---
cursor:
  subagentId: "bc-df131c91-63b4-57b1-9fe6-e0768319f732"
---

# Metrics sweep: training layer (Affine SN120, Teutonic SN3, IOTA SN9, Gradients SN56, Hone SN5, Albedo SN97, Reliquary SN81, ConnitoAI SN102, Refinery SN125, ChronoLLM SN38, cascade SN91, TPN SN65, Quasar SN24, Ralph SN40, Feval SN47, Capcomp SN103, plus Hugging Face downloads)

Accessed 24 Sep 2026. Machine-readable twin: `training.json` beside this file. Every series point carries its own source URL and a `verified` flag (true only where I read the value from a team-run API/dashboard/ledger or a public ledger such as the Hugging Face API).

Headline: the training layer has plenty of team-run JSON behind its dashboards, so most series here are verified, but few of them are honestly "up and to the right". The clean risers are the **IOTA** run counters (Orion-16B v8 1.1B -> 200.5B tokens in 51 days; v9 0.2B -> 24B in 13 days; Train-at-Home downloads 2.1K -> 45.9K since Feb 2026), **Reliquary** checkpoint publishing (4/day -> ~120/day since 18 Aug), and **Teutonic-II** king benchmarks going from chance to real numbers in three weeks (PIQA 52.8 -> 83.1, ARC-E 25.2 -> 83.5, GSM8K 0 -> 74). The king-of-the-hill composite scores are U-shaped, not rising: Affine's king total went 56.3 -> 43.3 -> 67.4 -> 63.2 (current king +7 over genesis after a month), and Albedo's SWE-bench Verified went 58.2 -> 2.4 -> 61.2 (+3 over genesis after three months). Connito baseline loss and cascade king error are not improving on their current data. Gradients' only public time series (HF model pushes) died in Nov 2025 when outputs moved off the public org. Hone, ChronoLLM, TPN, Feval have no measurable output.

## Series found

### affine-king-total — Affine (SN120): Composite king score (total), 0-100

- Unit: score (0-100) · kind: rate · source type: team · shape: **U** · points: 12 (2026-08-27 -> 2026-09-22, 56.26 -> 63.16)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 56.3 -> 43.3 (15 Sep, worst) -> 67.4 (reign 20, 19 Sep) -> 63.2 (reign 21). Honest shape is a U: the current king is above genesis by 7 points after a month, but the first hidden kings scored below genesis.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 56.26 |
| 2026-09-10 | 46.58 |
| 2026-09-14 | 47.75 |
| 2026-09-15 | 43.31 |
| 2026-09-17 | 49.36 |
| 2026-09-18 | 49.43 |
| 2026-09-18 | 57.59 |
| 2026-09-19 | 60.49 |
| 2026-09-19 | 60.78 |
| 2026-09-19 | 62.07 |
| 2026-09-19 | 67.39 |
| 2026-09-22 | 63.16 |

### affine-king-bench — Affine (SN120): King score on benchmark group (total:bench)

- Unit: score (0-100) · kind: rate · source type: team · shape: **U** · points: 12 (2026-08-27 -> 2026-09-22, 64.58 -> 67.8)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 64.6 -> 43.6 -> 70.3 -> 67.8.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 64.58 |
| 2026-09-10 | 50.21 |
| 2026-09-14 | 49.39 |
| 2026-09-15 | 43.55 |
| 2026-09-17 | 51.34 |
| 2026-09-18 | 51.28 |
| 2026-09-18 | 64.34 |
| 2026-09-19 | 66.31 |
| 2026-09-19 | 67.74 |
| 2026-09-19 | 68.54 |
| 2026-09-19 | 70.32 |
| 2026-09-22 | 67.8 |

### affine-king-env — Affine (SN120): King score on environment group (total:env)

- Unit: score (0-100) · kind: rate · source type: team · shape: **U** · points: 12 (2026-08-27 -> 2026-09-22, 50.63 -> 60.34)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 50.6 -> 43.2 -> 65.5 -> 60.3. Env group is the part that actually rose above genesis (+10).
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 50.63 |
| 2026-09-10 | 44.22 |
| 2026-09-14 | 46.77 |
| 2026-09-15 | 43.16 |
| 2026-09-17 | 48.1 |
| 2026-09-18 | 48.3 |
| 2026-09-18 | 53.47 |
| 2026-09-19 | 56.9 |
| 2026-09-19 | 56.31 |
| 2026-09-19 | 57.92 |
| 2026-09-19 | 65.51 |
| 2026-09-22 | 60.34 |

