How it works

From a scraped event
to a scored move,
to your Friday digest.

IDØ is an execution-signal pipeline. We ingest the public surface of every tracked company, score each observable move with a deterministic engine, capture the market reactions, and compress the noise into a weekly digest.

The pipeline

Five steps. Every score traces back to a URL.

  1. 01

    Ingest

    the public surface of every tracked company

    Scrapers run on a rolling schedule against websites, careers pages, blogs, GitHub, RSS, HN, X, Product Hunt, technical mirrors, regulatory filings, and a curated set of trade press. Every fetched artifact lands in the event store with a timestamp and a source URL.

  2. 02

    Diff

    compute what's actually new

    Each artifact is diffed against its prior snapshot. The diff is the unit of analysis — a pricing-page change, a careers-page role addition, a release tag, a press mention. Diffs are normalised into a typed event schema (ship / hire / press / raise / open-source).

  3. 03

    Score

    the engine assigns a numeric impact

    Every event runs through a deterministic scoring function. Weights are calibrated against historical outcomes (acceptance into accelerators, downstream funding rounds, retention proxies). The score is broken into three sub-scores: Speed, Effectiveness, Combat intel.

  4. 04

    Capture reactions

    what the market said

    For each scored event, the platform searches community, expert, and VC surfaces for quoted reactions. Each reaction is tagged with author kind, source attribution, and a sentiment signal. The reaction is part of the score; how the market reacted is half the story.

  5. 05

    Compress

    weekly digest

    Friday morning, the LLM compresses every event collected during the week into a structured summary — the eight to twelve moves that should change your week, with full traceability to the underlying events.

The engine

Deterministic by design. No LLM in the source path.

The single most important architectural choice we made: the scoring engine is deterministic. Every number on every page maps to a specific event with a URL and a timestamp. The language model only summarises what the engine already recorded — it never makes up facts, never invents signals, never embellishes the data.

The consequence is that we can hand you a dashboard and you can audit it. You can click any score and see the events that produced it. You can click any event and see the source URL. You can verify the entire chain in ten seconds, every time. This is non-negotiable for a tool that exists to make you more confident, not less.

The trade-off is that the engine is more work to build than "ask GPT about it." We think that's the right trade.

The reaction layer

What the market said matters as much as what they did.

A pricing-page change is one signal. A pricing-page change followed by 312 positive community mentions, a Forrester quote, and two named VC tweets is a different signal entirely. The reaction layer captures both halves.

For each scored event, the platform pulls quoted reactions from three categories:

  • communityHN, Product Hunt, Reddit, X / Twitter, comment volume and sentiment.
  • expertsPersonal blogs and substacks from recognised practitioners + analyst notes.
  • VCsPublicly attributable VC commentary on Twitter, Substack, podcasts.
What you see

One dashboard. Two product modes.

Founder mode

Add competitors, see their moves, read the reactions, compare yourself to the field, get the digest every Friday. Active monitoring.

Investor mode

Search any private company, read its six-month execution report, see its competitive set. Read-only by design — no watchlists, no alerts.

Design choices

The things we refused to build.

No demographic features

We never train on age, gender, race, school prestige, or location. The score reflects what your team has built and what your idea resembles — never who you are.

No LinkedIn scraping

We do not scrape LinkedIn. Public LinkedIn fields are accessed only if you opt in via your own session. We never hold third-party credentials centrally.

Investor mode is read-only

Investors cannot follow, watchlist, or trigger alerts on any company. The platform never tells a company that an investor opened its page. By architecture, not policy.

No binary verdicts

We do not return yes/no decisions about companies, ideas, or founders. Every output is a probability with a confidence interval and an explanation. Decisions remain yours.

See the dashboard, decide for yourself.

The /app workspace is open with seeded demo data. Toggle between founder and investor mode in the sidebar. No login required to look around.

Open the demo →