Dalhousie In-House Conference Archive (1975–2026)

A reproducible pipeline that turns ~50 years of Dalhousie Department of Psychology & Neuroscience In-House Conference programs into a single structured BibTeX archive, plus tooling and a 50th-anniversary talk built on the resulting data.

The conference was founded in 1975 and has been the Graham Goddard In-House Conference since 2011 (named for Dr. Graham V. Goddard, 1938–1987, who developed the kindling model of epilepsy at Dalhousie). The 2020 and 2021 conferences were not held due to COVID; 2026 is the 50th.

What's in this repository

Source data (the hard-to-replace stuff)

Path	Description
source_programs/In-House Program YYYY.{pdf,doc,docx,rtf,txt}	One program per year, 1975–2026 (no 2020/2021). Rescued from filing cabinets and email archives by Richard Brown, Suzanne King, and Susan Lowerison.
corrections.jsonl	Human-curated patch/add/delete operations applied on top of the parser output.
descriptions of specific issues to fix.md	Reviewer-supplied breakpoint markers used by the entry-splitter to disambiguate records the OCR fused together.
slides/figures/{GrahamGoddard_photo.jpg, Hebb.jpg, klein_graphic.png, program_1975*.{jpg,png}}	Hand-curated source images embedded in the slide deck.

Pipeline (regenerates everything else)

Path	Stage
pipeline/01_extract.py	Source files → extracted/YYYY.txt (one UTF-8 text file per year).
pipeline/02_parse.py	Extracted text → records.jsonl (structured records, one per presentation).
pipeline/03_export_bib.py	Records + corrections.jsonl → inhouse_conference.bib + qa_report.md.
pipeline/04_diagnose.py	Records → diagnostics.md (per-record QA flags).
pipeline/05_review.py	Records + diagnose → review_needed.bib (subset for human review).
pipeline/06_diff_review.py	Diff a hand-edited reviewed_and_fixed.bib back into corrections.jsonl patches.
pipeline/07_split_entries.py	Apply reviewer-supplied breakpoints to split fused records.
pipeline/booktitles.py	Per-year conference title lookup.
pipeline/run_pipeline.sh	Convenience runner for the three core stages.

Final output (tracked for convenience)

• inhouse_conference.bib — 2,155 @inproceedings entries covering 50 conferences.

Talk deck

Path	Description
slides/slides.md	Marp markdown source for the 50th-anniversary talk.
slides/analyze.py	Reads inhouse_conference.bib, writes slides/figures/0[1-5]_*.png and slides/stats.json.
slides/figures/	Source images committed; generated figures gitignored.

How to rebuild from scratch

# 1. Run the pipeline (extract → parse → export)
bash pipeline/run_pipeline.sh

# 2. Generate the talk figures + stats
python3 slides/analyze.py

# 3. Render the slide deck (requires marp-cli)
cd slides && npx --yes @marp-team/marp-cli@latest --allow-local-files --html slides.md -o slides.pdf

extracted/, records.jsonl, qa_report.md, diagnostics.md, and the generated figures are deliberately not tracked — they are pure functions of the source files + pipeline scripts and can be rebuilt in ~30 seconds.

License

• Code (pipeline/, slides/analyze.py) — MIT, see LICENSE-CODE.
• Data and program scans (source PDFs/docs, inhouse_conference.bib, corrections.jsonl, image files) — CC BY 4.0, see LICENSE-DATA. Original copyright in each abstract remains with its author(s).

Provenance

See CLAUDE.md for the detailed format-group breakdown (14+ distinct parser dialects across the 50-year corpus) and SESSION_HANDOFF.md for the state of the cleanup effort as of May 2026 (~10% of records still flagged for review).