### affine-king-coding — Affine (SN120): King score, coding group

- Unit: score (0-100) · kind: rate · source type: team · shape: **rising** · points: 12 (2026-08-27 -> 2026-09-22, 47.4 -> 71.96)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 47.4 -> 42.6 (10 Sep) -> 79.5 (19 Sep) -> 72.0. Clearest upward group; still noisy king-to-king.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 47.4 |
| 2026-09-10 | 42.61 |
| 2026-09-14 | 55.64 |
| 2026-09-15 | 56.98 |
| 2026-09-17 | 63.31 |
| 2026-09-18 | 72.76 |
| 2026-09-18 | 60.83 |
| 2026-09-19 | 67.62 |
| 2026-09-19 | 65.02 |
| 2026-09-19 | 69.1 |
| 2026-09-19 | 79.5 |
| 2026-09-22 | 71.96 |

### affine-king-terminal — Affine (SN120): King score, terminal group

- Unit: score (0-100) · kind: rate · source type: team · shape: **U** · points: 12 (2026-08-27 -> 2026-09-22, 78.57 -> 91.56)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 78.6 -> 57.6 (19 Sep) -> 91.6 (22 Sep).
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 78.57 |
| 2026-09-10 | 68.73 |
| 2026-09-14 | 67.89 |
| 2026-09-15 | 76.73 |
| 2026-09-17 | 67.14 |
| 2026-09-18 | 65.14 |
| 2026-09-18 | 72.32 |
| 2026-09-19 | 63.37 |
| 2026-09-19 | 73.05 |
| 2026-09-19 | 57.6 |
| 2026-09-19 | 83.05 |
| 2026-09-22 | 91.56 |

### affine-king-math — Affine (SN120): King score, math group

- Unit: score (0-100) · kind: rate · source type: team · shape: **peaked** · points: 12 (2026-08-27 -> 2026-09-22, 44 -> 77.97)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 44.0 -> 93.5 (10 Sep) -> 68.8 -> 91.8 -> 78.0. Jumped on first hidden king, oscillates since.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 44 |
| 2026-09-10 | 93.46 |
| 2026-09-14 | 69.09 |
| 2026-09-15 | 78.33 |
| 2026-09-17 | 75.41 |
| 2026-09-18 | 72 |
| 2026-09-18 | 83.85 |
| 2026-09-19 | 73.08 |
| 2026-09-19 | 68.75 |
| 2026-09-19 | 91.67 |
| 2026-09-19 | 91.75 |
| 2026-09-22 | 77.97 |

### affine-king-tool-use — Affine (SN120): King score, tool-use group

- Unit: score (0-100) · kind: rate · source type: team · shape: **U** · points: 12 (2026-08-27 -> 2026-09-22, 65.5 -> 72.26)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 65.5 -> 24.9 (17 Sep) -> 80.7 (18 Sep) -> 72.3.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 65.5 |
| 2026-09-10 | 29.11 |
| 2026-09-14 | 36.11 |
| 2026-09-15 | 48.62 |
| 2026-09-17 | 24.9 |
| 2026-09-18 | 27.92 |
| 2026-09-18 | 80.65 |
| 2026-09-19 | 75.45 |
| 2026-09-19 | 79.29 |
| 2026-09-19 | 75.25 |
| 2026-09-19 | 74.32 |
| 2026-09-22 | 72.26 |

