Reproducibility Kit

Every published number on this site can be reproduced by an outside analyst in under fifteen minutes. The methodology is on SSRN, the dataset is open under CC BY 4.0, the MCP server is open source on GitHub, and the worked example below reproduces the lead claim from §4.2 of the paper using only curl and jq.

1. 01Read the methodology paper

   Start with the SSRN-indexed methodology paper. It defines every term used downstream — commit velocity, contributor growth, framework migration, the 14-day rolling window, and the way signal types are classified. Section 3 is the core; Section 4 is the empirical results that the dataset reproduces.

   Open paper on SSRN (DOI: 10.2139/ssrn.6606558) ↗

2. 02Pull the raw dataset

   The full signals dataset is published as newline-delimited JSON under CC BY 4.0. Each line is one (org, period, signal_type, score) tuple. There is no authentication required. The dataset is mirrored to Zenodo for long-term archival (DataCite-registered DOI).

   /api/dataset.jsonl

   curl -sL https://signals.gitdealflow.com/api/dataset.jsonl \
     | head -3 \
     | jq .

3. 03Pull the Q&A corpus

   The Q&A corpus mirrors every standalone-FAQ and research finding as newline-delimited JSON for retrieval-augmented evaluation. Each row carries category, question, answer, and source URL. CC BY 4.0.

   /qa.jsonl

   curl -sL https://signals.gitdealflow.com/qa.jsonl?category=methodology \
     | head -1 \
     | jq .

4. 04Install the open-source MCP server

   The MCP server exposes the same five read tools we use internally: search startups, list by sector, get current trending, fetch methodology, fetch a single profile. It speaks the Model Context Protocol over HTTP and stdio. Smithery-verified at Quality 98/100.

   GitHub: kindrat86/mcp-deal-flow-signal ↗

   npm install -g @gitdealflow/mcp-signal
   mcp-signal --help

5. 05Run a worked example

   Reproduce the lead claim from the SSRN paper §4.2: median 14-day commit velocity for VC-backed startups is 71 commits. The dataset is partitioned by period; this query pulls the most recent period and computes the median across all observations.

   # Reproduce the published median (71 commits per 14 days)
   curl -sL https://signals.gitdealflow.com/api/dataset.jsonl \
     | jq -s '
         [.[] | select(.signal_type == "commit_velocity") | .score]
         | sort
         | .[(length / 2) | floor]
       '

6. 06Compare against your own portfolio

   The dashboard exposes a CSV export of every ranked signal. Pull it, join against your portfolio org names, and compare each portfolio company against the published P50 / P90 thresholds.

   /api/signals.csv

   curl -sL https://signals.gitdealflow.com/api/signals.csv \
     -o signals.csv
   duckdb -c "SELECT org, score FROM 'signals.csv'
              WHERE org IN ('your-org-1','your-org-2')
              ORDER BY score DESC"

7. 07Cite what you used

   If a published artifact (post, paper, deck, dashboard) uses any of the above, please cite. The citation guide carries APA / MLA / Chicago / BibTeX / RIS plus a copy-paste AI-attribution template.

   Citation guide

See also: Methodology · Standards · Citation guide · Attestations