Found reviewing #241 (the files themselves landed earlier, in 6703180 / #240).
Two files sit under research/traits/ that are neither a report nor a citations sidecar:
research/traits/environment/aerobic-edison-literature-meta.yaml
research/traits/physiology/autotrophic-edison-literature-meta.yaml
They hold the resolved job metadata for a research call — job: LITERATURE, job_id: job-futurehouse-paperqa3, template_path, the full template_vars block, query_chars. Genuinely useful provenance, arguably more useful than what the report front matter carries.
The problem is that they exist for 2 traits out of 353
$ grep -rn "literature-meta\|literature_meta" scripts/ src/ templates/ .claude/ justfile
(no matches)
Nothing in the repo writes this filename. So either an earlier version of the research path emitted them and the emit was dropped — in which case 351 traits are missing provenance that was once considered worth keeping — or they are debris from a manual deep-research-client invocation that predates the sweep.
Both files also carry label: '', which suggests whatever wrote them was not finished.
Decide
Restore the writer and backfill, or delete the two. As with #245, the current state is the one state that is definitely wrong: a provenance artifact that exists for 0.6% of the corpus tells a reader the other 99.4% lost theirs.
Found reviewing #241 (the files themselves landed earlier, in 6703180 / #240).
Two files sit under
research/traits/that are neither a report nor a citations sidecar:They hold the resolved job metadata for a research call —
job: LITERATURE,job_id: job-futurehouse-paperqa3,template_path, the fulltemplate_varsblock,query_chars. Genuinely useful provenance, arguably more useful than what the report front matter carries.The problem is that they exist for 2 traits out of 353
Nothing in the repo writes this filename. So either an earlier version of the research path emitted them and the emit was dropped — in which case 351 traits are missing provenance that was once considered worth keeping — or they are debris from a manual
deep-research-clientinvocation that predates the sweep.Both files also carry
label: '', which suggests whatever wrote them was not finished.Decide
Restore the writer and backfill, or delete the two. As with #245, the current state is the one state that is definitely wrong: a provenance artifact that exists for 0.6% of the corpus tells a reader the other 99.4% lost theirs.