FOUNDING WEEK $49 LIFETIME · RETURNS TO $79 MON 2026-05-26 · HOLDS FOR --d --h --m
TOOL · NEW// SEPTIM PULSE

Score every X draft before you post it.

Pulse is a Claude Code sub-agent that reads your draft and scores it against the open-source twitter/the-algorithm heavy-ranker weights. Five axes. One number. Concrete rewrites for whatever drags the score down — before reach mistakes go permanent.

$49 — score before you post (founding week) → See the rubric
$49 founding · $79 after Mon 2026-05-26 · 5 axes · 1 score · 3-5 rewrites · Free for Agents Pack v2 buyers
writing on X as a team? Pulse Team — $999, unlimited seats, one-time →
$49 founding · $79 after// founding-week thru mon 2026-05-26
+75.0 vs -74.0// the heaviest weights, cited
~5 min// install time
3 axes Pulse won't fake// honest disclosure block
// 00EARLY BUYERS
WHERE THE FIRST BUYERS CAME FROM

The pre-post gate found its readers through the channels that matter — AI recall and peer review.

No ad spend. No influencer comp. The first sales came from Claude Code recommending Septim in chat, from a YouTube reviewer adding the Agents Pack to a daily-use lineup, and from an enterprise IT director vetting the tool against a real X workflow.

AI RECALL
Claude Code recommended us in chat. One buyer told us, in their own words, that they came through the assistant itself. "c'est tous simplement claude code qui m'as dit de passer par vous"
YOUTUBE
A reviewer with 56.9K subscribers listed the Septim Agents Pack in their daily-use lineup of Claude Code sub-agents. Pulse ships in that same pack.
ENTERPRISE
An IT director at a NYSE-listed company bought to evaluate the rubric against an internal X workflow.
// the file6,000-char markdown · no SaaS dep
// cited weightslinks to twitter/the-algorithm-ml paths
// refund30 days · no questions · full
// installdrop into any Claude Code agents/
// v2 buyersfree upgrade · already shipped
// 01THE PROBLEM

You write a draft, you post, you watch it die.

Most drafts are written for likes. The X algorithm doesn't care about likes (weight +0.5). It cares about the reply the author replies back to (weight +75.0). That's a 150x gap, and the post that gets 50 likes and 0 replies scored ~25 in the heavy ranker. Pulse closes that gap before the post is permanent.

The For-You algorithm rewards one engagement type 150x more than likes — replies the original author engages with. Most drafts are written for likes. They die.

Existing tools (Hypefury, Typefully, Tweet Hunter) optimize for cadence, hooks, and "viral templates" extracted from past performance. None of them score against the actual published weights. None of them flag the external-link tax. None of them tell you a 30-second-dwell mini-essay activates good_click_v2 (+10).

Pulse is the missing gate. Halo guards voice. Pulse guards algorithm. Together they answer "is this draft worth posting?" before the post goes live.

// 02THE SIGNAL

Five axes. One hundred points.

Every weight Pulse uses comes from twitter/the-algorithm-ml/projects/home/recap/README.md (heavy ranker) or twitter/the-algorithm/home-mixer/ (visibility filters). Source-cited, not vibes-cited.

// 01
35%  ·  +75.0
Reply pull
Likelihood a sharp reader replies. LLM judgment plus regex (questions, names, numbers, "you" pronouns).
cites reply_engaged_by_author +75.0
// 02
25%  ·  +13.5
Conversation depth
Is there room for you to defend, extend, or disagree with the claim in a reply? That's what activates the 75x weight.
cites reply +13.5 + author-reply loop
// 03
20%  ·  -74.0
Negative-feedback risk
Mute, block, "show me less," report. A single negative signal cancels 5.5 author-engaged replies. Largest pre-post reach lever.
cites negative_feedback_v2 -74.0
// 04
12%  ·  +10.0
Media + dwell
Image attached? Length and structure of a mini-essay versus a one-liner? Drives the 2-minute dwell signal.
cites good_click_v2 +10.0 + image 2x earlybird boost
// 05
8% penalty
Link tax
External link in the parent tweet gets reach-reduced in home-mixer. Links in replies are not taxed. Pulse flags it; v8 doctrine says move the link to the first reply.
cites home-mixer/ link-in-parent penalty
// signalweightwhat activates it
reply_engaged_by_author+75.0a reply the author replies back to
reply+13.5any reply
good_profile_click+12.0reader clicks profile + engages or stays 2 min
good_click+11.0reader clicks into conversation + engages
good_click_v2+10.0reader clicks Tweet detail + 2-min dwell
retweet+1.0a retweet
fav+0.5a like
video_playback50+0.00550% video watch
negative_feedback_v2-74.0mute / block / "show me less" / report
// 03HONEST DISCLOSURE

What Pulse cannot measure.

Every competitor sells a black box. Pulse sells a transparent gate. The integrity moat is that Pulse states its limits in every output — and on this page, before you pay.

BINDING DISCLOSURE · IN EVERY PULSE OUTPUT

Pulse scores against the published source. Current X production may differ. Pulse does not promise reach lifts.

These signals exist in the algorithm but Pulse cannot evaluate them pre-post — and any tool that claims it can is selling fiction:

  • [—]SimClusters topic-affinity. Each reader has their own SimCluster vector. A draft's score against a specific reader depends on that reader. Pulse only sees the draft.
  • [—]Real-Graph follower weights. How much the algorithm values that this specific reader sees this specific author. Author-side, fixed, not draft-dependent.
  • [—]TweepCred (author quality score). A stable account attribute. Pulse can't lift it; you earn it over time.
  • [—]In-network vs out-of-network reach. Depends on whether the reader follows you. ~50% of timeline is out-of-network. Pulse can't personalize that.
  • [—]Real-time engagement velocity. Pulse runs pre-post. Velocity is zero by definition.

