There is a version of this that every team recognizes. A report goes out with a number that was right six months ago. A board presentation cites a competitor's market position that the company quietly changed two quarters prior. A policy brief references a study that was later retracted.
None of these are negligence. They are research debt. The team knew the sourcing was imperfect when they shipped it. They made a judgment call that getting the work out mattered more than getting the sourcing airtight. And they were probably right, except the debt stayed on the books, and eventually someone collected.
What research debt actually costs
The cost of research debt is not the moment when bad sourcing ships. It is everything that happens after. The client challenge. The board question no one can answer cleanly. The regulator who wants to see the methodology. The competitor who noticed a number was wrong and said so publicly.
By then, the people who did the original work have moved on. The version that shipped cannot be told from the three drafts around it. The sourcing, such as it was, lives in someone's browser history.
"This is the pattern we see consistently," says Holly Corbett, Chief Strategy Officer at Bricolage AI. "Teams ship under pressure, they know the sourcing isn't perfect, and they tell themselves they'll clean it up later. Later never comes. And then the challenge arrives, and the clean-up costs ten times what the original rigor would have."
Why the debt accumulates faster now
Research debt has always existed. What has changed is the rate at which it accumulates.
Generic AI tools make it worse, not better. They produce polished, confident output that looks thoroughly sourced and is often not. The fluency hides the gap. So the team ships faster, with more apparent confidence, and the underlying sourcing is thinner than it was before. The debt compounds faster than it ever did when research was done manually.
The organizations that are furthest ahead on this are not the ones that have slowed down. They are the ones that changed what "good enough to ship" means at the foundation, so the output arrives already sourced, and the debt never accumulates in the first place.
Build it in, not on
The fix is not a stricter review process. Review processes slow the work down and still miss things, because manual checking is inherently incomplete. The fix is sourcing that is built into the research as it happens, not reconstructed afterward by someone trying to remember what they looked at.
"The architectural choice we made at Bricolage was to treat provenance as a first-class output, not a byproduct," says Chris Connors, Chief Technology Officer at Bricolage AI. "Every claim the system produces is linked to the source it came from at the moment it runs. The audit receipt seals the whole record. You cannot produce an unsourced claim and call it finished, the system won't let you. That's not a policy. It's how the pipeline works."
That distinction matters. A policy can be overridden under pressure. A pipeline that does not produce unsourced output cannot be overridden. The sourcing is there because the work cannot be completed without it.
The compounding effect of getting this right
Research debt is a liability that stays on the books until something forces the collection. Getting the sourcing right the first time is not just risk reduction, it is a compounding asset.
Every run that produces a verifiable, cited deliverable is a run the team can reference, quote, and defend forever. The work does not expire the moment it ships. It becomes a durable record that survives the questions that come months and years later.
The teams that understand this do not think of sourcing as a cost. They think of it as the thing that makes the work worth doing in the first place.
Bricolage AI delivers autonomous research with cryptographically verifiable audit trails. Every claim cited. Every run sealed. Learn more at bricolageai.com.