### affine-king-swebench-verified — Affine (SN120): King score on SWE-bench Verified subset

- Unit: score (0-100) · kind: rate · source type: team · shape: **flat** · points: 12 (2026-08-27 -> 2026-09-22, 79.5 -> 73.7)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 79.5, max 85.2 (reign 16), current 73.7; one crash to 48.2 (reign 17). Not rising.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 79.5 |
| 2026-09-10 | 80.25 |
| 2026-09-14 | 76.72 |
| 2026-09-15 | 79.5 |
| 2026-09-17 | 83.69 |
| 2026-09-18 | 83.1 |
| 2026-09-18 | 85.19 |
| 2026-09-19 | 48.23 |
| 2026-09-19 | 80.95 |
| 2026-09-19 | 84.97 |
| 2026-09-19 | 73.91 |
| 2026-09-22 | 73.7 |

### affine-king-terminal-bench-2 — Affine (SN120): King score on Terminal-Bench 2

- Unit: score (0-100) · kind: rate · source type: team · shape: **flat** · points: 12 (2026-08-27 -> 2026-09-22, 36.36 -> 37.93)
- Note: Read from affine.io/api/v1/matrix (team eval server). One point per publicly listed king, dated by coronation; kings 1-10 (29 Aug-10 Sep) are hidden in the matrix (reigns 12 and 13 removed as spam/model copy), so the gap 27 Aug-10 Sep is real. Reign 0 is the genesis model. Teacher Qwen/Qwen3.8-27B scores 68.82 total; Occamy-1.0 59.97; Albedo CXXVII 61.49. Genesis 36.4, min 16.1, current 37.9. Flat.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 36.36 |
| 2026-09-10 | 32.94 |
| 2026-09-14 | 27.59 |
| 2026-09-15 | 27.59 |
| 2026-09-17 | 29.55 |
| 2026-09-18 | 16.09 |
| 2026-09-18 | 25.29 |
| 2026-09-19 | 24.39 |
| 2026-09-19 | 29.89 |
| 2026-09-19 | 23.26 |
| 2026-09-19 | 30.23 |
| 2026-09-22 | 37.93 |

### affine-verdicts-per-week — Affine (SN120): Evaluation verdicts issued per week (duels)

- Unit: verdicts/week · kind: rate · source type: team · shape: **peaked** · points: 5 (2026-08-24 -> 2026-09-21, 91 -> 55)
- Note: From affine.io/api/v1/history (650 events, 27 Aug-24 Sep 2026): 592 verdicts total, 22 crowned, 31 failed, 2 revoked. Week starting 21 Sep is partial (4 days). Weekly crowns: 2, 7, 3, 9, 1. Queue stats: 637 queued, 19 accepted, 593 rejected. Peaked in the week of 7 Sep.
- Source: https://affine.io/api/v1/history (verified=True)

| date | value |
|---|---|
| 2026-08-24 | 91 |
| 2026-08-31 | 151 |
| 2026-09-07 | 172 |
| 2026-09-14 | 123 |
| 2026-09-21 | 55 |

### affine-crowns-cumulative — Affine (SN120): Cumulative kings crowned since genesis

- Unit: crowns · kind: cumulative · source type: team · shape: **rising** · points: 15 (2026-08-27 -> 2026-09-22, 1 -> 22)
- Note: Coronation dates from affine.io/api/v1/matrix (kings 0-21, genesis 27 Aug 2026). Cumulative by construction; 22 crowns in 26 days, 5 of them on 19 Sep alone. Two reigns later removed (12 spam, 13 model copy) are still counted as crowns here.
- Source: https://affine.io/api/v1/matrix (verified=True)

| date | value |
|---|---|
| 2026-08-27 | 1 |
| 2026-08-29 | 2 |
| 2026-08-31 | 4 |
| 2026-09-01 | 5 |
| 2026-09-04 | 6 |
| 2026-09-05 | 7 |
| 2026-09-06 | 9 |
| 2026-09-08 | 10 |
| 2026-09-10 | 12 |
| 2026-09-14 | 13 |
| 2026-09-15 | 14 |
| 2026-09-17 | 15 |
| 2026-09-18 | 17 |
| 2026-09-19 | 21 |
| 2026-09-22 | 22 |

