Why the dashboard's EVLS / per-node projections diverged from reality, and what the on-chain data now says.
Live from dashboard-data.json. The analysis below is a point-in-time snapshot (see the "Prepared" date); this box tracks the daily-average cake, which is still settling through the month, so it can differ from the figures in the report body.
The dashboard's validator model assumed two size effects: a baseline of ~4.9 validations / node / day and an EVLS penalty that grew with validator size (-10% at 100 nodes, -44% at 221). Live data from 15 validators shows neither holds today. Post-fix, every validator from 3 to 132 nodes earns the same ~7.64 val/node/day, and earnings are now governed by a simple relationship: per-node = network demand cap ÷ total nodes.
The "efficiency increase" was not a new feature and not a size effect. A bug had been suppressing how many of the available validations the network could fulfil since ~Aug 2025; it was fixed around 21-22 Apr 2026, snapping fulfilment back to the demand cap. Combined with the validator count falling to an all-time low (76), each remaining node now captures a much larger slice of a fixed cake.
What it costs and what it means. Over the nine-month bug window the network missed roughly 1.25 million validations (~125,000 MNW) in unpaid rewards, about 316 per node (~20-25% of what an operator would otherwise have earned). The flip side: today's high per-node rate is partly a scarcity premium from the shrunken network. It will compress if nodes return, and rise as query demand grows - June demand has already lifted the daily cake to ~22,497/day, past the old ~20,000 ceiling (which was never a hard limit). Note: efficiency is confirmed flat only to 132 nodes; behaviour of 200-500 node validators is not yet measured.
The "Daily Validations Cap vs Fulfilled" chart tells the story. Demand (the cap) stayed flat at ~20,000/day the whole period. Fulfilled validations tracked it until ~Aug 2025, then decayed to ~10,000 by Apr 2026 as a bug throttled validator throughput - then snapped straight back to the cap on ~21-22 Apr 2026 when the bug was fixed.
A single validator's history makes the same shape visible in val/node/day: a long decline into a Feb 2026 trough of 2.78, then recovery and a sharp May-Jun jump to ~7.64.
How to read: each point is one validator's validations per node per day. The slide from ~4.9 to the 2.78 trough is the bug worsening; the sharp rise in May-Jun is the fix, amplified by fewer nodes sharing the work.
Crucially, the recovery coincided with the validator and node counts hitting all-time lows. Nodes roughly halved from the 5,331 peak (Jul 2025) to 2,943; validators fell from 332 to 76. Fewer participants share the same fixed cake - so the post-fix per-node rate is partly a scarcity premium.
How to read: total nodes (left axis) and validators (right axis). Both fell sharply to all-time lows, which is the main reason each surviving node now earns more of the same fixed cake.
Fifteen validators spanning a 44× size range (3 → 132 nodes) all land at the same per-node rate. The old model predicted a steep decline with size; the data is a flat line. This is the single most important chart in the report.
How to read: the red line is what the old model predicted at each size; the cyan dots are real validators. They sit on a flat line, not the declining curve, so the size penalty does not exist. The shaded zone above 132 nodes is not yet measured.
| Validator | Nodes | May 2026 | Jun 2026 | Reward/val |
|---|---|---|---|---|
| A | 3 | 5.97 | 7.70 | 0.10 |
| B | 3 | 5.93 | 7.55 | 0.10 |
| C | 4 | 6.03 | 7.76 | 0.10 |
| D | 10 | 5.91 | 7.57 | 0.10 |
| E | 17 | 5.98 | 7.65 | 0.10 |
| F | 20 | 5.99 | 7.66 | 0.10 |
| G | 90 | 5.95 | 7.64 | 0.10 |
| H | 100 | 5.99 | 7.65 | 0.10 |
| I | 132 | 5.94 | 7.66 | 0.10 |
| Mean | - | ~5.97 | ~7.64 | 0.10 |
Spread across the whole range is 7.55-7.76, within ±1.5%, which is just day-to-day noise. Query distribution is therefore equal per node, independent of validator size.
