# Profectus — Community Opportunity Bot & Dashboard Plan

Human-in-the-loop Reddit/forum posting system. Goal: consistent, credible, value-first presence in fitness/method/lifehacking communities that funnels curious readers to Profectus, without platform-manipulation risk and without AI-sounding content.

## Architecture (self-hosted on the Pi)

New service (not part of the app): `/home/rpi/projects/profectus-growth/`
- SQLite DB (opportunities, drafts, posted log, accounts profile)
- Hermes cron (every 4h): discovery + scoring + draft generation
- Tiny HTTP dashboard served LAN/Tailscale-only (no public exposure, no auth needed beyond network boundary), e.g. `http://100.102.237.9:5186`

## 1. Discovery & scoring (the bot's read pass)

Sources (unauthenticated endpoints, rotating polite rate): r/<sub>/new.json for the target list, plus RSS for legacy forums (T-Nation, EliteFitness, IronMagazine), plus Google Alerts/serp for "rest pause program" blog comments (phase 2).

Opportunity scoring per new thread:
- keyword hits (method terms: rest pause, myo reps, doggcrapp, partial reps, RP vs drop sets, "no time to train", plateau, program advice)
- question-shape bonus: "how do I / which program / am I doing it right / why am I stalling" beats statement posts
- timing: age < 48h (fresh enough to be seen), comment count < 15 (early = top reply possible), upvote velocity positive
- fit: sub-specific relevance weight (HIT/DC/myo-rep communities score highest, generic fitness lower)
- output: score 0-100 + one-line "why this fits" + link + suggested reply angle

## 2. Draft generation (anti-AI-tells ruleset)

Every draft must pass a checklist before it reaches the dashboard:
- no em dashes, no "delve/leverage/unlock/great question", no emoji, no bullet lists in comments, no bold
- varied length (40-150 words), varied sentence rhythm, contractions, at most one specific technical point (one concrete rep scheme or rest length beats five abstractions)
- no invented personal history: placeholders like [X weeks] / [my current lift] wherever a personal claim would go — YOU fill the truth. The bot never fabricates "I ran this for 3 months"
- plain register per the brand: engineering-note tone, opinion allowed, no marketing sentence at all unless the sub rules and the thread explicitly invite a mention
- mention policy: a Profectus mention is allowed only when the thread asks exactly what we do AND sub rules permit; otherwise the mention is the profile click, never the comment

## 3. Your training profile (the authenticity engine)

One-time intake (I interview you, stored locally only): real training history, current numbers per lift, how long you have used partial-set methods, honest opinions (what you like/dislike about DC, myo-reps, 531). Drafts cite only facts from this sheet, in your voice. This is the single biggest quality lever: the bot writes as you, truthfully.

## 4. Scheduling (natural-human cadence)

- max 1-2 substantive posts/day total, never two in the same hour
- send slots drawn from your wake hours (CEST morning 7-9, lunch, evening 18-22), jittered +-20 min, no pattern across days
- per-sub pacing: same subreddit max once per 5-7 days; method-expert subs (small) max twice a month
- ramp: weeks 1-2 comments only, no profile mention; week 3+ profile mention allowed; first explicit "we built this" post only after ~20 value comments and only in self-promo-permitted venues
- dashboard shows each draft with a suggested send time (the jittered slot) and a one-click thread link; you review, edit if needed, post manually, hit "posted"

## 5. Dashboard (spec)

One page, table of pending opportunities:
- columns: sub/forum, thread title (link), score + why, draft (inline editable), suggested send time, status (new / edited / posted / skipped)
- history tab: posted log with date + link + karma snapshot (checked daily by bot) so you can see what resonated
- settings tab: sub list + weights, keyword list, send-window config
- server logs note: referrer traffic from reddit.com to the app measures the funnel (profile -> site), giving us real conversion data per community

## 6. Feedback loop

Weekly digest: opportunities acted on vs skipped, karma per posted reply, profile->site clicks from server logs, and any signup attributed to the reddit referrer. This is what turns "we think it works" into per-community CPS numbers.

## 7. My additional thoughts (what makes this powerful)

1. The intake profile is the whole game. A truthful personal voice from a real lifter cannot be faked at scale, and that is precisely what no competitor is doing.
2. Comments outperform posts: a great reply under a 500-upvote "how do I stop stalling" thread beats our own thread in every way (credibility, reach, durability).
3. The bot should optimize for being remembered, not clicked: a commenter people recognize ("that rest-pause guy is always precise") converts readers for years. Play the long identity game.
4. Never vote-manipulate, never argue via second account, never cross-post identical text. One violation ends the channel.
5. Time budget: your part is ~10-15 min/day at the dashboard; the bot handles the rest.
6. Kill criteria: if after 90 days there are <5 site visits/week from referrers, the channel mix is wrong, not the concept; reshuffle the sub list rather than abandoning.

## Build order

1. Pi service skeleton + DB + dashboard shell (half day)
2. Discovery + scoring cron (half day)
3. Draft generator + checklist + intake interview (half day)
4. Scheduling + history + digest (half day)
Can be built now; it does not depend on the public launch. Execution (posting) starts only after you have done the intake and the first two weeks of comment-only ramp.