Position: what is known, what is not, and what this repository adds
Step 4 of the review. It keeps "nobody has done this" apart from "we have not done this here". Terms are in frame.
Shape: position
What the field already answers
- How to report one case: the guidelines of Runeson 2009 and the Empirical Standards (Ralph 2020). Convention.
- How an experience report should read: a take-home message, both outcomes, the authors' roles (ICSE 2003). Convention.
- How a history misleads: the perils of Kalliamvakou 2014, the edits of Bird 2009, the dataset bias of Bird 2009b. Measured.
- How a postmortem is run and written (Collier 1996; Beyer 2016). Convention.
What the review did not find
| Gap | Said as | Grade |
|---|---|---|
| A guideline for a case written by an AI-assisted team about its own day | not found in 42 sources; not shown absent | supported with caveat |
| A measure of the questions that reached a person in a working day | not found in this pool; the company posts do not count interventions | supported with caveat |
| A study of several agent sessions coordinated in one day | not found in this pool | supported with caveat |
| A test that a checklist makes a case more trusted | not found | convention, untested |
- "Not found" is a statement about this search. The protocol lists what was not searched.
What this repository adds, and under which conditions
The repository asks what the field does not answer here: the cost of a day of the method, measured where a program can measure it, and said by hand where not.
- A case is an experience report with repository measures, written from inside, and says so.
- A case meets the reader's checklist of Runeson 2009, and states for each of its 12 items where it is met or why not.
- A program takes every count it can from the history and prints its own cautions beside the number; a count by hand is marked as such.
- A language model analyses no data of a case; it may draft the prose, which a person reads.
- A case names its comparison: the method's earlier cases or a contrast inside the case, never a control it does not have.
- These are choices, graded convention. Conditions 3 and 4 follow the Repository Mining and Case Study standards (Ralph 2020).
- Where practice runs ahead of evidence: company posts report sessions and cost with no checklist. Where evidence is ignored: the perils of mining a history are rarely printed beside a count.