See how every FalcoScan score is calculated.

The Pulse Heat Score

Every tool's Heat Score is a weighted sum of five inputs. Each input is rescaled to 0–100 before the weights are applied, so a score of 100 would require a tool to lead on every dimension simultaneously. In practice the top of the distribution sits in the high 80s. This is the ranking signal behind the Pulse issues.

heat_score =
    0.30 × growth_signal_weight
  + 0.25 × trend_direction_weight
  + 0.20 × opportunity_score
  + 0.15 × recency_boost
  + 0.10 × rating_normalized

trend_direction_weight

Three-state direction of category/tool movement: up → 100, flat → 50, down → 10.

recency_boost

Exponential decay with a 14-day time constant, measured from the most recent observation of the tool: every 14 days the boost falls to about 37% of what it was (so it halves roughly every 9.7 days). A tool observed today scores 100; one observed 14 days ago scores 37; one observed 30 days ago scores 12. Tools marked dead score 0 on this term.

recency_boost = 100 × exp(-days_since_observed / 14)

rating_normalized

The FalcoScan rating (0 to 5) rescaled to 0 to 100. The smallest weight in the composite, because ratings barely separate one product from another; treated as a tie-breaker, not a primary signal.

Momentum

Categorical momentum label maintained by the FalcoScan editorial data layer. Four states: hot → 100, rising → 70, stable → 40, declining → 10. This label is the heaviest single input into Heat Score (30%) and the largest weight in the Rising 10 composite (40%). It reflects real-time editorial judgment synthesized from launch velocity, funding activity, search volume, and qualitative review — not a pure historical metric.

Opportunity score

A 0–100 estimate of addressable opportunity for a tool or category, assigned by the FalcoScan opportunity model. It considers addressable-market signals, competitive saturation, and differentiation. High opportunity plus low saturation is the “sweet spot” corner of the Niches chart on every Pulse issue.

Saturation score

A 0–100 estimate of how crowded a category is. Higher means more competitors, more commoditization pressure, less room for a new entrant. Low saturation on top of high opportunity is the pattern FalcoScan flags as a buildable niche. Published per tool (inherited from its category) and per category directly.

Wrapper risk score

A 0–100 estimate of commodity risk — the probability a tool is a thin layer over someone else's model and will be absorbed by the platform it wraps. High wrapper risk is the primary signal for the Risk Watch section of every Pulse issue. Shown per-tool on browse cards and in category deep-dives.

Moat rating

A product's moat score (0 to 100) estimates how well it is protected from rivals and from general AI models. It blends four FalcoScan inputs: switching cost (30%, which feeds the switching costs measure), stickiness (25%, habit and retention), differentiation (25%, hard to copy) and the inverse of wrapper risk (20%, wrapper exposure). A score of 75 or more is a wide moat, 55 to 74 is narrow, and below 55 is none. Public pages show the band; the exact score is included with paid plans, the API and the MCP server.

A market's rating on AI Moats is the average of its products' scores, on the same bands. Each market is also rated on five moat sources (switching costs, habit and retention, hard to copy, regulatory barrier and ecosystem) and two threats (wrapper exposure and platform dependency), each judged against the other AI markets. Regulatory barrier, ecosystem and platform dependency describe the market and are not part of a product's score. The sources follow Morningstar's moat sources and Hamilton Helmer's 7 Powers, limited to what FalcoScan measures on every product; network effects, proprietary data, scale and brand are not scored. Every live AI product is rated. Products whose scores are still estimated from their market's peers count, and the Moats pages say how many; products that shut down or were acquired, and the AI directories market, are not rated.

Category Heat

Category Heat scores a market as a whole. It is not an average of tool Heat Scores: every market starts at 50, and five properties of the market move it up or down.

category_heat = 50
  + 8 × (3 × hot + rising) / active_products
  + 0.18 × (avg_opportunity − 50)
  − 0.20 × (avg_saturation − 50)
  − 15 × wrapper_share
  + min(6, 0.12 × launches_90d)

rounded, and held between 0 and 100

Momentum: each hot product counts three times and each rising product once; that total is divided by the market's active products (those not shut down) and multiplied by 8. Opportunity: 0.18 points for every point the market's average opportunity score sits above 50, taken off when it sits below. Saturation: 0.20 points taken off for every point average saturation sits above 50, added when it sits below. Wrappers: up to 15 points taken off, in proportion to the share of the market's products with a wrapper risk score of 60 or more. Launches: 0.12 points for every product added to the market in the last 90 days, up to 6.

The inputs are recomputed from the catalog every day, and the per-market heat values in each Pulse issue are these scores as they stood in that week's snapshot. (The overall Heat Index at the top of an issue is a separate figure, built from tool Heat Scores.) High means momentum; low means crowded, commoditized, or both. Compare every market in the Market Terminal; the per-category breakdown lives at /browse/[category].

Rising 10 composite

The monthly Rising 10 ranking uses a different composite than Pulse Heat Score — narrower, and designed to be citable in a single sentence. Three inputs, declared weights, deterministic output.

rising_10_composite =
    0.40 × signal_weight       (hot → 100, rising → 70)
  + 0.35 × opportunity_score   (already 0–100)
  + 0.25 × rating_rescaled     (0–5 → 0–100)

Filter: only tools tagged hot or rising with a non-null opportunity score that have not shut down are eligible. Ties go to the higher rating, then the higher opportunity score, then the name in alphabetical order. Weights are declared in lib/articles/rising-10-schema.ts; changing them is a methodology change and gets a changelog entry below.