### affine-registrations-per-month — Affine (SN120): Miner registrations per month (chain events)

- Unit: registrations/month · kind: rate · source type: chain · shape: **flat** · points: 6 (2026-04-30 -> 2026-09-24, 11 -> 68)
- Note: affine.io/api/v1/market/reg-history (team API, sourced from taomarketcap chain data). Sep 2026 is partial (24 days). Ramped Apr-Jun 2026 then flat at ~90/month since June. Current reg cost 0.629 tao; 252 miners, 4 validators on the metagraph (subnetradar).
- Source: https://affine.io/api/v1/market/reg-history (verified=True)

| date | value |
|---|---|
| 2026-04-30 | 11 |
| 2026-05-31 | 50 |
| 2026-06-30 | 89 |
| 2026-07-31 | 91 |
| 2026-08-31 | 88 |
| 2026-09-24 | 68 |

### teutonic-ii-king-loss — Teutonic (SN3): Teutonic-II-110B king eval loss at each coronation (challenger loss that won)

- Unit: nats/token (lower is better) · kind: rate · source type: team · shape: **falling** · points: 28 (2026-08-24 -> 2026-09-23, 7.89 -> 0.932)
- Note: teutonic.ai/dashboard.json history (accepted duels). Genesis 21 Aug 2026 was a random-init 110B model (loss ~12.5). Loss fell 12.5 -> 2.05 by 2 Sep on the original mix, then the eval dataset mix changed (24 Aug, 27 Aug, 2 Sep, 9 Sep, 15 Sep, 18 Sep, 19 Sep, 21 Sep, 22 Sep; eval_n 2000 -> 30000, required delta 0.5 -> 0.003), so cross-date comparisons after 9 Sep are on a different, easier mix (1.06 -> 0.93). Falling is the good direction here. 28 coronations, 29 reigns.
- Source: https://teutonic.ai/dashboard.json (verified=True)

| date | value |
|---|---|
| 2026-08-24 | 7.89 |
| 2026-08-24 | 6.61 |
| 2026-08-24 | 4.03 |
| 2026-08-25 | 3.38 |
| 2026-08-27 | 2.51 |
| 2026-08-27 | 2.56 |
| 2026-08-28 | 2.35 |
| 2026-08-29 | 2.24 |
| 2026-09-02 | 2.05 |
| 2026-09-08 | 2.04 |
| 2026-09-09 | 1.06 |
| 2026-09-10 | 1.05 |
| 2026-09-10 | 1.03 |
| 2026-09-11 | 1.02 |
| 2026-09-14 | 1.09 |
| 2026-09-17 | 0.909 |
| 2026-09-21 | 0.948 |
| 2026-09-21 | 0.929 |
| 2026-09-22 | 0.937 |
| 2026-09-23 | 0.932 |

(subsampled; all 28 points are in training.json)

### teutonic-ii-piqa — Teutonic (SN3): Teutonic-II king PIQA accuracy

- Unit: percent · kind: rate · source type: team · shape: **rising** · points: 5 (2026-08-21 -> 2026-09-23, 52.8 -> 83.1)
- Note: Team king-benchmark job (results.json on R2, read via the dashboard bundle). Reign 0 (random init) 52.8 = chance; jump to 74.4 at reign 8 (27 Aug) coincides with the first non-random-looking king, i.e. pretrained weights very likely entered. Flat at ~83 since 8 Sep.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-08-21 | 52.8 |
| 2026-08-27 | 74.4 |
| 2026-09-02 | 82.8 |
| 2026-09-08 | 83.2 |
| 2026-09-23 | 83.1 |

### teutonic-ii-gsm8k — Teutonic (SN3): Teutonic-II king GSM8K accuracy

- Unit: percent · kind: rate · source type: team · shape: **peaked** · points: 7 (2026-08-21 -> 2026-09-23, 0 -> 63.7)
- Note: Same source. Peaked 74.3 on 9 Sep, then the king that won on the new eval mix (12 Sep) scored 12 points lower; not recovered by 23 Sep.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-08-21 | 0 |
| 2026-08-27 | 14.3 |
| 2026-08-28 | 53.9 |
| 2026-09-02 | 72.8 |
| 2026-09-09 | 74.3 |
| 2026-09-12 | 61.9 |
| 2026-09-23 | 63.7 |

