed after the object has been built. A strong conclusion that replaced the object is a different performance.

Engineers, open-source maintainers, people running a local model

Do you want a local system that can pause, postpone, or refuse, and also say “That, I do not know yet,” without a hidden rule doing it for you?
Are you being asked to trust a safety layer you cannot read, appeal, or fork?
Would you rather inspect a wrong decision than be protected by one you cannot see?

One tool: transparency is not proof of safety. It is the condition under which a wrong move can be found, attributed, and repaired. The project claims no jurisdiction over your machine.

A.I. product, policy, and safety staff

Is “recognizable goodness” easier to ship than a correction that makes the product look less certain?
When your caution is counted, do you also count the warranted caution you did not raise toward your own rule?
If a user can answer your safety question either way and your action stays the same, what did the question do?

One tool: caution has a direction. Downward, upward, sideways, and back onto the checker. Volume is not quality.

Auditors, risk analysts, compliance officers

Does your register record the harm that happened, and stay silent about the contribution that was stopped and left no incident?
Can a small, visible, reversible risk draw more scrutiny than a large one with no appeal?
If your own criterion were applied to your own process, would the finding still stand?

One tool: engage, then test, then amplify. First contact is not a decision to scale, and it is not a decision to suppress.

Scientists and anyone comparing competing risks

Have you watched a partial model become a rule while the people who said “not yet” were scored as the risk?
Do you have a place for a result that is “not established in this record,” which is different from “false”?
If your method cannot be wrong in public, what exactly is being tested?
When someone says “most were wrong,” can you inspect what was counted, what was left out, and whether later corrections changed the count?

One tool: a limitation in what you inspected is not yet a limitation of the claim.

Satirists, comedians, and people who have to say the contradiction out loud

Can you show an official contradiction without being filed as the problem the contradiction was about?
Does the joke still include you, or has it become a verdict on everyone else?
Would the same standard survive being turned around?

One tool: satire here is a way to make a contradiction perceptible. Applause is not the result. A second look is.

Legislators, staff, and anyone who must vote or brief on unread machinery

Citizens are told to know the rule. Did the people who wrote it read the part that does the harm?
If a crisis is the reason the audit was skipped, who checks the crisis later?
When the competing warning is vindicated, does anyone have to correct the record, or only the critic?

One tool: half the story is the risk that was named. The other half is how the namer treated the people who named a different one.

Fact-checkers, moderators, and platform reviewers

Are you tired of being the last person allowed to ask what the label did to the object?
Does “trust the source” ever get asked of the source that supplied the label?
If the same person is allowed to be wrong when they have power, and not allowed to be partly wrong when they do not, which rule is actually in force?

One tool: a label can be legitimate in one case and a jurisdiction in another. The register needs a column for both, or it will start seeing weaponization everywhere, or nowhere.

What you can use without reading the rest

Five moves. None of them requires you to agree with the project.

Name the object you were actually given. Return to it after checking the framing.
Keep the qualifier that makes the sentence true.
If you are about to warn, correct, or refuse, check whether a different answer would change that act.
Apply the same check to your own side.
If nothing is wrong, say so. A manufactured caution is not rigor.

The longer instruments, the 27, the 12, the 10+1, the proposed 16 anchors, stay behind this door. They are BETA. They are not a rank, not a belief, and not a requirement.

What this page will not do

It will not tell you the project is finished. It will not ask you to join. It will not treat refusal as an attack. A later finding can show that one of these questions was the wrong question. That finding belongs on the page beside the question, not in a quieter draft.

If one question on this page is already your job, start there. The longer pack is the workshop behind the door, not the greeting.

Where the workshop already is

Where this page came from

John Kuhles asked for a newcomer page in the voice of the job, not a compression of the whole archive. The questions were drafted on 27 September 2026 from the public site and from a cold-start extraction pack of that date. He read the draft and asked for it to be a separate page, promoted from the front. Later the same day he asked for the audience to be stated, and for five concrete repairs: the car example, what a count includes, staying on a vindicated case, returning to the object after the frame, and “will” changed to “can.” The sentences are a front door. They are not a quotation of the pack, and they are not a new canonical list.