pbProductBench

Every score is a claim. Here's how we earn each one.

ProductBench measures customer trust in Saudi fintech from governed, verifiable evidence — never scraped sentiment, never model guesswork. This page is the full contract behind the number.

5Trust pillars
n ≥ 400For HIGH confidence
n ≥ 100Evidence floor
0Ungoverned sources
SAMPLE · illustrative, not a governed live reading. Pipeline percentages, F1 benchmark, snapshot IDs, evidence IDs/dates, and freshness values below are demonstration values until connected to a verified data adapter.
01 · Pipeline

Raw voice in, governed evidence out.

Every review passes four gates before it can move a score. Each gate discards — what survives is what we call n governed.

GATE 1

Ingest

App Store, Google Play and X/KSA reviews collected on a fixed schedule, in the original language — Arabic stays Arabic.

drops ~0% everything enters
GATE 2

Verify

Duplicate, bot, incentivized and off-market submissions removed. Signals must match a real KSA customer.

drops ~22% of raw volume
GATE 3

Classify

Each item maps to one driver and one pillar, held to F1 ≥ 0.85 on a KSA fintech benchmark. Below that, counts are suppressed.

drops ~6% as unclassifiable
GATE 4

Score

Governed evidence rolls up into pillar scores, then a weighted trust score, stamped to an immutable snapshot ID.

output sealed snapshot
02 · Pillars

Five pillars, fixed weights.

Weights are published in advance and never tuned to move a specific entity. A pillar with insufficient evidence is marked unscored — it is never imputed.

ReliabilityDoes it work, every time
28%
SupportSpeed and quality of resolution
22%
Trust & safetyMoney and data feel safe
20%
Fees & clarityNo surprises in pricing
16%
OnboardingFrom download to first use
14%
03 · Confidence

Confidence is a rule, not a feeling.

Every score carries a tier computed from mechanical thresholds. The tier controls what ProductBench is allowed to show — a LOW-confidence entity can never be ranked or charted authoritatively.

HIGH
  • n ≥ 400 governed items in period
  • 5 / 5 pillars scored
  • ≥ 2 independent sources
permits: score · rank · trend · anomaly markers
MEDIUM
  • n 150–399 governed items
  • ≥ 4 / 5 pillars scored
  • coverage gaps disclosed inline
permits: score · trend — excluded from rankings by default
LOW
  • n < 150 governed items
  • any pillar count
  • single-source coverage allowed
permits: raw evidence only — no score chart is ever rendered
04 · Refusals

What ProductBench will not claim.

The system is built to refuse. If a question has no governed data path, the answer says so — visibly, in the product.

NOT CLAIMED

Revenue or churn impact

Trust movement correlates with business outcomes, but we hold no governed financial data. Answers state the boundary rather than extrapolate.

NOT CLAIMED

Causes without evidence

A driver is named only when the evidence names it. “63% of items cite checkout declines” is a claim; “users are frustrated” is not.

NOT CLAIMED

Scores from stale sources

A source more than 48h behind its refresh schedule is excluded and flagged. Its absence is disclosed on every affected answer.

NOT CLAIMED

Anything without a citation

Every generated sentence resolves to evidence IDs you can open. If the citation set is empty, the sentence is not shown.

05 · Evidence

Source language is the source of truth.

Evidence is stored and displayed in the customer's own words. English glosses are aids, never substitutes — analysis runs on the original Arabic.

تم رفض بطاقتي ثلاث مرات عند الدفع رغم وجود رصيد كافٍ. التطبيق بعد التحديث الأخير أصبح لا يعتمد عليه.
"My card was declined three times at checkout despite sufficient balance. After the last update the app is unreliable."
ev-9012App Store · KSA2026-07-30driver: Technical Issue✓ verified
06 · Freshness

Data ages in public.

Every snapshot states when each source was last refreshed. Stale sources are excluded from computation, and the exclusion is printed on the answer that would have used them.

SourceRefresh cadenceStale thresholdCurrent status
App Store · KSAevery 6h> 48h● fresh · 0.1h ago
Google Play · KSAevery 6h> 48h● fresh · 0.1h ago
X / KSAevery 12h> 48h● fresh · 0.5h ago
News / KSAevery 12h> 48h● fresh · 0.5h ago

Excluded now: SAMA · 185h DEAD · regulatory · 318h COLD

The contract behind the number

Audit the number yourself.

Start watching your marketmethodology v2 · snapshot 2026-08-02 05:00Z