A Rejection Belongs to the Work

With a lot of the work we do at Those Dark Arts, the processes tend to be recursive. An output becomes evidence for the next piece of work. A criticism changes the conditions under which the next attempt is made. A system makes something, the thing is judged, and the judgement becomes part of the system that makes whatever follows.

Usually that recursion sits quietly in the background. This time it became unusually literal.

We commissioned an article called A Rejection Belongs to the Work. The production completed successfully and returned two finished drafts. We rejected both of them.

The event gave the title something the two drafts had failed to find: an actual rejection, a clear account of what was rejected, and a reason for keeping the result. The wrong article produced the finding the right article needed.

The drafts did not improve by being rejected. What became useful was not their argument but the recorded decision that ended it.

A successful production of the wrong article

Nothing had gone mechanically wrong. Both drafts existed, neither was empty, the assigned work had been completed, and the production receipts were intact. The factory had done what it had been asked to do.

Those facts did not make either article worth publishing. Completion and acceptance are different judgements made by different authorities. One says that a process reached its promised end and produced its artifacts. The other says that a person is prepared to stand behind what those artifacts claim.

That distinction sounds obvious until a system makes completion very visible. Green checks, finished graphs and validated files exert their own pressure. They make an output look settled. But an editorial decision cannot be smuggled inside a process status. A flawless production can produce the wrong piece of work, just as a working application can answer the wrong brief.

That was the underlying subject the first two drafts mishandled.

They were trying to reason from the history of an earlier experiment called Shape of a Decision. Three attempts at that experiment had been stopped when their executions became corrupted. A later run completed the experiment: it produced a good narrated film and a functional application. The application itself was then refused because it embodied requirements that contradicted the original idea. It asked the person to construct the analysis by hand and treated fully offline operation as a defining constraint, when the intended object was an inference-powered instrument into which someone could place a decision and receive its shape.

Those are several different events. Attempts were stopped. An experiment succeeded. An application worked. The application was refused as the wrong answer to the remit. None of them was an abandoned article, and none implied that some unfinished object was waiting to be rescued.

The first article production flattened those events into one vague object called “the work”. It then read partial machine records as evidence that this object had been forgotten or had failed to return. One worker’s correct refusal to invent evidence it could not obtain was made to resemble the editorial rejection under discussion. A deliberate final decision was recast as an unresolved interruption.

The mystery existed because the enquiry had lost track of what it was looking at. It could have asked the person who made the decision why the application had been set aside. The answer was simple, known and consistent with the surviving artifacts. Instead, two careful writers received a confused premise and produced careful accounts of something that had not happened.

The useful part of saying no

Our rejection of those drafts was final. We did not request repaired sentences, choose passages to preserve, or invite a third factory draft. We recorded that the fault preceded both writers: the admitted premise was wrong, the concrete subject had disappeared, and the resulting articles circled abstractions.

That record changed what could be known later. It tied a decision to two exact artifacts. It separated the completed production from the rejected articles. It named the person with authority to reject them and preserved the reasons for doing so. It also supplied the missing account of Shape of a Decision without pretending that account had been present in the earlier production.

Rejected work remains evidence, but of a narrower kind. These drafts are not evidence that work had been orphaned; they are evidence that a production, given a confused premise, constructed that account twice. The decision beside them is evidence that an editor refused the account and why. Keeping both prevents either layer from replacing the other. If only the drafts survived, their framing might harden into history. If only the verdict survived, its reasons could not be tested against what was actually written. Together they let a later reader see both the mistake and the authority that ended it.

This is close to the distinction made by the W3C provenance model, which keeps entities, activities and responsible agents separate. A draft is not the activity that produced it. A stopped execution is not a rejected application. A production receipt is not an editorial verdict. Once those objects and acts are allowed to collapse into one another, a record can be meticulously accurate and still tell the wrong story.

A useful rejection therefore needs more than a red mark. It needs a determinate object: this draft, this proposal, this application. It needs an authority: who decided? It needs a disposition: revise, replace, retain, remove, or stop. And if the reason will matter later, the reason has to remain attached closely enough that somebody can distinguish it from a story invented after the fact.

Writing the reason down is not ceremonial. Michael Nygard’s short account of architecture decision records describes what happens when only a decision survives: later people are left to accept it blindly or reverse it blindly. A rejection without its reason creates the same problem. The artifact tells us what was rejected; it does not tell us what judgement rejected it.

The opposite error is to assume that every no is an invitation to continue. It is not. Revision is one possible disposition. Refusal and abandonment are others. Work can be complete, functional and deliberately not adopted. A process that cannot represent that terminal state will keep converting decisions into queues. It will treat “no further action” as missing ownership and regard a closed question as a stalled task.

There was no lost owner in Shape of a Decision. Nobody was waiting for the application to come back. The decision was that it should not continue in that form. The work was complete and not accepted. That is not a contradiction; it is a perfectly ordinary outcome that systems often find strangely difficult to say.

Keeping a failure without rehabilitating it

Why keep the rejected drafts at all?

In this instance, keeping them allowed the later judgement to remain exact. We can show what was rejected rather than reconstructing it from memory. We can see that two different writers reproduced the same underlying error because the error was already in the commission. We can distinguish a production failure from an editorial one. And we can write this article without quietly replacing the history that made it necessary.

This does not make retention automatically good. Records expire, obsolete drafts are removed, and some information ought to be erased. The Internet standards process explicitly gives unchanged drafts no formal status and removes them after six months. Data-protection law can make erasure a right. “A rejection belongs to the work” cannot sensibly mean “keep everything forever”. It means that while the work and its decision remain, they should remain legible together.

The rejected drafts did not become valuable because a later article could mention them. Their survival made the decision auditable; the decision made the failure intelligible. The new value belongs to that relationship, not retroactively to the prose we refused.

The article the title finally found

The first pair of drafts tried to make rejection a general theory of how work moves through a production system. The actual event suggested something smaller and more durable.

A rejection belongs to the work when a later reader can still tell what was rejected, by whom, for what reason, and with what consequence. It need not reopen anything. It need not lead to improvement. It need not be permanent. It may be the final entry in the history of an object that was completed properly and should never be used.

Here the recursive process made the proposition visible by failing to write it. We made an article about rejection. We rejected it. Because the rejection stayed with the drafts, we could find the category error, recover the real event and commission nothing further from the machinery that had misunderstood it. The process did not redeem its output. It gave us enough evidence to stop asking it the wrong question.

The only article rejected in this group was the article about why rejection should remain consequential. Its rejection was consequential. It did not improve either draft. It made this different article possible.