skip to content

the natural ranking index

133,000+ coffee spots,
one honest number.

the nri is a bayesian-weighted merit index. seven signals, logarithmic volume scaling, specialty classification, and a locals-versus-tourists credibility layer. ranked on quality, never on who paid.

the decomposition

NRI0 - 100
review quality35
review volume25
social + buzz13
profile depth10
web presence8
photo coverage7
specialty + local13

weights shown as max points per signal; the raw sum is clamped to 100. on top, a verified "find us on nas" backlink adds up to +8 and editor visits add their own bump.

review quality

35 pts

the bayesian-shrunk star rating. a 4.8 across 600 reviews beats a 5.0 across 9.

review volume

25 pts

logarithmic, with diminishing returns. confidence, not a popularity contest.

social + buzz

13 pts

an instagram link (+4), verified follower reach (up to +3), and buzz-engine velocity (up to +6). momentum, not vanity.

profile depth

10 pts

claimed ownership (+5), hours, contact, description (+5). a serious operation fills the gaps.

web presence

8 pts

a live, working domain. a digital footprint that proves the lights are on.

photo coverage

7 pts

owner + editorial imagery. you eat with your eyes before the first sip.

specialty + local

13 pts

coffee-led classification (+5, or a penalty for food-first) plus the locals-versus-tourists ratio (up to +8). the soul of the score.

the math behind it

most directories rank on raw stars times review count. that surfaces the loudest, not the best. the nri runs four corrections that the run-of-the-mill index doesn't.

bayesian shrinkage

raw averages lie at low sample sizes. every rating is blended against a global prior (μ ≈ 4.3 over 50 phantom reviews), so a 5.0 with nine ratings regresses toward the mean while a 4.7 with two thousand holds firm. confidence is earned, not claimed.

logarithmic volume

review count enters on a log curve that saturates near two thousand, and past four thousand reviews a venue with no coffee word in its name has its volume points capped outright. a fifty-thousand-review tourist magnet cannot bury a six-hundred-review specialty bar on sheer footfall. we reward depth, then stop counting.

specialty classification

a coffee-led signal layer reads the name and editorial copy: single-origin, pour-over, roastery, named equipment. food-first markers (brunch, dessert, full kitchen) are penalised. the index measures coffee, not popularity.

local credibility

the strongest discriminator we track: the ratio of local regulars to one-time tourists in the review base. a great roaster's reviewers are insiders; an observation-deck cafe's are visitors. it separates the real from the merely busy.

the formulae

the closed form, matching the scoring kernel. shrinkage against a global prior, a concave volume transform, a piecewise specialty classifier, an additive local-credibility term, and the disclosed backlink and editor-visit bumps, summed and clamped to a bounded 0–100 interval.

bayesian rating estimator

r̂ = (W·μ + n·x̄) ∕ (W + n)

the posterior mean of a beta-binomial shrinkage model: each venue's sample mean x̄ over n ratings is regularised toward the population prior μ = 4.3 with pseudocount W = 50. quality points then map linearly: Q = clamp(((r̂ − 3.5) ∕ 1.5) · 35, 0, 35).

log-saturated volume

V = min(c, (log₁₀(n + 1) ∕ log₁₀(2000)) · 25), c = 14 if n > 4000 ∧ ¬coffee-named, else 25

a monotone-increasing, strictly concave transform. the marginal worth of each additional review decays toward zero past n ≈ 2,000, and a tourist-magnet cap holds venues with landmark-scale volume but no coffee word in the name to 14 points, so footfall cannot dominate craft.

specialty modifier (piecewise)

S = −8·I[food-first] + 5·I[coffee-led] − 4·I[neither]

an indicator function over a tokenised pass of the name and editorial copy. branches are mutually exclusive; the unclassifiable default carries a penalty so ambiguity is never rewarded.

local-credibility term

L = (ℓ ∕ 100) · 8, ℓ ∈ [0, 100]

linear in ℓ, the resident-to-visitor ratio of the review base. purely additive - a venue is lifted for local loyalty, never penalised for its absence.

backlink + editor bumps

B = backlink_bump (≤ 8) + editor_visit_bump

trigger-maintained, disclosed because our pricing page sells the first one: add your 'find us on nas' badge or a plain link to your site, we verify it daily, and up to +8 lands automatically. editor-visit bumps follow in-person visits from our team. both are earned, neither can be bought directly.

composite index

NRI = round(clamp(Q + V + P + S + L + B, 0, 100))

P aggregates the presence vector - web (8), photo (7), social & buzz (13: 4 for an instagram link, up to 3 for verified follower reach, up to 6 from the buzz engine), profile depth (10: 5 for a verified claim, 5 for completeness). the sum is clamped to the unit-hundred interval and re-evaluated on a weekly cadence.

notation: x̄ sample mean over n ratings · μ prior mean (4.3) · W prior pseudocount (50) · ℓ resident-to-visitor ratio · I[·] the indicator function · clamp(z, a, b) bounds z to [a, b].

how an elite score is built

a worked example. a london roaster with a 4.8 rating over ~600 reviews, full claimed profile, live site, owner photos, an active instagram, and a strong local following assembles a 94 like this.

review quality29 / 35
review volume21 / 25
social + buzz10 / 13
profile depth10 / 10
web presence8 / 8
photo coverage7 / 7
specialty + local9 / 13
total nri94

badge tiers

the score maps to a badge, visible on every card, detail page, and search result. the distribution is deliberately top-heavy with rigour: elite is rare by design.

elitenri 85-100

the rarefied top. consistent quality, deep review base, full presence, and a local following. the places other roasters follow.

featurednri 70-84

strong fundamentals across the board. high ratings, real volume, an active footprint.

verifiednri 50-69

solid spots with good reviews. often missing profile depth or still building volume.

claimednri 30-49

owner-claimed, actively building presence. early but engaged.

basicnri 0-29

newly indexed or thin. claim the profile to start compounding the score.

how to move up

the score recalculates weekly. these are the levers, in order of impact.

biggest

earn reviews at 4.5+

the bayesian rating is the heaviest signal. consistent high ratings at volume move it most.

+5

claim your listing

verified ownership unlocks the profile-depth and claim components.

+7

add owner photos

real imagery of the room, the bar, the cup. fills the photo-coverage signal.

+5

complete the profile

hours, contact, description, social links. close every gap.

+8

keep the site live

a working domain is the web-presence signal. dead links cost you.

+5

lean into specialty

single-origin, pour-over, your roaster, your method. the classifier rewards coffee focus.

what the nri is not

not pay-to-win. a premium plan unlocks badges and editorial features, but the score itself is computed from public quality signals. a $200 featured listing with a real local following outranks a sponsor with mediocre reviews, every time.

not an ad auction. we do not sell ranking positions. there is no slot to buy at the top of a city page.

not a chain index. starbucks, costa, pret, blank street and the rest are flagged and excluded from every ranking. indie only, on purpose.

ready to compound your nri?

claim your listing and start building the score that ranks you.