B3 · Coding: fleet

Read the material independently, twice

B3 applies the shared scheme across the analysed communities. Two independent readers examine every local theme, making differences in interpretation visible before the aggregator settles them.

Where this stage starts

B2 has frozen the categories and tested their instructions. B3 receives that scheme and all the local themes, with the concepts and participants’ words beneath them. A “reader” is a separate model conversation. The collection of reader jobs is called the fleet.

What the aggregator and handler do

The aggregator prepares batches that keep each community together. A large community may require several files, but one reader receives all its parts. Its second reading goes to a different reader. The allocation connects communities and readers across batches, allowing the analysis to examine whether particular readers or batch contexts produce different answers.

Each reader gets the scheme and its assigned evidence. It can see the labels of the community’s other themes for context, but not another reader’s answers, the aggregator’s notes or the canvas outcome. It assigns themes from their underlying concepts rather than relying on a community coder’s label alone. If one category does not capture the material, multiple placements need reasons.

The handler runs the jobs separately. The aggregator then checks what returned: whether every required item was read twice, whether replies are complete, and whether all category references are valid. A missing or truncated reply is a delivery problem, not evidence that two readers disagree.

Reference items measure how closely readers follow the author’s intended distinctions. Further checks examine changes across a batch and variation among readers. These results reveal how consistently the instrument works. They cannot establish that the categories are the best explanation of organizing.

What the next stage receives

B4 receives both readings, their reasons, and a record of coverage and uncertainty. B3 preserves disagreement instead of resolving it by majority vote. B4 must return to the evidence to decide what each disagreement means.