Public · Methodology changelog

Methodology changelog

Material changes to scoring weights, thresholds, and methodology. Cosmetic or UI-only changes are not logged here.

See also: methodology · errata · freshness · data sources.

• 2026-05-27v1.2.0

  Free tier → metered monthly access

  The free-tier lookup meter moved from a weekly cap back to a calendar month, and free accounts now get full feature depth within that budget (pain clusters, component breakdowns, advanced-mode weights). A 'lookup' is one distinct category or app opened in the month re-opening one you've already seen is free, so refreshing a page never burns the meter twice. The cap is 3 lookups per month (now 3 - lowered 2026-06-10), resetting on the 1st at 00:00 UTC.

  lookups_per_period
• 2026-05-25v1.1.0

  Methodology v1.1 NLP improvements + opportunity normalization

  Wish clustering now filters wish-trigger words from cluster labels, raises the minimum document frequency to reduce noise, and selects the number of clusters via silhouette scoring rather than a fixed value. The distinctive-phrases signal now uses Monroe et al. keyness (not raw frequency), making category phrase indexes more informative. Opportunity-score normalization is winsorized at p95 before min-max, preventing a single outlier category from compressing the rest of the corpus. All NLP pipelines are now explicitly English-only (non-English reviews still count toward rating and velocity signals). The /methodology page was updated to match.

  opportunity_scorephrase_indexwish_clusters
• 2026-05-24v1.0.0

  Methodology page published

  First public version of the scoring methodology. Vulnerability (8 components), opportunity (6), shape (5).

  vulnerability_scoreopportunity_scoreshape_score
• 2026-04-30v0.9.4

  Weekly usage cap

  Lookup cap reset cadence moved from monthly to weekly. The free cap was set to 20 lookups per week, resetting Monday 00:00 UTC.

  lookups_per_period
• 2026-04-15v0.9.0

  Three intent presets

  Opportunity score now exposes three presets (gap / beat / validate), each emphasizing the underlying signals differently. The UI defaults to 'gap'.

  opportunity_scorepreset_scores
• 2026-03-22v0.8.0

  Low-review hard-floor for vulnerability

  Apps with very few reviews (under ~50) are now hard-floored to a vulnerability score of zero; the earlier formula surfaced unproven, near-zero-revenue micro-apps, which is not the intended use case.

  vulnerability_score
• 2026-02-18v0.7.0

  Category-level pain themes

  Negative reviews in each category are now grouped into recurring complaint themes, surfaced as short theme labels (never verbatim review text). Themes need a minimum number of distinct apps before they're shown, so a single noisy app can't manufacture a theme.

  pain_themes
• 2026-01-20v0.6.0

  Winsorized opportunity normalization

  Opportunity components are now robustly normalized across all real categories before being combined, so a single extreme category no longer compresses the rest of the scale. Scores shifted slightly; relative rankings were largely preserved.

  opportunity_score
• 2025-12-08v0.5.0

  Platform-owned apps excluded from leaderboards

  Apps published by Shopify and other large platforms (Meta, Microsoft, Google and similar) are now excluded from vulnerability leaderboards. They aren't realistic challenge targets for an independent builder and were crowding out genuinely contestable incumbents.

  vulnerability_score
• 2025-11-10v0.4.0

  Meta-category shape score

  Introduced a bucket-level shape score over the hand-curated meta-categories, blending mean vulnerability, pain density, price floor, demand, and how crowded the bucket is.

  shape_score
• 2025-10-06v0.3.0

  Confidence and sample-size surfacing

  Every score now travels with the sample size behind it (apps, reviews), and a low-confidence indicator appears when the sample is small, so thin-data scores are read with appropriate caution.

  confidence
• 2025-09-01v0.2.0

  Category noise filtering

  Country names and other non-category tags are now filtered out before anything is scored or displayed, so they never appear as if they were product categories.

  categories
• 2025-08-04v0.1.0

  First opportunity and vulnerability scores

  Initial deterministic scoring: a per-category opportunity score and a per-app vulnerability score, both composites of public-listing signals. No machine learning in the loop; the same inputs always produce the same score.

  opportunity_scorevulnerability_score