Trust Learning Starts With a Prior
A laboratory study separated people’s opening expectations from how they adapted to partners who changed behavior.

Inside this article
01Two people can choose the same amount for different reasons ↓02The reversal matters more than the first impression ↓03A small commitment can produce a clear observation ↓Event / source date: 2 November 2026 — PNAS publishes the trust-learning study online · Published 12 NOVEMBER 2026
Two people can choose the same amount for different reasons
Two people join an unfamiliar collaboration and agree to put the same small amount of money into a first shared purchase. One expects a reliable partner but prefers to keep the first step simple. The other expects a broken promise and considers the small sum an acceptable loss. Their choices look identical on day one. What distinguishes them may become clearer when the partner delivers on time, misses a deadline or changes how they handle the money.
A study by Jonas P. Nitschke and colleagues, published online in PNAS on November 2, tried to separate these parts of trust. The researchers assessed lifetime stress through a structured interview and had 168 adults play a repeated trust game. Participants decided how many credits to transfer to partners whose responses were computer-controlled. After two baseline trials, one partner generally made more generous returns and the other kept more of the transfer. The partners changed roles without warning. That reversal gave researchers a way to compare the starting expectation with the response to new evidence.
The reversal matters more than the first impression
The paper reports an association between greater lifetime stress and lower initial trust. Lower investment was specific to the selfish trustee in the behavioral comparison. The computational model also found that greater lifetime stress was associated with more volatile beliefs about the initially generous partner, consistent with faster updating in that model. A change in modeled expectations and a change in the amount a participant transferred are separate outcomes. The study assessed stress retrospectively, and the authors describe the associations as modest. Its mostly young, healthy sample and programmed partners limit what can be inferred about individual behavior in daily life.
Imagine that the game ended immediately after participants met the initially selfish partner. A low transfer could reflect a low initial expectation, a strong response to the last disappointing outcome or a preference to risk little even after several fair outcomes. Reversal gives the model more information. If a player starts cautiously and increases transfers after repeated good outcomes, the history shows room to update. If a player starts at the same level and makes little change through both phases, the resulting trajectory suggests a different decision process. The model still represents choices in the task and depends on the assumptions built into it.
A small commitment can produce a clear observation
Consider a practical example in which two departments need to share responsibility for a customer handover. One team has experienced repeated missed handoffs elsewhere and keeps a second person on standby. Asking it to remove that backup immediately raises the cost of a miss. A clear first commitment could be sending a complete, checked file by noon on Thursday, with a named person responsible and a place both teams can verify the delivery. The backup stays in place until the team sees enough work to reconsider it.
The important feature is the trace left by the commitment. If a file arrives late because the customer supplied information after noon, record that reason and check when the other team disclosed the delay. If the file arrives on time with a missing attachment, the agreed checklist shows what was incomplete. An unexplained late handoff, an early disclosure of a blocked handoff and a complete handoff are different observations. Teams can preserve those distinctions by writing down the expected behavior before the deadline, while it is easy to agree on what each person has promised.
Make it possible to revise in both directions
Suppose the partner meets the deadline three weeks running, including one week when someone was absent. The cautious team can lower a single part of the backup burden and leave the rest in place. If the next deadline slips, both sides can inspect the new event and decide whether the original arrangement should return. A reversible adjustment lets evidence affect the size of the commitment while keeping a workable fallback.
Asking a partner to prove themselves through a hidden loyalty test would change the relationship being observed. In ordinary collaboration, make expectations visible and allow a partner to explain legitimate exceptions. Keep a history long enough to distinguish a change in incentives from an unlucky day. The laboratory reversal was a tool for measuring adaptation. Deliberately creating a breach in a real project would give both parties a reason to distrust the person who set the test.
Separate the belief from the decision to take risk
A team might believe a colleague is likely to deliver and still keep an inexpensive backup because a missed deadline would harm a customer. The same team might accept a greater probability of error on a small reversible assignment. A trust judgment and the size of an authorized exposure answer different questions. Keeping them distinct helps explain why one person receives more access to one workflow while another remains subject to routine checks in a higher-stakes setting.
CYPAPER’s discussion of restoring access addressed the choices that follow a damaged relationship. The laboratory work examined what can be learned from a sequence of choices as evidence changes. The practical question for an ongoing collaboration is what observation would justify moving one safeguard, by how much, and for which part of the work. Stating that threshold while there is time to plan makes it easier to recognize a real change when it happens.
Gabriel Nicolaev — CyGuru
Sources & further reading
University of Vienna SCAN Unit research summary ↗
Nitschke and colleagues’ open-access PNAS study ↗
A move in a game of chess. Illustrative photograph by Elias Derksen / Unsplash.
Another angle on this subject
Continue through related ideas, decisions and consequences.
Human BehaviourA Memory Reward Effect Does Not Identify Its Mechanism
A study of reward-linked learning asks whether deliberate rehearsal after learning explains the later memory advantage.
Human BehaviourDisappointment Becomes Betrayal Only When a Real Commitment Was Broken
A detailed ethical essay on expectations, promises and relationships: name the commitment before treating an unwelcome outcome as proof that trust was violated.
Human BehaviourForgiveness Does Not Decide Who Receives Access Again
A reflective essay on separating release from renewed trust, and making a future relationship depend on the conduct it can responsibly support.
