Sense
Collectors pull search trends, news, local media, Wikipedia, GDELT, charts, X trends, YouTube and TikTok per country, every few minutes to every few hours. Raw is kept; sm ingest normalises it into signals every 15 minutes.
SignalMachina watches what is rising in your markets, what your competitors ship and what their users complain about, keeps brand-unsafe topics out, and turns every opening into a growth hack your brand can run now — hook, format, timing, and the loop that makes it compound. Scored with ICE, in your audience's language, ready for the creator to pick up.
Suggestions only · it never posts, spends or contacts anyone · hands Act ideas to ContentMachina
The clásico is peaking in Argentina. Bridge: turn the match tension into a freestyle challenge where fans predict the winner in two bars.
“No me digas quién gana Boca–Racing. Rimámelo en dos barras.”
loop fans stitch their two bars → the best get reposted → their followers join · ICE 7.3 (impact 8 · confidence 7 · ease 7) · facts to check are marked [verify]
App Store reviews keep saying the assistant freezes and bills unexpectedly. That weak spot becomes a play: show your product doing well exactly what they hate — next to releases, share of voice and TikTok mentions for every brand you track.
enters the same bridge check as a trend · up to 5 competitor moves per brand per day
01 The shift
Trending lists are noise: the same football match shows up as ten keywords, tragedies sit next to memes, and "post about it" is not a plan. What a brand needs is fewer, cleaner trends, a hard filter on what is off limits, and an honest answer to one question — is there a natural way in for us, what exactly do we run, and how does it keep bringing new people in?
02 How it works
A deterministic, watermarked pipeline on a scheduler, local models by default, every step measured. The only agentic step is ideation.
Collectors pull search trends, news, local media, Wikipedia, GDELT, charts, X trends, YouTube and TikTok per country, every few minutes to every few hours. Raw is kept; sm ingest normalises it into signals every 15 minutes.
Every signal is reduced to a canonical key (a Wikidata entity when it resolves) before embedding, so ten ways of writing the same match become one trend with a time series and a lifecycle: emerging, rising, peak, fading.
Deaths, disasters, violence and hard politics are labelled blocked before any brand sees them; caution and safe are the other two labels. Checked against 244 real headlines: 100 % of the unsafe ones blocked, none over-blocked. When in doubt, it blocks.
Per trend and brand, a local model judges whether there is a credible, non-forced way in. Lateral bridges are welcome; forced ones are rejected. Competitor moves — a new version, a listener jump, a review theme — enter the same check.
sm probe writes the day's ideas at 06:00: hook, format, timing, cost and a required growth loop (or an explicit one-off), each scored on impact, confidence and ease. Facts it cannot verify are marked [verify], never invented.
You rate Act / Maybe / No in the panel. Ratings calibrate ICE; Act ideas are picked up by ContentMachina within 15 minutes; outcomes from DistroMachina come back as winners and losers for the next round.
03 What it surfaces
Clean, labelled trends with stage, momentum and sources — ranked for the brand's family (music first for artists, tech and consumer first for apps). Brand-agnostic until a bridge personalises them.
Holidays, curated dates and fixtures per market, surfaced days ahead so the idea is ready when the moment arrives.
Hook, format, timing and cost that match the brand, the loop that makes it compound (trigger → action → share or invite → new people arrive → they repeat) and an ICE score with its reason. When the brand's pace vs target is known, one line steers the prompt toward acquisition or retention.
Self, three curated peers and anyone you add, side by side: listeners, ratings, releases, chart entries, press share of voice, TikTok mentions in your geos, and what their users complain about in app reviews — themes, sentiment, gist.
Every trend is labelled safe, caution or blocked and the labeller is evaluated on a labeled set of real headlines. Only safe trends reach the idea stage.
Families (tech, artist; fashion planned) define the feature catalogue; each brand is a folder — brief, voice, watchlists, channels, strategies — with its own markets and languages. A new brand needs no code.
04 Where it listens
Logged-out public data with rate limits and an identifying user agent. No scraping behind logins, no cookies, no CAPTCHA solving, no proxy rotation. A source that blocks is skipped and flagged — never evaded.
Paid sources (X API, TikTok data, frontier models for challenger ideas) each run under their own monthly cap; local models do the default work.
05 Autonomy & safety
SignalMachina's autonomy level is fixed at suggest: its output is a recommendation to a person, and it has no write access to any external platform.
06 In the panel
Local-network only; owner-facing text in the owner's language, publishable text in the brand's.
/signalPer-geo board, family-aware ranking, stage and safety chips, Upcoming lane.
/signal/ideasCards with ICE and its I/C/E chips, the loop, the evidence; Act / Maybe / No with a why; 'Draft content' hands an idea to ContentMachina.
/signal/competitionSide-by-side table with 7-day deltas and 30-day sparks, the moves feed, review themes, press share of voice, add or pause entities.
/signal/sourcesHealth per collector, last run, partial runs when a host blocked, spend per paid source against its cap.
/strategyWhere a signal meets a lever: each strategy names the sources and bridges it depends on.
/growthThe north star and its pacing — the same status the ideation prompt receives.
07 Status · 2026-10-04
SignalMachina is where the loop starts. Rate its ideas for a week and ContentMachina and DistroMachina take it from there.