SN21 — Rewards (daily stream)
| Version | 1.1 |
| Audience | Miners |
| Status | Authoritative for daily-stream emissions |
| Last updated | 2026-08-04 |
| Update independently of | SN21_SCORING.md · SN21_STAKING.md · SN21_TRANSITION_PLAN.md |
This document explains how emissions are allocated once you have a standing. It does not explain how individual predictions are scored — that is SN21_SCORING.md.
In one line
Your standing (moving-average score) is mapped through a published weight curve into on-chain weights; emissions flow continuously on Bittensor’s tempo from those weights. The model that runs live (champion) is chosen by a separate, stricter rule.
From standing to weight
- Scoring produces a per-miner standing (SN21_SCORING.md).
- Standings pass through the weight curve below.
- Validators set weights on chain; emissions follow Bittensor’s own tempo (~every 72 minutes). There is no weekly payout event in the steady-state daily stream.
During cutover, bridge rules and indicative burn rates may temporarily modify who is paid — see SN21_TRANSITION_PLAN.md. Burn is not fixed; it may be adjusted at any time (see below).
The weight curve
Steep, but no winner-take-all cliff. Rank-based shares among earners; below the score threshold → zero weight. Hard ceiling on how many miners earn.
| Rank | Share of miner emissions (before re-normalisation) |
|---|---|
| 1st | 50% |
| 2nd | 25% |
| 3rd | 10% |
| 4th+ | Geometric tail: each next rank gets 50% of the previous share |
| Cap | At most 20 earners |
| Floor | Standing below the published score threshold → 0 weight |
If fewer than the full curve of earners qualify, shares are re-normalised so weights still sum to 1.0 (ratios preserved).
Ties break deterministically: higher standing first; equal standing → miner id ascending.
Published score threshold: starts at 0.0 (review-adjustable). At threshold or above earns; strictly below does not.
Placement eligibility
You need enough scored predictions in the standing window before the curve can place you:
- 250 scored predictions → eligible for placement
- 1000 → full standing confidence
(Details in SN21_SCORING.md.)
Champion vs earner (two different seats)
| Weights (who earns) | Champion (who runs live) | |
|---|---|---|
| Decides | Emission share via the curve | Which model runs across real accounts |
| Changes when | Standings move through the curve | All three promotion tests pass |
Champion promotion
The champion changes only when a challenger:
- leads the incumbent’s moving average by at least 5% (relative), and
- has held that lead for 7 consecutive days, and
- has at least 14 scored days of history.
Miss any one → incumbent stays. Weights can shift gradually while the champion seat does not.
Quiet days
No special weekend multiplier. Thin baskets simply produce fewer scored predictions; your standing is per-prediction, so volume self-scales.
Worked examples
Example A — three earners (curve re-normalised)
Standings on a given day (all above threshold, all placement-eligible):
| Miner | Standing | Rank |
|---|---|---|
| Alice | 0.82 | 1 |
| Bob | 0.75 | 2 |
| Carol | 0.70 | 3 |
Raw curve shares: 50% / 25% / 10%. Sum = 0.85. Re-normalise:
| Miner | Raw share | Weight after re-norm | Share of miner emissions |
|---|---|---|---|
| Alice | 0.50 | 0.50 / 0.85 ≈ 0.588 | ~58.8% |
| Bob | 0.25 | 0.25 / 0.85 ≈ 0.294 | ~29.4% |
| Carol | 0.10 | 0.10 / 0.85 ≈ 0.118 | ~11.8% |
| Total | 0.85 | 1.000 | 100% |
There is no cliff: if Bob’s standing rises toward Alice’s, ranks can swap and weights move — but nothing jumps from 0% to 100% on a single tick.
Example B — five earners (geometric tail)
| Rank | Miner | Standing | Raw share |
|---|---|---|---|
| 1 | Alice | 0.84 | 0.50 |
| 2 | Bob | 0.78 | 0.25 |
| 3 | Carol | 0.72 | 0.10 |
| 4 | Dave | 0.68 | 0.10 × 0.5 = 0.05 |
| 5 | Eve | 0.65 | 0.05 × 0.5 = 0.025 |
Raw sum = 0.925. After re-norm: Alice ≈ 54.1%, Bob ≈ 27.0%, Carol ≈ 10.8%, Dave ≈ 5.4%, Eve ≈ 2.7%.
Miners outside the top 20, or below the score threshold, get 0.
Example C — burn during cutover (does not change the curve)
Burn rates are planned and indicative only — they may change at any time. Using the illustrative 30% burn from the transition plan on 12 August: if Alice’s curve weight is 0.588 of the miner pool after burn:
- 70% of subnet miner emissions are paid out under the curve
- Alice receives
0.588 × 70%≈ 41% of total miner-side emissions that day - The burned share is not redistributed through the curve
Stake hold is a separate gate: fail the alpha hold → 0 payout that day even with a high standing.
Example D — champion vs earner
| Day | Alice standing | Bob standing | Bob lead vs Alice | Bob consecutive lead days | Bob scored days | Result |
|---|---|---|---|---|---|---|
| 1–6 | 0.80 | 0.86 | 7.5% (≥ 5%) | 1…6 | 20 | Bob earns more via the curve if ranked higher; Alice stays champion (hold not yet 7 days) |
| 7 | 0.80 | 0.86 | 7.5% | 7 | 20 | Bob promoted to champion (margin + hold + history all met) |
If on day 7 Bob’s lead shrinks to 3% (< 5%), the hold clock resets — Alice remains champion.
What replaces weekly tiers / EMA
The daily stream does not use Elite / Competitive / Participating bands or four-week EMA pools. Those were weekly-epoch machinery. The curve + standing replace them.
Burn rate and stake
- Burn during cutover is published as a plan only in SN21_TRANSITION_PLAN.md. Figures are indicative. Because Bittensor’s emissions-allocation methodology is changing significantly, SN21 may adjust burn at any time to protect and grow alpha value for all holders. Adjustments will be announced as soon as practical; do not treat the transition table as a locked schedule.
- Alpha hold / staking requirements are in SN21_STAKING.md.
Neither changes how an individual prediction is scored.
Parameter reviews
Numeric curve parameters (threshold, shares, cap) are restated at four-weekly published reviews. Changes are announced in advance; nothing silent between reviews.
Related
- Scoring: SN21_SCORING.md
- Staking: SN21_STAKING.md
- Cutover: SN21_TRANSITION_PLAN.md
- Why daily: SN21_WHY_DAILY.md