Glueguy Lab

Spreadsheet-style adhesive selector plus a larger McMaster-derived engineering reference library.

Run locally

python3 -m http.server 4317 --bind 0.0.0.0

Open http://127.0.0.1:4317/ on the host, or from another machine on the same LAN/VPN open http://<host-ip>:4317/.

Catalog pipeline

1. Crawl McMaster glue families into raw JSON.

python scripts/mcmaster_glue_pipeline.py --max-pages 220 --output data/mcmaster-glues.json

2. Enrich the representative McMaster part page for each family into a reusable detail cache.

python scripts/enrich_mcmaster_product_details.py

This enriches the representative McMaster part page for each family into data/mcmaster-product-details.json, adding higher-fidelity fields like part numbers, package data, mix ratio, coverage, consistency, peel, For Use On, and direct McMaster spec-page URLs.

3. Transform the raw crawl plus the detail cache into site-ready selector products and reference families.

python scripts/build_mcmaster_site_catalog.py

4. Build the manual TDS-backed additions that are not yet covered by the McMaster crawl.

python scripts/build_tds_manual_catalog.py

5. Build the compact first-load selector catalog.

python scripts/build_selector_catalog.py

6. Build the minified browser runtime used by index.html.

npx --yes terser app.js -c -m -o app.min.js

7. Build the machine-readable catalog intended for LLM/tool agents.

python scripts/build_agent_catalog.py

This writes data/agent-catalog.json, a compact JSON endpoint with source URLs, cached TDS text paths, exact specs, missing-field lists, and agent notes. Agents should treat missing fields as unavailable, not infer them from profile defaults.

8. Normalize any Digi-Key electronics adhesive export into distributor/TDS leads.

python scripts/import_digikey_adhesive_applicators.py

9. Run the autonomous coverage audit and backlog report.

python scripts/run_glue_autoresearch.py

This writes:

• data/mcmaster-glues.json
• data/mcmaster-product-details.json
• data/mcmaster-site-summary.json
• data/mcmaster-site-catalog.js
• data/tds-manual-catalog.js
• data/selector-catalog.json
• data/selector-catalog.js
• app.min.js
• data/agent-catalog.json
• data/autonomous-discovered-products.json
• data/digikey-electronics-adhesives.json
• data/autonomous-research-report.json
• data/autonomous-research-report.md
• data/tds-extraction-suggestions.json
• data/tds-extraction-suggestions.md
• data/autonomous-results.tsv

10. Mine cached TDS text for reviewable field candidates when working the manual backlog.

python scripts/extract_tds_field_candidates.py

This reads data/tds-cache-manifest.json plus the current missing-field audit and writes source-snippet-backed candidates for viscosity, electrical, mechanical, cure, service-temperature, and thermal fields. It deliberately does not edit data/tds-manual-source.json; the output is a ranked verification queue that makes exact TDS transcription faster without inventing values.

The cache pipeline is Mistral-first for PDFs when an OCR sidecar exists: scripts/cache_tds_sources.py prefers data/tds-cache/<id>.mistral.md over pdftotext and records textExtraction: mistral-ocr-markdown in data/tds-cache-manifest.json. pdftotext remains only as the fallback for entries that do not yet have Mistral OCR output.

To import a Mistral OCR MCP result into the cache sidecars:

python scripts/import_mistral_ocr_result.py <entry-id> --input /path/to/mistral-result.json

To generate OCR sidecars directly from cached PDFs using the local Codex MCP Mistral credential:

/home/scandium/.local/share/mcp-mistral-ocr-venv/bin/python scripts/ocr_tds_with_mistral.py --limit 25
python scripts/cache_tds_sources.py
python scripts/extract_tds_field_candidates.py

To print the next highest-impact products with candidate fields:

python scripts/tds_gap_queue.py --electronics-only --limit 25

Verification

python -m py_compile scripts/enrich_mcmaster_product_details.py scripts/build_mcmaster_site_catalog.py scripts/build_tds_manual_catalog.py scripts/build_selector_catalog.py scripts/build_agent_catalog.py scripts/discover_official_glue_products.py scripts/autonomous_glue_research.py scripts/mcmaster_glue_pipeline.py
python -m py_compile scripts/cache_tds_sources.py scripts/extract_tds_field_candidates.py scripts/import_mistral_ocr_result.py scripts/ocr_tds_with_mistral.py scripts/tds_gap_queue.py
node --check app.js
node --check app.min.js

Autonomous loop

• program.md defines the autoresearch-style operating loop for this repo.
• data/autonomous-research-seeds.json defines target manufacturers and products.
• data/autonomous-discovery-config.json defines which official sitemap surfaces should be mined automatically.
• scripts/discover_official_glue_products.py expands the backlog from official manufacturer surfaces.
• scripts/run_glue_autoresearch.py executes one measurable discovery-plus-audit experiment and appends it to data/autonomous-results.tsv.
• scripts/autonomous_glue_research.py merges curated seeds plus discovered official leads into a ranked backlog report.

Current pipeline snapshot

• 428 McMaster-derived selector products
• 503 McMaster reference families
• 11 manual TDS-backed entries
• 298 McMaster families matched to an official TDS
• 42 McMaster families with product-page-only fallback
• 881 observed McMaster package offers