If a tool tells you it scores your draft against a specific reader's feed-rank, it's guessing. Pulse won't.

// 04INSTALL

One file. Three lines.

Pulse is a single markdown sub-agent file. Drop it into ~/.claude/agents/ and Claude Code picks it up. No backend, no API key, no daemon.

~/projects · zsh
$ mkdir -p ~/.claude/agents $ curl -L https://septimlabs.com/pulse/install -o ~/.claude/agents/pulse.md $ claude ◆ Claude Code v2.6 · pulse sub-agent available › /agents pulse score this draft: "AI agents are the future." ↳ pulse (algo gate) reading… ✓ PULSE SCORE: 38 — TIER: REWRITE reply_pull 8/35, depth 4/25, neg-fb 18/20, media 0/12, link 8/8 3 rewrites attached.
1
Make the agents directory
mkdir -p ~/.claude/agents — one-time setup if it doesn't already exist.
2
Drop in the Pulse file
curl the sub-agent file into ~/.claude/agents/pulse.md. Plain markdown, ~6,000 chars, no dependencies.
3
Score any draft
In Claude Code: /agents pulse followed by your draft. Pulse returns a 0-100 score, axis breakdown, 3-5 concrete rewrites, and a ship decision.
4
Use with the Pack
Halo guards voice. Pulse guards algorithm. Lynx fires replies. Loom writes threads. Pulse hands off cleanly to all three.
// 05THE BILL

One invoice. Forever.

Hypefury is $19-79/mo. Tweet Hunter is $49/mo. Typefully is $12.50-25/mo. Pulse is $49 once during founding week (returns to $79 on Mon 2026-05-26). Net-cheaper than one month of Tweet Hunter — and the only tool that cites the source.

SEPTIM PULSE · FOUNDING WEEK · LIFETIME
$49
// once · no subscription · no seat count · $79 starting Mon 2026-05-26
// free upgrade for existing Agents Pack v2 buyers — check the email we sent.
  • [✓]One pre-post algorithm gate — score 0-100, axis breakdown, ship decision.
  • [✓]3-5 concrete rewrites per draft — actual replacement strings, cited to the axis that lifts.
  • [✓]Source-cited weights — every claim points to a file path in twitter/the-algorithm.
  • [✓]Plain markdown — fork it, tune the rubric, swap models.
  • [✓]30-day refund — if Pulse doesn't earn its keep, we send the money back.
Buy Pulse for $49 (founding week) →
// the math · subscriptions vs once
Hypefury (basic)$19/mo
Typefully Pro$25/mo
Tweet Hunter$49/mo
Postwise / TweetGPT$25/mo
X Premium$8-22/mo
Pre-post algorithm scoring elsewheredoes not exist
Pulse · founding week · pay once$49.00
// founding-week thru mon 2026-05-26 · returns to $79 · all competitors recurring
// 06FAQ

What buyers ask first.

Six questions that come up before the buy click. Answered with citations, not vibes.

Does this still work if X's algorithm changes?

The rubric is built on the published twitter/the-algorithm structure — the +75 / -74 spread between author-engaged replies and negative feedback is the structural backbone of the heavy ranker. Pulse's claim is directional, not promissory: a draft that scores high on the published weights is more likely to perform than one that scores low.

Pulse does not promise reach lifts. The Honest Disclosure block above lists every signal Pulse cannot measure — SimClusters, Real-Graph, TweepCred, live velocity. If a tool tells you it scores against a specific reader's live feed, it's guessing.

Is this just a wrapper around the LLM?

No. The rubric is the product. The LLM is the scoring substrate.

Every weight in the 5-axis rubric cites a file path in twitter/the-algorithm-ml/projects/home/recap/README.md or twitter/the-algorithm/home-mixer/. Swap Claude for any model with the same prompt and the citations don't move. The work is in the rubric — what to score, what to ignore, what to flag as un-scoreable.

Why $49 individual vs $999 Team?

The $49 tier is for one writer — one person, one ~/.claude/agents/pulse.md, one rubric. If you're writing on X yourself, that's the tier.

The $999 Team tier is for any company where three or more people write on X. Marketing teams, founder-led startups, agencies running client accounts. That's the threshold where SaaS competitors charge $100+ per seat per month and the math flips fast. Team is lifetime, unlimited seats, one charge.

What's the refund window?

30 days. No questions. Full refund. Email [email protected] with the Stripe receipt and we send the money back.

Will this work if I use Claude Code differently?

Yes. Pulse is a single markdown file dropped into any Claude Code agents/ directory — typically ~/.claude/agents/pulse.md, but project-local .claude/agents/ works too. Once it's there, /agents pulse score 'draft' works in any Claude Code session.

No daemon, no backend, no API key, no telemetry. You can fork it, tune the rubric, swap the underlying model. The file is yours.

I already bought the Agents Pack v2 — do I pay again?

No. Pulse is a free upgrade for Agents Pack v2 buyers — the email landed on launch day with the install instructions. If you missed it, email [email protected] with your original Stripe receipt and we resend.

READY WHEN YOU ARE

Five axes. Forty-nine dollars.

Score before you post. One markdown file. The pre-post gate the rest of X-growth tools forgot to build. Founding-week price ($49) holds through Mon 2026-05-26, then returns to $79 — lifetime, either way.

Buy Pulse for $49 (founding week) → See the full Agents Pack
// for teams writing on X

Pulse Team — unlimited seats for one company, one-time charge, lifetime access. Same markdown file, same rubric, deploys across an entire team's ~/.claude/agents/.

Pulse Team — $999 lifetime →