### teutonic-ii-mmlu — Teutonic (SN3): Teutonic-II king MMLU accuracy

- Unit: percent · kind: rate · source type: team · shape: **rising** · points: 4 (2026-09-08 -> 2026-09-23, 72.4 -> 74.2)
- Note: Same source; MMLU only benchmarked from 8 Sep. Small (+1.8) rise over two weeks.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-09-08 | 72.4 |
| 2026-09-12 | 71.5 |
| 2026-09-17 | 73.1 |
| 2026-09-23 | 74.2 |

### teutonic-ii-arc-easy — Teutonic (SN3): Teutonic-II king ARC-Easy accuracy

- Unit: percent · kind: rate · source type: team · shape: **rising** · points: 4 (2026-08-21 -> 2026-09-23, 25.2 -> 83.5)
- Note: Same source. Chance -> 84 in two weeks, flat since 5 Sep.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-08-21 | 25.2 |
| 2026-08-27 | 72.3 |
| 2026-09-05 | 84.2 |
| 2026-09-23 | 83.5 |

### teutonic-ii-humaneval — Teutonic (SN3): Teutonic-II king HumanEval pass@1

- Unit: percent · kind: rate · source type: team · shape: **rising** · points: 3 (2026-09-09 -> 2026-09-23, 23.4 -> 44.3)
- Note: Same source; first benchmarked 9 Sep. 23 -> 45 in one week then flat.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-09-09 | 23.4 |
| 2026-09-17 | 45.1 |
| 2026-09-23 | 44.3 |

### teutonic-ii-math500 — Teutonic (SN3): Teutonic-II king MATH-500 accuracy

- Unit: percent · kind: rate · source type: team · shape: **rising** · points: 3 (2026-08-21 -> 2026-09-23, 0 -> 42)
- Note: Same source. Three points only.
- Source: https://pub-c982d552b8044578b4a79e653700ec73.r2.dev/king-benchmark-daily/all-kings/results.json (verified=True)

| date | value |
|---|---|
| 2026-08-21 | 0 |
| 2026-09-08 | 33 |
| 2026-09-23 | 42 |

### teutonic-ii-coronations-cumulative — Teutonic (SN3): Cumulative Teutonic-II coronations since genesis

- Unit: coronations · kind: cumulative · source type: team · shape: **rising** · points: 17 (2026-08-24 -> 2026-09-23, 4 -> 28)
- Note: From dashboard.json history (accepted=true). Rising by construction; the rate fell from 11 in the first 9 days to ~1 every 3 days after 12 Sep, with a burst of 4 on 21 Sep after another dataset change.
- Source: https://teutonic.ai/dashboard.json (verified=True)

| date | value |
|---|---|
| 2026-08-24 | 4 |
| 2026-08-25 | 6 |
| 2026-08-27 | 9 |
| 2026-08-28 | 10 |
| 2026-08-29 | 11 |
| 2026-09-02 | 12 |
| 2026-09-05 | 13 |
| 2026-09-08 | 14 |
| 2026-09-09 | 16 |
| 2026-09-10 | 18 |
| 2026-09-11 | 19 |
| 2026-09-12 | 20 |
| 2026-09-14 | 21 |
| 2026-09-17 | 22 |
| 2026-09-21 | 26 |
| 2026-09-22 | 27 |
| 2026-09-23 | 28 |

### teutonic-ii-duels-per-day — Teutonic (SN3): Teutonic-II evaluations (duels) completed per day

- Unit: duels/day · kind: rate · source type: team · shape: **peaked** · points: 28 (2026-08-24 -> 2026-09-23, 13 -> 1)
- Note: Count of dashboard.json history entries per UTC day (659 total: 560 rejected, 71 error, 28 accepted; 31 unique coldkeys; stats.completed_evaluations 644). Peaked 10 Sep, single digits since.
- Source: https://teutonic.ai/dashboard.json (verified=True)

| date | value |
|---|---|
| 2026-08-24 | 13 |
| 2026-08-25 | 24 |
| 2026-08-27 | 32 |
| 2026-08-28 | 57 |
| 2026-08-30 | 37 |
| 2026-08-31 | 8 |
| 2026-09-02 | 36 |
| 2026-09-03 | 32 |
| 2026-09-04 | 31 |
| 2026-09-06 | 25 |
| 2026-09-07 | 30 |
| 2026-09-09 | 44 |
| 2026-09-10 | 80 |
| 2026-09-11 | 22 |
| 2026-09-13 | 2 |
| 2026-09-14 | 6 |
| 2026-09-16 | 1 |
| 2026-09-17 | 5 |
| 2026-09-22 | 1 |
| 2026-09-23 | 1 |

