Not an RSS reader.
A conviction engine.
FL1 is a five-stage pipeline — discover, filter, verify, classify, deliver — built around a single fund's specific watchlist. Hard keyword gates. Up to three sequential gpt-5.4 verification passes. Direct SEC EDGAR polling that catches a filing before any journalist writes it up. A self-maintaining momentum index, a 13D/13G whale-watch, prediction-market-derived IPO odds, and a warm/cold ledger that tracks whether FL1's own early calls actually played out. This is what it does, feature by feature, with the real numbers behind it.
FL1 Logo
Mascot
// Pipeline_Architecture
A collection of fetchers, not one big scraper
Sixteen-plus intelligence fetchers each run on their own cadence: RSS-based news search across two parallel tracks, SEC EDGAR full-text search, Form 4 insider-buy filings, Greenhouse job boards, GitHub org activity, Polymarket's Gamma API, Tranco rank exports, iTunes podcast search, and court records — roughly 1,100+ search queries per ingestion cycle, run multiple times a day. The General Track covers regional editions per company; the Authoritative Track runs a dedicated site-level query against every credibility-tiered domain, for every company, every cycle.
Cheap, deterministic, free — and it kills most of the noise
Every candidate is checked against per-company allow/deny lists before a single model token is spent. This is stage zero of a pipeline that gets more expensive and more discriminating at every later stage — necessary, but nowhere near sufficient on its own. LinkedIn gets its own layer: only IPO, funding, and M&A signals pass; job posts, open roles, and "excited to share" personal posts are killed on arrival, across the entire watchlist.
A backstop for the ambiguity hard gates can't anticipate
Run only where a company's name history has actually shown it's needed — Kraken, Plaid, ElevenLabs, Klarna, Supabase, and a handful of others — gpt-5.4 confirms the article is substantively about the company, not a coincidental name match or a passing one-line mention. Running it universally would be needless spend on names that never collide with anything.
The workhorse — and where most rejections happen
Runs on every article that reaches it, not a scoped subset. Rejects four recurring patterns: duplicate coverage of an event twelve outlets already wrote up, boilerplate press releases restating a prior announcement, opinion pieces with no new factual content, and a company appearing as one incidental name in a "20 startups to watch" roundup rather than being the actual subject.
Two independent channels, deliberately redundant
A dedicated pass reads for S-1/F-1 filings, underwriter selection, crossover rounds, confidential filings, direct listing or SPAC announcements, and roadshow confirmations — in parallel with a direct, independent poll of SEC EDGAR's full-text search API. A registration statement can be caught purely from landing on EDGAR, before any journalist has written about it at all. A signal missed by one channel is often caught by the other; one confirmed by both carries higher confidence than either alone.
3,500 words in. 3 sentences out.
readability-lxml strips ads and boilerplate, feeding up to 3,500 words of actual article text to gpt-5.4 — not a 200-character RSS snippet. Output is a fixed structure: fact + key numbers · investment implication · one forward trigger. Translated into RU · EN · AR · ES via gpt-5.4-mini and disk-cached — the same translation is never generated twice.
Fact: ElevenLabs closed a Series C at an $11B valuation, extending its lead as the dominant AI voice synthesis platform with no close competitor at scale.
Implication: At this valuation, strategic acquirers face a rapidly closing window — consolidation cost rises materially with each round.
Watch: Underwriter selection and S-1 timeline, given the crossover structure of the round.
Fact: SpaceX's Starlink unit is trading at approximately $2.6T in secondary markets ahead of a potential direct listing.
Implication: Secondary premium at this scale implies strong institutional demand — a direct listing would be the largest in US market history.
Watch: SEC registration statement and any Musk commentary on structure or timeline.
Outcome: SpaceX debuted on NASDAQ (SPCX) in June 2026. Subscribers had the secondary-pricing signal in their digest months before the bell.
Four daily slots. One weekly IPO digest. Infrastructure built to actually stay up.
Digests drop at 08:00, 14:00, 20:00, and 23:00 Moscow time — each subscriber picks their own slots. Every Monday at 10:00 MSK a dedicated IPO/S-1 digest compiles all pre-IPO signals from the week. Beyond the shared digest, each subscriber can maintain a personal watchlist, and FL1 generates a personalized brief after every send scoped to their own tracked subset. Every item is simultaneously mirrored to the live web dashboard with company filters, signal-type tabs, date range, and full-text search.
Reliability engineering
// Live_Dashboard
Dashboard Screenshot
api.klymo.net/fl1/
// Signal_Intelligence_Suite · Beyond the digest
The three-pass verification pipeline is the core, but it's surrounded by sixteen-plus purpose-built features that turn raw coverage into an ambient sense of what's moving across the whole watchlist.
// Signal_Taxonomy · Auto-classified on every article
Auto-detection runs on title and full body before brief generation. The IPO scanner monitors the IPO Signal category across all 37 companies simultaneously, every cycle.
// Company_Profiles · Weekly deep-refresh
Not just headlines. A living profile for every company.
Every Monday a separate pipeline rebuilds a structured profile for all 37 companies — stage, latest round, valuation, and IPO likelihood — and keeps it on the dashboard next to the news feed.
Tracked to the bell: two watchlist companies have gone public under FL1 coverage — Klarna (NYSE:KLAR, Sep 2025) and SpaceX (NASDAQ:SPCX, Jun 2026). Their profiles switched to live 5-minute quotes the week they listed.
// Command_Reference · Everything works on-demand, not just on a schedule
A digest you can query, not just receive
Every command runs against the same live archive that feeds the digests, so an analyst can pull history, check the system's own accuracy, or scope a personal watchlist without waiting for the next scheduled send. Two authenticated JSON endpoints — /api/momentum and /api/receipts — expose the momentum index and the edge KPI directly, for tooling that shouldn't have to go through Telegram at all.
// Real_Numbers · Pulled directly from the live system, not estimated
Every brief in the archive passed three consecutive filter layers — duplicate coverage, PR filler, and passing mentions are calibrated out before anything reaches you. Roughly 1,100+ raw queries run per ingestion cycle; on the order of 2% of that scanned volume survives to become an actual delivered brief — about 85 verified briefs a day across the active watchlist. What's in the archive earned its way there. Your analysts decide instead of discard.
Top tracked companies by archived coverage
Polymarket · SpaceX · Klarna · Kraken · Apptronik · Crusoe · Plaid · Payward · Discord · ElevenLabs
The impact was immediate. We've reduced repetitive tasks dramatically, and the system saves our team approximately 15 hours every single day. That time is now reinvested into growth and customer-facing work.
The pipeline is live.
See it yourself.
Every brief from the last 30 days is on the dashboard now — searchable by company, signal type, and date. Book an audit to discuss a pipeline built around your fund's specific thesis.