Our measurements span 3 to 132 nodes. We have not sampled validators in the 200-500 node range, so we cannot guarantee they receive the same ~7.64 val/node/day. It remains possible that very large validators are throttled or capped in a way we have not observed. The flat per-node rate is confirmed only up to ~132 nodes; beyond that it is an extrapolation, not a measurement (shown as a dotted line / hatched bar in the charts).
| Parameter | Old model | New (data-driven) |
|---|---|---|
| Baseline val/node/day (≤100) | 4.90 - 4.97 | 7.64 (= cake ÷ nodes) |
| EVLS penalty | -10.26% @100, -0.279%/node | 0% - flat to ≥132 nodes |
| Val/node/day @132 nodes | 4.01 | 7.66 |
| Network-size scaling | √(4,338 ÷ nodes) | linear: cake ÷ nodes |
| Total network validations | scale up with network | limited by demand (~22.5k/day) |
| Active nodes baseline | 4,338 | 2,943 |
| Active validators baseline | 108 | 76 |
| Reward / validation (12-mo) | 0.10 | 0.10 (unchanged) |
| Stake / node | 1,800 MNW | 1,800 (unchanged) |
How to read: red is the old model, green is reality. The gap widens with size, so the old model most badly understated large validators. The 200-node green bar is assumed (not measured).
| Metric (per node, 12-mo) | Old | New | Change |
|---|---|---|---|
| Val/node/day (@100 nodes) | 4.46 | 7.64 | +71% |
| Annual validations / node | 1,628 | 2,789 | +71% |
| Annual MNW / node | 162.8 | 278.9 | +71% |
| MNW yield on 1,800 stake | 9.0% | 15.5% | +72% |
| Break-even nodes / VPS @ $0.062 | ~8 | ~4 | easier |
| Val/node/day @132 nodes | 4.01 | 7.66 | +91% |
Per-node figures; a 100-node validator is 100× these. Yield is token-denominated (MNW reward ÷ MNW staked), not USD. The old model under-states earnings everywhere and over-penalises large validators specifically.
The old EVLS curve was calibrated on two anchors - 4.9662 val/node/day ("baseline") and 2.78 ("221 nodes, -44%"). But look at a single 132-node validator's own history: it sat at 4.94 in Aug 2025 and decayed to exactly 2.78 by Feb 2026. Those aren't two different-sized validators - they're the same validator at two points in time, during the bug's decline. The model mistook throughput decay over time for a penalty that grows with size.
Three compounding factors explain the gap:
How many validations did the bug cost? Without it, each node would have validated at cap / nodes per day (full demand met, shared equally per node). The shortfall versus that counterfactual, summed Aug 2025 to Apr 2026, is the missed total. This is an estimate: cap is taken as ~20,000/day, node counts are month-start snapshots, and April is a blended month (bug for ~3 weeks, fixed for the last ~9 days).
How to read: green dashed is what each node could have done with no bug (cap / nodes); cyan is what it actually did. The shaded area between them is the missed throughput, which closes once the bug is fixed in May.
About ~316 validations missed per node (~32 MNW) over the nine-month window, roughly 20-25% of the rewards an operator would otherwise have earned. Network-wide that is about 1.25 million validations (~125k MNW) in unpaid rewards.
| Validator size | Validations missed | MNW missed (×0.10) |
|---|---|---|
| 3 nodes | ~950 | ~95 |
| 10 nodes | ~3,160 | ~316 |
| 20 nodes | ~6,320 | ~632 |
| 50 nodes | ~15,800 | ~1,580 |
| 90 nodes | ~28,460 | ~2,846 |
| 100 nodes | ~31,620 | ~3,162 |
| 132 nodes | ~41,740 | ~4,174 |
How to read: missed validations scale linearly with size because the per-node shortfall is uniform. A 100-node validator lost roughly 31,600 validations (~3,160 MNW) over the window.
Recommended model: val/node/day = demand_cap ÷ total_nodes · reward 0.10 · EVLS = 1.0 · baselines 2,943 nodes / 76 validators · demand cap surfaced as an explicit, adjustable lever so growth scenarios show the dilution effect.