Reading list
What I'm reading toward
Solo research starts in fall 2026 and the topic is not settled. These are the sources I am working through to settle it. Every entry carries a DOI or an arXiv id, which the build enforces, so anything here can be looked up rather than taken on trust.
5 sources listed, 0 read so far. A read entry carries the full extraction. The rest are queued, with a note on why each one earned its place.
Comment staleness
Whether a comment still describes the code beneath it, and how often it stops doing so. The candidate both primers currently favour, mostly because it is tractable in two semesters.
Ratol and Robillard. ASE 2017.doi:10.1109/ASE.2017.8115624
Comments made fragile by identifier renaming. A narrow and precisely defined class, which makes it the most obviously implementable starting point of anything on this list.
Fluri et al.. WCRE 2007.doi:10.1109/WCRE.2007.21
The canonical co-evolution study, and the closest existing work to the thing I would actually be measuring. Worth reading early enough to find out whether the question is already answered.
Tan et al.. SOSP 2007.doi:10.1145/1294261.1294276
The origin point for automated comment and code consistency analysis. Combines NLP with program analysis over lock-related comments, which is the shape the whole subfield inherited.
Wen et al.. ICPC 2019.doi:10.1109/ICPC.2019.00019
Needed for the base rate. Any detector I build is only interesting relative to how often inconsistency actually occurs, and guessing at that number would undercut everything downstream of it.
Indirect prompt injection
Detecting instructions aimed at an AI browsing agent rather than at the reader. The candidate that follows on from Sneppard Sniffer.
Greshake et al.. AISec 2023.doi:10.1145/3605764.3623985
The paper that named indirect prompt injection and set out a taxonomy for it. Sneppard Sniffer was built against this threat model without my having read the source, which is the wrong order.