(subsampled; all 28 points are in training.json)

### iota-train-at-home-downloads — IOTA (SN9): Train-at-Home client downloads, cumulative

- Unit: downloads · kind: cumulative · source type: team · shape: **rising** · points: 5 (2026-02-24 -> 2026-09-24, 2,100 -> 45,916)
- Note: Latest value from the team stats API (train_at_home_downloads). Historical points: archived overview pages (Wayback 14 May and 7 Jun 2026, values baked into HTML) and team posts (24 Feb 2026 "2100+ downloads", week 2 of launch; 20 Apr 2026 dashboard launch, 23K+ per team roster note, post id not verified). Growth slowed: 32.4K -> 32.9K in the 3.5 weeks May-Jun, then 32.9K -> 45.9K in 3.5 months.
- Source: https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics; https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview; https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview; https://x.com/IOTA_SN9/status/2026634822026715565; https://x.com/IOTA_SN9/status/2046261387437949259 (verified=False/True)

| date | value | source |
|---|---|---|
| 2026-02-24 | 2,100 | https://x.com/IOTA_SN9/status/2026634822026715565 (team post, unverified) |
| 2026-04-20 | 23,000 | https://x.com/IOTA_SN9/status/2046261387437949259 (team post, unverified) |
| 2026-05-14 | 32,400 | https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-06-07 | 32,940 | https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-09-24 | 45,916 | https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics (verified) |

### iota-tokens-processed — IOTA (SN9): Tokens processed across all IOTA runs, cumulative

- Unit: tokens · kind: cumulative · source type: team · shape: **rising** · points: 3 (2026-05-14 -> 2026-09-24, 16.77T -> 19,816,428,784,000)
- Note: Team stats API tokens_processed; earlier values from archived overview pages. 16.77T -> 19.82T in 4.3 months (~0.7T/month).
- Source: https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics; https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview; https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview (verified=True)

| date | value | source |
|---|---|---|
| 2026-05-14 | 16.77T | https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-06-07 | 17.10T | https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-09-24 | 19,816,428,784,000 | https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics (verified) |

### iota-completed-experiments — IOTA (SN9): Completed experiments counter on the overview dashboard

- Unit: experiments · kind: cumulative · source type: team · shape: **falling** · points: 3 (2026-05-14 -> 2026-09-24, 730 -> 218)
- Note: Counter went 730 (14 May) -> 798 (7 Jun) -> 218 (24 Sep); the drop means the counter was reset or redefined (unique_compute_nodes is also 0 now), not that experiments were undone. Not usable as a growth series.
- Source: https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics; https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview; https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview (verified=True)

| date | value | source |
|---|---|---|
| 2026-05-14 | 730 | https://web.archive.org/web/20260514224455id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-06-07 | 798 | https://web.archive.org/web/20260607081623id_/https://iota.macrocosmos.ai/overview (verified) |
| 2026-09-24 | 218 | https://iota-web.api.macrocosmos.ai/mainnet/v1/stats_overview_metrics (verified) |

### iota-orion-16b-v8-tokens — IOTA (SN9): Orion-16B run v8: cumulative training tokens

- Unit: tokens · kind: cumulative · source type: team · shape: **rising** · points: 13 (2026-07-21 -> 2026-09-10, 1.10B -> 200.50B)
- Note: Team run dashboard throughput chart (cumulative-token customdata). 116 consumer-GPU nodes. Launched 21 Jul 2026, rolled back to epoch 19 on 24 Jul, ended ~10 Sep at 200B tokens; team post 23 Sep says ~80k tok/s, ~20% MFU, ~18 B200-equivalent. Steady ~3.5B tokens/day Jul-Aug, slowing to ~2.4B/day in Sep.
- Source: https://iota.macrocosmos.ai/api/data/research/orion-16b-v8_rollback_epoch19_20260724_172500/ov_throughput.json (verified=True)