FalcoScan rating

The rating on a product page, and the rating term in the Heat Score and the Rising 10, is one figure: the FalcoScan rating, from 0 to 5 to one decimal. It is recorded with the product when it is added to the catalog. It is not a live average of reviews: FalcoScan does not collect ratings from its visitors, and none of the formulas on this page recalculate it. A product submitted by its maker starts without one, and a product without a rating shows none.

Ratings bunch tightly near the top of the scale, so the rating separates products far less than the other measures here. That is why it carries the smallest weight in the Heat Score and why we treat it as weak evidence. Our note on the rating column shows how little it discriminates.

How a product is marked shut down or acquired

The AI Graveyard lists a product only with evidence. A catalog label is not enough on its own, so a product the catalog marks as gone stays out of the graveyard until the evidence is there.

Shut down

A published source says the product closed, such as the company's own notice, a regulatory filing or a press report. Or its site is gone (the domain no longer answers, or the home page returns an error) or parked, and the Internet Archive holds an earlier capture of it working, which proves the product existed. The site route is never used for a feature of a public company, or for a product that shares its website with one still running.

Acquired

A published source reports the acquisition and names the buyer. A merger of equals counts as acquired, but never as a purchase by either company. Whether an acquired product still runs is recorded only when a source says so.

Bought, then closed

A product that was bought and later closed is recorded as shut down, and keeps its buyer. It counts once, as a shut-down, and also among that buyer's purchases, with the year it was bought and the year it closed when sources give them.

Review

Every sourced verdict is reviewed by a person. Research records the verdict, the buyer, the year and the source; a second check re-opens every source; and a person reads each verdict the graveyard publishes. A shut-down shown by the site route instead carries the month the Internet Archive last saw its site working. Renamed products, retired features, companies still running on their own, and anything a source does not settle are left out. The year is the year a product shut down or was sold; when only the Internet Archive dates a shut-down, it is the year of the last working capture. The catalog sometimes lists one product more than once, so the graveyard counts products, not listings.

Corrections

If a record is wrong, or a shutdown or acquisition is missing, email friends@falcoscan.com with a link to a published source. We check it the same way, and correct the record if it holds up. The graveyard rebuilds from the database once a day.

Pulse issue curation

Each Pulse issue is generated from a frozen snapshot of the dataset, then curated by the editorial desk. The sections — Top Movers, Breakouts, Hidden Opportunities, Risk Watch, Graveyard, and Category Heat — use different filters of the same source data. Breakouts require a created_at inside the last 30 days; Hidden Opportunities require opp > 75 and saturation < 40; Risk Watch surfaces declining momentum and wrapper risk > 75. The Brief at the top is written by a human against the snapshot — not LLM-generated.

Update cadence

Underlying tool signals refresh continuously as the ingest pipeline pulls new evidence. The Heat Score view recomputes nightly at 5:30am CT. A full platform snapshot is written at 5:45am CT. Each Pulse issue is built from that snapshot. Rising 10 publishes on the 1st of each month, also from a frozen snapshot so the ranking is reproducible.

Citation format

When citing FalcoScan numbers in articles, threads, or research, please use the canonical form: “Source: FalcoScan, [series] [issue label].”

For Pulse numbers: link to the dated issue URL (/articles/pulse/YYYY-MM-DD) or the number alias (/articles/pulse/issue/N). For Rising 10: link to the monthly issue (/articles/rising-10/[month]-[year]). For category-level citations: link to the category page (/browse/[category]). All three URLs are permanent.

Changelog

9 October 2026: moat rating added. A new moat score for every product and a moat rating for every market, built from existing measures. No existing formula or weight changed, and no published score moved.

7 October 2026: two additions. The rules for marking a product shut down or acquired in the AI Graveyard are now set out here. The rating was described as a user rating; it is the FalcoScan rating, recorded when a product is added, and is now defined in one place. No formula or weight changed, and no published score moved.

25 September 2026 — corrections. Two descriptions on this page did not match the formulas in use, and now do. The Heat Score recency term was called a 14-day half-life; the formula, exp(−days / 14), has a 14-day time constant (37% left after 14 days, half after about 9.7), which is what the worked examples always showed. Category Heat was described as an aggregate of tool Heat Scores; it is a 50-base composite of market-level momentum, opportunity, saturation, wrapper share and launches, now set out term by term. No formula or weight changed, and no published score moved.

v1.1 — consolidated methodology. Heat Score, Rising 10 composite, Opportunity, Saturation, Wrapper Risk, and Growth Signal now documented at one canonical URL (/methodology). /pulse/methodology 301s here. No formula weights changed in this revision.

v1.0 — April 2026. Initial public methodology. Heat Score formula fixed at the weights above. Category Heat uses a 50-base composite with saturation and wrapper-risk penalties. Recency term degrades gracefully when observation timestamps are missing, via a cascade through the most recent refresh event.

Limitations, honestly

Every score here is an opinionated signal, not an oracle. Momentum depends on editorial judgment, which means it reflects the FalcoScan team's read of the market — not an impartial measurement. Opportunity and Saturation models are tuned against our own definition of “buildable” and will not be identical to a VC's TAM model or a Gartner category map. Wrapper Risk is blunt; a tool with high wrapper risk may still win its category on distribution or UX, and a tool with low wrapper risk can still lose. If you disagree with a categorization, you'll be disagreeing with a human, and we'll hear you out. Any material change to the methodology is versioned and announced in the changelog above.