Methodology
No black box.
A benchmark is only worth sharing if you can defend it. This page is the complete recipe behind every BenXmark report card — the tiers, the distributions, the percentile model, the metric weights, and the grade bands. This page renders from the same code that scores you, so it can never be out of date.
1 · You're only compared to your tier
Reach and engagement scale non-linearly with audience size — a 500-follower account routinely posts 5% engagement while a 500K account rarely clears 1%. Comparing across sizes is noise. BenXmark buckets creators into seven follower tiers (0–1K, 1K–5K, 5K–10K, 10K–25K, 25K–50K, 50K–100K, 100K+) and every percentile is computed strictly within your tier.
2 · Where the distributions come from
Each tier starts from curated baseline quartiles (25th, 50th, 75th percentile per metric) assembled from public engagement research and hand-verified submissions. You can inspect every number on the benchmarks page.
As creators benchmark themselves, anonymous submissions accumulate per tier. Once a tier crosses 30 submissions, its live community quartiles replace the baseline — and your report card explicitly says which one scored you. We never fabricate sample sizes, testimonials, or counters anywhere in the product.
3 · The percentile model
Creator metrics are ratios of counts and empirically right-skewed, so each metric is modeled as log-normal. From a tier's quartiles we derive the distribution parameters:
μ = ln(p50) σ = (ln(p75) − ln(p25)) / (2 × 0.6745) percentile(x) = Φ((ln(x) − μ) / σ) × 100
where Φ is the standard normal CDF. Percentiles are clamped to 1–99 — nobody is a flawless 100 and nobody is a hopeless 0. The distribution curves on your report card are drawn from these exact parameters, so the picture and the number always agree.
4 · The metrics and their weights
| Metric | Formula | Weight |
|---|---|---|
| Engagement rate How often people who see your posts interact with them. | engagements ÷ impressions | 1.50 |
| Reach multiplier Impressions relative to your audience size — how far past your followers your content travels. | impressions ÷ followers | 1.25 |
| Engagement quality Interactions weighted by effort — a reply counts 3×, a save or repost 2×, a like 1×. | (replies×3 + saves×2 + reposts×2 + likes) ÷ impressions × 1,000 | 1.25 |
| Save rate Bookmarks per 1,000 impressions — the strongest signal of genuinely useful content. | bookmarks ÷ impressions × 1,000 | 1.00 |
| Conversation rate Replies per 1,000 impressions — how much discussion your content starts. | replies ÷ impressions × 1,000 | 1.00 |
| Profile pull How often your content makes people click through to your profile. | profile visits ÷ followers | 1.00 |
| Click-through How well your content drives traffic off-platform. | link clicks ÷ followers | 0.75 |
| Audience quality Share of your followers that are verified — a rough proxy for real, invested humans. | verified followers ÷ total followers | 0.50 |
Your overall score is the weighted mean of your metric percentiles. Metrics that aren't visible in your screenshot are simply left out — they neither help nor hurt you.
5 · The grade
score ≥ 97
Top 3% — outlier
score ≥ 90
Top 10% — elite
score ≥ 75
Top 25% — outperforming
score ≥ 55
Above the median
score ≥ 35
Around the median
score ≥ 0
Below the median
6 · Data handling
- Screenshots are read once by an AI model to extract the numbers, then discarded. They are never stored.
- Submitted metrics enter the anonymous community pool only when you press submit — with no handle, name, or account attached.
- Result links carry the metrics inside the URL itself. Share one and you're sharing those numbers; that's the deal, stated plainly.
- No login, no tracking pixels in share cards, no dark patterns.
7 · Known limitations
- Baseline quartiles are estimates until a tier has real community volume. They're good context; they are not a census.
- Self-selected submissions skew ambitious — live medians will likely sit above the true population median. Percentiles against this pool are therefore conservative: the real crowd is easier to beat.
- The log-normal fit is a model. It's a good one for engagement-style ratios, but tails are approximate — treat P97 vs P99 as the same message.