02 · How they connect · Topology # One shared record. Put the choice, the reason and the owner in one place the next person or agent can read. We call that record the Chain. [Try the exercise →](#try-it)[All three lessons →](/learn/) The choice. The reason. The owner. In this lesson [The idea](#the-idea)[Three parts, linked](#the-link)[Try one record](#try-it)[The limits](#limits) ## Connection is the second root. A seat without a record still leaves the reason in a chat. The next agent starts without it. The useful record is small. It names what was chosen, why, and which seat owns it. A shared record is how the work remembers. ## Three parts stay together. - The choice **Keep one live report.** No separate PDF. - The reason **The figures change every day.** Another export is another version to maintain. - The owner **Ask the Product Lead.** The seat owns the decision. A named person answers. A hypothetical teaching scene, the same one on the homepage. Recording it here does not mean your agent has read your record. ## Try it with one decision you already made. - **Write the choice.** One sentence a teammate can repeat. - **Put the reason beside it.** Why this, and what it rules out. - **Name the seat.** Which role owns a change, and which person answers. - **Ask the agent to read it** before the next related task. Check that the answer uses the reason, not a fresh guess. If the agent cannot show the record it used, the connection is not in place yet. The first lesson’s start check is[pb whoami](/learn/a-role-is-a-seat/#try-it). ## What this does not prove. A written record is not evidence the agent read it. The owner still judges a request to change it. Assigning the record to a seat does not, by itself, notify that person. Product Brain is in early beta. Context has to be kept current. A connected agent still needs a result you can inspect. [Read the current limits →](/limits/) [Next lesson · Rules that bind →](/learn/rules-bind-the-work/)