The Work Changes the Question

When we began developing HALO Journal, we had three ideas about what our work might have taught us.

One concerned permission: a system that can do something should not automatically be allowed to do it. Another concerned evidence: what we say we know should remain within what our sources support. The third concerned results: completing one stage of a process does not prove that the next stage happened.

These became three proposed doctrines. Before presenting them as something HALO had earned the right to say, we asked the research to examine the projects behind them.

The trouble with the third became clearer when we tried to explain how it differed from the second.

Suppose a system finishes a local operation and sends something to another system. That can establish completion of the local work. It cannot, on its own, establish that the other system stored what was sent. We had treated that distinction as a separate concern about proof at successive stages. But it was also an instance of the evidence rule: the stronger statement needed support the local result did not supply.

The distinction was useful. Its independence was harder to defend. A third heading had given it a place in the proposed architecture without establishing that it needed a separate doctrine.

Other parts of the research put pressure on the wording. In a demonstration, reconstructed dates could help arrange material on a timeline. They could not recover the publication history of that material. Several reports could describe the same connected program without becoming several independent confirmations. And a rule against acting before uncertainty was resolved would rule out experiments whose purpose was to investigate something unknown.

These were consequential differences for the Journal we were trying to make. We could not use the archive simply to populate three persuasive sections. The examples were changing what the sections could reasonably contain.

The September assessment returned two narrower concerns. We adopted them as Governed Action and Warranted Representation: how consequential action is permitted, and how claims remain answerable to their evidence. The concern about proof at successive stages became an application of the second. Permission remained distinct because knowing something and being authorized to act on it answer different questions.

The resulting language also allowed for deliberate, bounded experiments under uncertainty. An experiment could have an owner, limits and conditions without having a known outcome. That mattered more than preserving a broad formulation which sounded rigorous until it prohibited the very inquiry needed to learn anything.

There was a further change in what we could claim about ourselves. The names recorded an interpretation we were adopting in September. They did not establish that one unified philosophy had consciously governed all the earlier work. Some connections were visible to us afterward. They could still be worth making public, provided we did not rewrite their discovery into the beginning of the story.

Writing helped expose this problem. A topic such as evidence leaves considerable room to agree in principle. A sentence about a particular result is less accommodating. It makes us choose whether the thing was observed, reported, inferred or adopted. Put that sentence beside its source and the relation can become clearer, sometimes by disappearing.

Our material was documentary, but the experience has something in common with the design work Donald Schön examined. In his account of an architectural design episode, a change made for one purpose revealed other spatial possibilities that the designer then evaluated. The consequences helped develop the intention. The design was doing more than implementing a complete thought brought to it at the outset.

That is an appealing way to understand making. It is also a convenient story for someone who would rather revise the question than accept an unwelcome answer.

A project can acquire remarkable resilience that way. If the original claim fails, its failure becomes proof that a more sophisticated inquiry was required. If the new claim fails, perhaps the real discovery was the difficulty of asking it. The work keeps producing explanations for its continuation while surrendering very little.

I do not think calling the process reflective protects us from that. The useful challenge is to identify what the revision actually costs the original argument.

In this case, we gave up the claim to three independent doctrines. We could not represent a present synthesis as an established historical worldview. We also had to leave examples in the account that qualified the proposed principles, rather than treating every project as positive confirmation. A version of the Journal that quietly restored those claims would undo the work of the investigation, however polished the prose became.

Even those concessions did not validate what remained. In our account of revising Homebridge's replacement estimate, removing unjustified scope made the model smaller without independently establishing its surviving assumptions. The same objection applies here. Two narrower concerns can still be incomplete or unhelpful beyond the situations that suggested them.

The purpose of the undertaking matters. If our task had been to demonstrate that HALO had always operated under three established doctrines, we would have to report that we had not done it. The Journal's purpose was to make public what the work allowed us to say. Discovering a weakness in the arrangement was therefore material to the publication, even though it frustrated the original shape of the argument.

That does not make every revised question a better one. It gives this revision something to remain accountable to. We still owe the reader an account of the work, including the limits that changed our interpretation. We cannot substitute a more attractive question merely because it is easier to answer.

There was no need to restore the third doctrine for symmetry. Nor did we have to abandon interpretation and offer only a chronology. We could make connections while showing when and how we had come to make them.

The third doctrine still has a place in that account. It records a distinction we thought deserved independence, then found more useful inside another concern. Keeping the change visible says more about how the Journal acquired its argument than presenting two immaculate principles with no history of their own.

See also

• A coherent plan for the wrong problem — Offers a separate case in which reconsidering the problem narrowed the work, without proving the resulting product was complete.