| date | value |
|---|---|
| 2026-07-21 | 1.10B |
| 2026-07-22 | 4.20B |
| 2026-07-24 | 6.40B |
| 2026-07-27 | 16.70B |
| 2026-07-31 | 35.10B |
| 2026-08-05 | 57.90B |
| 2026-08-10 | 78.70B |
| 2026-08-16 | 101.10B |
| 2026-08-20 | 120.40B |
| 2026-08-25 | 145.60B |
| 2026-08-31 | 173.70B |
| 2026-09-05 | 188.70B |
| 2026-09-10 | 200.50B |

### iota-orion-16b-v9-tokens — IOTA (SN9): Orion-16B run v9: cumulative training tokens

- Unit: tokens · kind: cumulative · source type: team · shape: **rising** · points: 14 (2026-09-11 -> 2026-09-24, 200,800,000 -> 24.00B)
- Note: Team run dashboard throughput chart. Llama-3.2-style 16.2B params, 44 layers, 134 active nodes (RTX 4090/5090). Started 11 Sep 2026; ~2B tokens/day, accelerating to ~3B/day in the last week.
- Source: https://iota.macrocosmos.ai/api/data/research/orion-16b-v9/ov_throughput.json (verified=True)

| date | value |
|---|---|
| 2026-09-11 | 200,800,000 |
| 2026-09-12 | 879,000,000 |
| 2026-09-13 | 1.50B |
| 2026-09-14 | 3.00B |
| 2026-09-15 | 5.10B |
| 2026-09-16 | 7.00B |
| 2026-09-17 | 8.90B |
| 2026-09-18 | 11.10B |
| 2026-09-19 | 12.20B |
| 2026-09-20 | 13.90B |
| 2026-09-21 | 16.20B |
| 2026-09-22 | 19.10B |
| 2026-09-23 | 22.00B |
| 2026-09-24 | 24.00B |

### iota-orion-16b-v9-loss — IOTA (SN9): Orion-16B run v9: training loss, moving average

- Unit: loss (lower is better) · kind: rate · source type: team · shape: **falling** · points: 14 (2026-09-11 -> 2026-09-24, 12.2 -> 2.8)
- Note: Team run dashboard ov_loss_ma stream. 12.2 at start (11 Sep) -> 2.80 on 24 Sep. Falling is the good direction; the curve flattened around 2.8-2.9 after 19 Sep.
- Source: https://iota.macrocosmos.ai/api/data/research/orion-16b-v9/ov_loss_ma.ndjson (verified=True)

| date | value |
|---|---|
| 2026-09-11 | 12.2 |
| 2026-09-12 | 5.84 |
| 2026-09-13 | 4.92 |
| 2026-09-14 | 4.42 |
| 2026-09-15 | 3.79 |
| 2026-09-16 | 3.34 |
| 2026-09-17 | 3.18 |
| 2026-09-18 | 3.13 |
| 2026-09-19 | 2.96 |
| 2026-09-20 | 2.95 |
| 2026-09-21 | 2.94 |
| 2026-09-22 | 2.86 |
| 2026-09-23 | 2.86 |
| 2026-09-24 | 2.8 |

### iota-orion-model-scale — IOTA (SN9): Largest parameter count trained in a public IOTA run

- Unit: parameters · kind: rate · source type: team · shape: **peaked** · points: 4 (2026-06-01 -> 2026-09-11, 100.00B -> 16.20B)
- Note: Team posts: 3 Jun 2026 128-node 3B run (Jacob, 5 TAO cost); 1 Jun 2026 Orion-100B demo (65% efficiency on 48 A100s, 1.1B tokens; centralized-ish hardware); 27 Jul and 11 Sep Orion-16B v8/v9 on consumer GPUs. The 100B was a short demo, the sustained decentralized runs are 16B. Not a clean rising series.
- Source: https://iota.macrocosmos.ai/api/data/research/orion-16b-v9/ov_config.json; https://x.com/IOTA_SN9/status/2077790675932680220; https://x.com/MacrocosmosAI/status/2061493162582118695; https://x.com/const_reborn/status/2062258366739550618 (verified=False/True)

| date | value | source |
|---|---|---|
| 2026-06-01 | 100.00B | https://x.com/MacrocosmosAI/status/2061493162582118695 (team post, unverified) |
| 2026-06-03 | 3.00B | https://x.com/const_reborn/status/2062258366739550618 (team post, unverified) |
| 2026-07-27 | 16.20B | https://x.com/IOTA_SN9/status/2077790675932680220 (team post, unverified) |
| 2026-09-11 | 16.20B | https://iota.macrocosmos.ai/api/data/research/orion-16b-v9/ov_config.json (verified) |

