Methodology

How this feed is sourced

The feed surfaces aggregated movement from a curated set of public technologists. Identities are kept internal; cards show distilled, anonymised signal only.

Where we observe

We observe public posts from a hand-curated set of approximately 55technical voices — engineers, founders, fund managers focused on technology, open-source maintainers. The sources are free, public APIs and RSS feeds: Bluesky (AT Protocol), Mastodon (per-instance ActivityPub), Substack (per-publication RSS), YouTube (channel-RSS plus auto-captions for two discovery channels we watch — Fireship and Developer Voices), and SEC EDGAR (13F-HR filings for the fund-manager subset).

How we filter

Every voice in the universe passes four criteria: technical credibility, weekly-or-better public output cadence, occasional intersection with publicly traded companies, and explicit absence of a stock-picker public persona. The universe is reviewed annually.

How we aggregate

Each raw post is scored by an extraction model that pulls ticker mentions, sentiment, and a conviction estimate. Posts produce zero, one, or several signals; most posts produce zero. Signals roll up into per-ticker aggregate cards on rolling 24h, 7-day, and 30-day windows. A card writes only after at least two distinct voices have produced a qualifying signal on the ticker (one-voice cards are allowed only on the 24-hour window when the single voice's conviction is high). The card body itself is generated by a small language-model call from the aggregate counts plus 3–5 sample paraphrased summaries.

What we never do

• We never name an individual on a public card. The card body is generated by a constrained prompt that prohibits attribution; a compliance guard scans every card body and rewrites or redacts any name or @handle that slips through.
• The feed is never shaped by your holdings — every visitor sees the same cards. Holdings you model are stored only to your own account, encrypted, and never feed the cards, the AI, or our research.
• We never claim predictions (“X will outperform”). Cards describe what the curated universe is observing, not what the market will do.
• We never republish verbatim quotes longer than 8 words. The extractor's summary is a paraphrase; the aggregator compresses again.

Opt-out

If you're one of the technologists whose public posts are in our observation set and you'd like to be removed, contact us. We'll verify the request comes from you, soft-delete your entry, and prune the underlying posts within 30 days.

This page covers how the social feed is sourced. The standing verdicts and signals are a separate system — that page is the fundamental and regulatory track record, with the signal thresholds and SPY-relative results. For how a Buy/Hold/Trim/Avoid verdict is generated, see Behind the AI. The glossary defines the terms used on every card.