How the Opportunity Score works
This platform does not tell you what to build. It measures what people are publicly asking for, how much it hurts, whether the demand is growing, and who already serves it — then shows you the evidence behind every number.
The formula
Version v2-woodruff — Woodruff (1976) determinants; weights from Evanschitzky et al. (2012) meta-analysis.
The Opportunity Score combines seven independently measured sub-scores — demand, momentum, pain, competition, monetization, market size and evidence quality — into one deterministic number. No language model participates in scoring or adjusts a result.
The exact weighting formula is proprietary and not published here. What is published is every input that feeds it: every score change traces back to counted evidence, visible on each opportunity’s own history and evidence panels.
Weights are versioned internally. Recalibration ships as a new version so historical snapshots stay comparable, and every snapshot records the version that produced it.
What AI does and does not do
AI is used for
- · Semantic clustering of problem statements
- · Duplicate detection beyond exact text
- · Summarising a cluster from its own evidence
- · Topic and entity resolution
AI never decides
- · The Opportunity Score or any sub-score
- · Market size or financial value
- · Whether a source is telling the truth
- · Why a score moved — that comes from counted deltas
Pipeline
Internet → evidence → opportunity
- 01Internet
Public posts from Reddit, GitHub and Hacker News.
- 02Raw signals
Stored verbatim with source, URL, timestamp and payload — never overwritten.
- 03Signal extraction
Deterministic intent patterns + lexicon sentiment produce problem, intent, pain and purchase intent.
- 04Problem clustering
Embeddings + incremental cosine clustering group different phrasings of the same problem.
- 05Demand measurement
Mentions, questions, explicit requests, alternative searches and distinct people, normalised 0-100.
- 06Competition analysis
Observed products, entrants, ratings, feature coverage, pricing and funding signals.
- 07Momentum detection
Log-space regression over weekly volume, plus acceleration between halves of the window.
- 08Opportunity scoring
Deterministic, no LLM anywhere in the chain; every input is stored and shown.
- 09Historical tracking
One snapshot per hour per opportunity, with measured drivers for each move.
Evidence quality
Not every score is equally reliable
Each opportunity carries an evidence label. HIGH needs at least 30 signals across 2 sources with a strong evidence sub-score. MEDIUM needs 12+ signals. Everything else is LOW and should be treated as a lead, not a conclusion.
Near-duplicate posts are detected with SimHash and embedding similarity, and are excluded from demand counts — the same complaint reposted twice is not twice the demand.
Data collection principles
- · Only public, authorised endpoints are read; rate limits and source terms are respected.
- · No authentication, CAPTCHA or anti-bot control is ever bypassed.
- · Author identifiers are stored as one-way hashes, never as usernames.
- · Original payloads are retained so results can be re-derived when models improve.
- · Opportunity Scores are informational research output, not investment advice.
Frequently asked
- How is the Opportunity Score calculated?
- It combines seven independently measured sub-scores — demand, momentum, pain, competition, monetization potential, market size and evidence quality — into one deterministic number between 0 and 100. The formula is versioned, and no language model participates in scoring or adjusts a result.
- Where does the data come from?
- Public posts on Reddit, GitHub, Hacker News, app-store reviews and Vietnamese forums, read through authorised endpoints only. Original payloads are stored verbatim, so every figure can be traced back to the post it came from.
- Does an AI decide the scores?
- No. Language models are used to read and label text, never to produce a number. The score, the sub-scores, the market size and the explanation of why a score moved are all computed from counted quantities.
- What does the evidence label mean?
- It says how much the score can be trusted. HIGH means many signals across several independent sources; MEDIUM means enough to be worth reading; LOW means the score is a lead to check, not a conclusion to act on.
- Are these opportunities investments or securities?
- No. Nothing here can be bought, sold or traded. The scores are informational research output about where public demand is concentrated, and they are not financial instruments or investment advice.