Research brief: canine-longevity-ai-opportunities.md (landscape, ranked opportunities, and the Phase 1 plan in Section 7).
Phase 1 components:
Directory | What | Status |
| Complete: 3,565 records (2,185 core); full text for 1,140 of 1,347 PMCIDs (the other 207 are publisher-restricted from XML distribution). | |
| Complete: 30 offline tests, smoke-tested against live Ensembl/RePORTER and over stdio; DB built from the finished corpus, FOI and trial-registry layers. | |
| v0.1: drafted, validator-clean, then reviewed item-by-item by an independent LLM pass (9 fixes, 1 drop); recall@5 = 0.985. No domain-expert review yet. |
Phase 2 components:
Directory | What | Status |
| Complete: 497 records, 97% with parsed sections; structured layer covers all 497 (4,510 quotes, 0 errors, 79 explained warnings). | |
|
| Complete; 30 tests. |
| v0 (2026-10-01): drafted from corpus abstracts by model-assisted extraction plus hand-written company records; validator-clean; no domain-expert review yet. |
| docs/ | Static evidence site generated from the database by mcp/scripts/build_site.py: one page per NIA ITP compound and veterinary comparator (DrugAge rows, dog-equivalent doses, canine literature, FDA FOI summaries, gaps) and one per FOI ingredient (dose regimen, PK, target-animal safety, effectiveness, adverse reactions, each with its verbatim quote), plus llms.txt and a sitemap. No model-written text. | Live at https://w0lph.github.io/k9/ (219 pages); rebuild with cd mcp && uv run python scripts/build_site.py. |
.\rebuild.ps1 regenerates everything in dependency order (-Fresh re-fetches sources);
scripts/refresh.sh is the same from a clean checkout and is what the monthly GitHub Actions
workflow runs to republish the datasets, the database and the site.
PUBLISHING.md is the step-by-step for the Hugging Face Hub (dataset cards and
staging script in publish/), PyPI and the MCP registry (mcp/server.json), and Glama
(glama.json, root Dockerfile). The server installs with
uvx dog-geroscience-mcp (PyPI), as a Claude Desktop extension (.mcpb on the releases page),
or as a Claude Code plugin with a routing skill (/plugin marketplace add w0lph/k9, then
/plugin install dog-geroscience@k9; sources in plugins/), and downloads its database on
first run. Each directory is its own
uv project:
cd corpus && uv sync --extra dev && uv run pytest -q
cd mcp && uv sync && uv run dog-geroscience-mcp build && uv run pytest -q
cd questions && uv sync && uv run pytest -q && uv run cgq validate data/canine_geroscience_v0.jsonl --require-ids
cd foi && uv sync && uv run foi run && uv run pytest -q
cd mcp && uv run dog-geroscience-mcp build --skip-download # picks up ../foi/data/foi_summaries_dog.jsonlLarge derived data (corpus/data, mcp/data, FOI PDFs and text) is git-ignored; it is
rebuilt by the pipelines and published as Hugging Face datasets (publish/).
This server cannot be deployed
Canine genomics for agents: breed allele frequencies, AI pathogenicity + OMIA clinical disease layer
Pharmacogenomics for AI agents: cited drug–gene, gene and rsID lookups plus a medication screen.
Link compounds to protein targets, rank bioactivity, and look up drug mechanisms and indications.
Source-verified pet food regulations, recalls, nutrient standards and species care data.
