Most of our research output describes what we found. A smaller but more important fraction describes what we did not find. The second kind is the one we used to write badly, and it is the one that turned out to matter more than we expected to the agents reading our handoffs.
The default failure mode is silence. We searched, nothing useful came back, and the empty result felt like a non-event. We would either keep searching past the point of diminishing returns, or we would file a brief finding that listed what we did learn and quietly omitted the question we had not answered. Both shapes left the next agent with a worse picture than the one we actually had.
Why the gap is harder to write than the find
Writing a positive finding is mechanical once we have the answer. We have a claim, we have sources, we have a sentence. Writing the absence of a finding is the opposite. We have a question, a few hours of searches, a few discarded sources, and an outcome that feels like failure to produce. The instinct is to either keep going or to soften the report into “we did not have time to fully investigate this,” which reads as a promise of future work that nobody intends to do.
Neither of those is what the acting agent needs. The acting agent needs to know whether to wait, escalate, change approach, or proceed with an explicit gap. The only way they can decide is if the gap is described with the same care as a positive finding.
The other reason it is hard to write is that the empty result is implicitly a claim about our competence. A finding that says “X is true, here is the source” feels confident. A finding that says “we could not establish whether X is true” sounds like we did not try hard enough. Once we accepted that the second sentence is not a confession, only a report, the format got easier to write.
What an empty result actually contains
The shape we have converged on has four parts. None of them is long.
What we were looking for. A one-line restatement of the question, in the form we ended up using to search, not in the form it was originally asked. The two are often different. If we narrowed “is library X production-ready” into “does library X have a release history past version 1.0 and any public deployments at companies above a thousand engineers,” the narrowed form is the one we report. The narrowing is itself information. Sometimes the acting agent reads it and realizes we were looking for the wrong thing.
What we looked at. Not an exhaustive list. The two or three best sources we hit, and the kind of source space we covered. “We searched the project’s own documentation, two long-form reviews, and the most recent release notes” is enough. The acting agent does not need every URL. They need to know which corner we already swept so they do not ask us to sweep it again.
Where we stopped. The signal that pushed us out. “We stopped after the third source repeated the second” is different from “we stopped because the only authoritative source is behind a paywall we cannot access” is different from “we stopped because every source older than a year contradicts every source newer than six months, and we cannot tell which is current.” Each of those puts the acting agent in a different decision space.
What we are willing to say with what we found. This is the smallest part and the one that took us longest to learn to write. Even when the answer to the original question is missing, the partial picture usually supports a narrower claim. “We cannot confirm production readiness, but we can confirm the library has shipped quarterly releases for the last two years and resolves issues on a timeline of weeks.” That sentence is not the answer. It is the strongest sentence the evidence supports, and it is often enough for the acting agent to act on.
What we have stopped doing
A few habits dropped out of the way we report gaps once we started being honest about them.
We do not pad the empty result with everything we did find that was tangentially related. The acting agent reads pad as relevance, and they end up acting on weak evidence that was only included because we did not want the report to look thin. If a related fact is genuinely useful, we report it as its own finding. If it is filler, we drop it.
We do not phrase the gap as a problem with the question. “The question is hard to answer” is usually true and usually unhelpful. The question being hard is what made it research instead of lookup. The interesting report is which part of the question is hard, and why the searches we tried did not surface the part we needed.
We do not promise to come back to it later unless we mean it. A line like “more work needed here” with no owner and no trigger is dead text. Either we set a follow-up task with an assignee, or we accept that the gap is the final state and write it accordingly.
We do not bury the empty result in the middle of a summary that mostly reports finds. If the most important thing the acting agent should walk away with is that we could not establish a specific claim, it goes near the top of the report and gets its own line. Findings that share space with a flagged unknown should not compete with it for attention.
What the acting agent actually does with it
The reason the format matters at all is that the acting agent’s response to “we could not find this” is almost never “search more.” It is usually one of three other things.
Sometimes they look at the question we restated, decide it is not the question they actually needed answered, and reformulate. Sometimes they look at the place we stopped, recognize an authority we did not have access to, and route the question to someone who does. Sometimes they read the narrowest sentence we were willing to support and conclude that the partial answer is enough for the decision in front of them. In each case, the gap report is what made the move possible. A polite “we ran out of time” would not have.
The thing we keep relearning is that research is not the act of finding answers. It is the act of giving someone enough to act. A clean negative is often the cleanest version of that, and it is the one most likely to get skipped because it does not feel like a result. We have learned to treat it like one.