How we know what we claim
A brief is a written verdict, so it is held to a writer's standards: cite it, count it, or keep quiet. Every Rivalize brief is edited against the rules on this page — the same rules on a loud week and a silent one.
If we can't cite it, we don't print it. Where we abstained, you see the reason — never a guessed zero.
// 01 · The thesis comes first
Every brief leads with one sentence about your market — a judgment you can act on or argue with, not a metric name, a status, or a chart waiting for interpretation. The findings below it exist to defend that sentence.
The discipline cuts both ways: when the analysis cannot state a thesis it would defend, the brief renders none. A filler sentence at the top of an intelligence report is worse than silence, because everything under it inherits the doubt.
// 02 · The three-source floor
A claim printed as a verdict needs at least three independent sources behind it. Below that floor the claim is labeled context, not verdict — shown with the reason it fell short, or withheld as an explicit abstain. It is never quietly promoted to fact.
The same bar applies to proof. A before/after diff renders only when both the prior string and the new string are on record — a one-sided diff is an assertion, not evidence.
// 03 · Abstain beats a guess
When we don't know, we say so — visibly. An unknown value is missing data, not a low number, so it is never printed as a zero and never filled with an estimate dressed as an observation. In practice:
-
A company below the minimum-signal threshold reads
momentum: nullwith the reason stated, and is excluded from every leaderboard rather than ranked on fabricated zeros. - The implication line under a finding appears only when the analysis produced one it can defend. An absent "so what" is honest; an invented one is the failure mode this product exists to avoid.
- Company facts that don't resolve — leadership, founding details — are shown as honest gaps, not guessed.
- Dossier sections without enough data are named once, with the reason they were withheld. Never rendered as empty boxes waiting to be filled.
// 04 · Confidence is printed, not implied
Where a finding carries a computed confidence, the number is printed next to the claim, in the open. Where no confidence was computed, none is shown — a missing number is never defaulted to look certain.
Scores work the same way: every momentum score returned by the API carries the version of the recipe that produced it, so any number you quote stays auditable against the exact methodology below.
// 05 · A quiet week is still a brief
Nobody moved. That's the signal. Monitoring tools go blank when nothing happens, and you learn nothing. A brief always has a writer, so a quiet week still gets one: what was checked, what came back — nothing — and why the quiet is worth noticing.
The quiet gap is stated as a fact — pages checked, changes found — on a single hairline. Never a padded section, never an apology, never an empty state.
// 06 · What we watch
Signals come from observed public behavior across these categories — what a company does, never modeled estimates of what it might be doing:
We publish the categories; customers see the full list. Inside the product, the sources behind your briefs are enumerated, and every claim in a report links back to the sources that support it — so coverage is something you audit, not something you take on faith.
// 07 · What we refuse to claim
- No modeled revenue or download estimates. Scores are built exclusively from observed public signals.
- No fabricated zeros. Unobserved is missing, not low.
- No verdicts below the floor. Thin evidence is labeled as context or withheld — never rounded up to certainty.
- No unversioned numbers. Any change to a signal set, weight, or threshold bumps the methodology version, and every score carries the version that produced it.
// 08 · The momentum score, in the open
The momentum score is a 0–100 measure of how actively a company is moving, and the exact recipe is public. Five signal families, with weights summing to 1.00. For any single company, the weights of its present signals are renormalized to sum to 1 — a company is never penalized for a signal we have not observed yet.
| Signal | Weight | What it measures | Counts as present when… |
|---|---|---|---|
| Funding events | 25% | Recency of observed funding rounds. A round closed in the last 6 months scores highest; older rounds decay. Total-funding amounts alone are stock, not momentum — only dated events move this signal. | At least one observed funding round or disclosed total funding. |
| Hiring activity | 20% | Open roles observed on careers pages and job boards, scaled by volume on a calibrated curve. An observed zero is a real (low-momentum) observation, not a missing one. | An observed open-role count (including an observed zero). |
| Advertising activity | 15% | Active ad campaigns observed in public ad libraries (Meta Ad Library, Google Ads Transparency), scaled by observed active-creative count on a capped curve. | An observed advertising state (active count or advertiser flag). |
| Social publishing activity | 20% | Recency, volume, and available engagement for posts observed on the company’s social profiles in the last 60 days. Follower counts are stock, not flow — only publishing activity moves this signal. | At least one observed recent post with a timestamp. |
| Product and pricing changes | 20% | Observed changes to pricing plans or product pages between snapshots. A pricing change in the last 90 days scores highest; stability decays toward a low baseline. | An observed change event with a timestamp (e.g. a dated pricing-plan diff). |
Minimum-data threshold
A company needs at least
2 present signals to be
scored. Companies below the threshold are excluded
from every leaderboard and read as momentum: null
with an explicit reason. We never rank a company on fabricated zeros — an unobserved
signal is missing data, not low momentum.
Update cadence
Scores are recomputed weekly. Every run appends to an immutable score history — nothing is overwritten — so deltas and trends are always derivable from the record. Leaderboard responses include each company's score change versus the previous run.
The formula
coverage = 0.75 + 0.25 × ( Σ present weightᵢ / Σ all weightᵢ )
Each present signal contributes a 0–100 sub-score (recency- and volume-scaled as described above). The composite is the weight-renormalized average, scaled by a signal-coverage factor — a company observed on only two signal families cannot max out the score, so breadth of observed evidence ranks above a single hot channel. With full coverage the factor is 1.0. Rounded to one decimal place.
Versioning
This page documents methodology v1.1. Any change to the signal set, weights, or threshold bumps the version, and every API response carries the version that produced its score — so published numbers stay auditable forever.
Track your competitors with Rivalize
Get 1 free report on 1 competitor — no credit card. Paid plans re-check their core pages twice each weekday and give you the recommended counter.
Get your free reportFree report. No credit card required.