Track a tool's hype curve across any Substack (no API key)

The Stack

Tested Against

Prerequisites

• node
• A Substack publication to point it at, for the live step

Steps

1. 1

   Verify the mention-over-time analyzer (deterministic)

   The core is a counter: given dated posts, count how many per month mention a term. CI runs it on a fixture where the term peaks early then goes silent, the shape of a hype curve, and asserts the exact monthly timeline. No network, no key.

   cat > posts.json <<'JSON'
   [
     {"post_date":"2026-02-10","text":"Everyone is trying AutoGPT this month, it is all anyone writes about."},
     {"post_date":"2026-02-22","text":"More AutoGPT experiments, the autonomous agents hype is peaking."},
     {"post_date":"2026-03-05","text":"AutoGPT is cooling off already, people are moving to the next thing."},
     {"post_date":"2026-05-01","text":"Claude and MCP are everywhere now, the conversation has fully moved on."},
     {"post_date":"2026-06-09","text":"MCP servers keep multiplying; nobody brings up the old tool anymore."}
   ]
   JSON
   node -e '
   const term = (process.env.TERM_Q || "AutoGPT").toLowerCase();
   const posts = JSON.parse(require("fs").readFileSync("posts.json", "utf8"));
   const byMonth = {};
   for (const p of posts) {
     if (p.text.toLowerCase().includes(term)) {
       const m = p.post_date.slice(0, 7);
       byMonth[m] = (byMonth[m] || 0) + 1;
     }
   }
   const line = Object.keys(byMonth).sort().map((m) => m + ":" + byMonth[m]).join(" ");
   const expect = "2026-02:2 2026-03:1";
   if (line === expect) {
     console.log("analyzer OK: AutoGPT mentions " + line + " then silent (a hype curve that peaked and faded)");
   } else {
     console.log("BAD timeline: " + line + " (want " + expect + ")");
     process.exit(1);
   }
   '

2. 2

   Run it on a real Substack (the live step, not checked by CI)

   Point it at a publication: fetch the archive list (https://PUB.substack.com/api/v1/archive?sort=new&limit=25&offset=0, paginate with offset), then each post (/api/v1/posts/SLUG) for full body_html, strip tags, and feed the dated text through the same counter with your TERM_Q. The data changes and the endpoint throttles bulk pulls, so CI never claims a live result. Remember it is coverage over time, not subscribers.