Where does this loop end?
That was the first question in the review of a marathon's results flow. After the course closes, the timing contractor works through every runner's chip readings: compute the time, check the runner against the athletics federation's eligibility rules, add the time to the results, repeat. On the drawing this was a Merge, two Process boxes and a Decision titled "More?" whose yes edge climbed back up to the Merge. The logic was right. But to learn what it repeated over, the reviewer had to find the back edge and guess from a one-word label, and a loop over 12,000 runners looks just like a loop that never stops.
A Loop Limit marks the boundary of a repeating section, naming the condition that ends it: here "For each runner" (it ends when no runner is left), written on the symbol, where the reviewer was already looking. The review asked two more questions, and each has a symbol of its own. Of the three, the Loop Limit and the Predefined Process are ISO 5807 flowchart symbols; the Data object is borrowed from BPMN.
The three symbols, and the Subprocess that a Predefined Process is easy to confuse with.
Loop Limits usually come in a pair (ISO 5807, which lets the condition sit in either one, depending on where the loop tests it). Here the opening one holds it: "For each runner", or "Until it works, max 3 tries" where success can end the loop before the cap. The steps between the two are the body, which repeats; the closing one ("Next runner") marks where each iteration ends. The flow enters through the opening symbol and leaves through the closing one once that condition ends the loop, so no loop-back edge is drawn. It is not a Merge either: a Merge rejoins different paths, while the pair brackets one path that repeats. In RodGrid, place the same Loop Limit node twice with the body between them (one input and one output each); a single one in front of the body is also accepted.
Bounding a loop with a plain Decision and a loop-back edge when a Loop Limit would name the bound is the readability mistake this review caught: "More?" left the reader to reconstruct a fact the symbol could have stated. The pair has limits of its own. A body that branches has to meet at a Merge before the closing symbol, which takes one incoming edge. And with one way out, the pair cannot by itself send "it worked" and "we ran out of attempts" to different places; Loops & retries draws the loop that can, and why a retry needs a cap.
Whose box is the eligibility check?
On the old drawing the check was a plain Process, and the reviewer asked to see its steps. The contractor had none to show: the rules (age categories, a chip that missed the 30 km mat) belong to the federation, and every race it sanctions runs the same routine. A Predefined Process (a double-railed box) says exactly that: it calls a procedure defined elsewhere, like invoking a named subroutine instead of inlining its steps.
Use Predefined Process when the routine is defined elsewhere and shared: this box only calls it (vs. Subprocess, whose detail belongs to this flow alone). The ISO symbol itself asks only that the steps be specified elsewhere; shared is the usual case, and the line these articles draw against Subprocess. Computing a runner's time could grow into a dozen of the contractor's own steps, used nowhere else; collapsed into one box, those are a Subprocess. Call them a Predefined Process and the reader goes looking for a definition elsewhere that does not exist. Subprocesses & reusable steps covers collapsing and nesting.
What does the results step receive?
Each pass through the routine adds a time to a list or records why the runner was rejected. When the loop ends, Publish results needs that list, not the raw readings. A Data object (BPMN's folded page; the flowchart 'data' parallelogram is Input/Output here) marks information the flow reads or writes; Data, documents & manual input covers Input/Output. "Valid times" documents what moves between the two steps. In RodGrid it is an in-line node with one input and one output, so place it on the path where that data is handed over.
A Data object names what is handed between steps; where that data is kept past the run is Stored Data, which Stored data, displays & manual steps covers. Add one for every value (each split, each age category) and the one hand-over that matters disappears. CSV Import & Validation has this symbol in its palette, with a Loop Limit around rows that branch valid or invalid.
Two names in a backfill
The Data Backfill Migration challenge has no Data object in its palette, but it needs the other two symbols: bound the batch loop with a Loop Limit pair ("For each batch" … "Next batch"), and call validation as a Predefined Process after the last batch. A good run makes the loop's exit condition and the validation call both readable without opening a node.