Topic mode vs lab mode¶

The toolkit has two front ends. They differ only in where the corpus comes from; once a verified set of papers exists, both run the identical verify → canonicalize → count → cross-reference → families → figure → review backbone.

	Topic mode	Lab mode
Start from	a question	a lab's publication corpus
Direction	search outward from the topic	derive themes inward, then place them in the field
Front-end phases	1 scope → 2 search → 2b antecedents	L1 ingest → L2 prune → L3 derive themes
Shared backbone	3 verify → 3f canon → 5/5b → 6 xref → 6b families → 7 review → 8	(identical)
The question it answers	"What's known about X?"	"What has this lab done, and where does it sit in the field?"

Topic mode¶

The default. You give a question and a span; the agent searches outward, runs the required antecedents pass for the roots, and the backbone takes over. This is the flow described throughout The pipeline and Phases in detail.

"I want a literature review on the anatomical connections between the visual system and the cerebellum — any anatomy papers, primate or human, any tractography method, back to the 1970s."

Lab mode¶

Lab mode inverts the front end: instead of starting from a query, it starts from a known body of work — a lab's output — derives the lab's research themes and how they shifted over time, and only then searches outward to place that work in the surrounding field.

L1 — Ingest the corpus¶

python3 ../tools/lab_corpus.py --search "Jack Gallant"          # find the OpenAlex id
python3 ../tools/lab_corpus.py --author A5056348548 --out lab_papers.json

tools/lab_corpus.py pulls a PI's authored works from OpenAlex (pass several --author ids for PI + key members to widen coverage). Each row carries title / year / DOI / APA / venue / citation count / topics / abstract / coauthors / type.

L2 — Prune false positives (human decision)¶

Author disambiguation is the #1 correctness risk in lab mode — OpenAlex will fold in same-name authors. You review the ingested list and drop the false-positives before anything is themed.

L3 — Derive themes and their drift¶

The agent derives the lab's research themes from the (abstract-enriched) corpus and tracks how emphasis moved over time. OpenAlex's light topic metadata alone is not enough — abstracts are enriched first. The themes become the lanes the rest of the pipeline uses.

Lab mode antecedents include the lab's own pre-paradigm work

The antecedents pass (still required) must surface the lab's own foundational work — e.g. macaque physiology that predates the lab's current human-imaging paradigm. An inclusion filter like "human fMRI only" must not silently drop those papers; they're re-entered as lab-sourced (and starred).

Then: the shared backbone¶

From verification onward, lab mode is identical to topic mode — every reference verified and canonicalized, counts fetched, cross-citations mined, and optionally a families figure and review article. The figure and review are contextualized: they show the lab's work against the field it sits in.

Lab corpus themes over time — L3 output — a lab's research themes and how their emphasis shifted across decades.

Lab work placed in the context of the broader field — The contextualized lineage figure — the lab's papers (highlighted) placed within the surrounding literature.

See the gallant_lab worked example for the full run.