### gradients-hf-models-per-month — Gradients (SN56): Fine-tuned models pushed to the rayonlabs Hugging Face org per month

- Unit: models/month · kind: rate · source type: third-party · shape: **peaked** · points: 12 (2024-12-01 -> 2025-11-01, 52 -> 3)
- Note: Counted from HF API createdAt over 2,219 models (28,922 downloads all time). Rose to 477 in Oct 2025 then stopped (3 in Nov 2025, 0 since): outputs moved off the public org, so this is a proxy that died, not a demand signal. Team API network/status shows 25,145 successful jobs total but no history.
- Source: https://huggingface.co/api/models?author=rayonlabs&expand[]=createdAt&expand[]=downloadsAllTime (verified=True)

| date | value |
|---|---|
| 2024-12-01 | 52 |
| 2025-01-01 | 76 |
| 2025-02-01 | 78 |
| 2025-03-01 | 123 |
| 2025-04-01 | 94 |
| 2025-05-01 | 328 |
| 2025-06-01 | 264 |
| 2025-07-01 | 220 |
| 2025-08-01 | 140 |
| 2025-09-01 | 364 |
| 2025-10-01 | 477 |
| 2025-11-01 | 3 |

### albedo-swebench-verified-by-king — Albedo (SN97): SWE-bench Verified resolved rate of the reigning Albedo king (500 tasks)

- Unit: percent · kind: rate · source type: team · shape: **U** · points: 22 (2026-06-24 -> 2026-09-20, 58.2 -> 61.2)
- Note: albedo.tech/data/model-scores.json (team benchmark job), dated by the king repo createdAt on Hugging Face (dendriteholdings/albedo-*-king-<numeral>). Kings I-CXV are labelled by the team as scoring below genesis; mid-July kings scored 2-4% (reward-hacked the judge). Dates for kings between listed coronations are interpolated from monthly counts (Jun 41, Jul 56, Aug 19, Sep 6) and are +/-2 days. Genesis 58.2 -> 2.4 (24 Jul) -> 61.2 (CXXVII, 20 Sep); GLM-5.2 reference 73.8. Honest shape: U, current king +3 over genesis after 3 months.
- Source: https://albedo.tech/data/model-scores.json (verified=True)

| date | value |
|---|---|
| 2026-06-24 | 58.2 |
| 2026-06-24 | 38.4 |
| 2026-06-25 | 26.4 |
| 2026-07-03 | 32.2 |
| 2026-07-08 | 26.8 |
| 2026-07-19 | 12 |
| 2026-07-24 | 3.8 |
| 2026-07-26 | 31 |
| 2026-07-31 | 41.4 |
| 2026-08-04 | 32 |
| 2026-08-10 | 43.4 |
| 2026-08-12 | 45.8 |
| 2026-08-25 | 0 |
| 2026-08-26 | 55.4 |
| 2026-08-27 | 59.2 |
| 2026-08-29 | 60.4 |
| 2026-09-01 | 40.4 |
| 2026-09-11 | 59.8 |
| 2026-09-14 | 56.2 |
| 2026-09-20 | 61.2 |

(subsampled; all 22 points are in training.json)

### albedo-kings-cumulative — Albedo (SN97): Cumulative Albedo kings crowned

- Unit: kings · kind: cumulative · source type: team · shape: **rising** · points: 14 (2026-06-24 -> 2026-09-20, 8 -> 127)
- Note: Coronation dates from HF createdAt of the king repos. Rising by construction, but the rate collapsed: 41 in the first week (24-30 Jun), 56 in Jul, 19 in Aug, 6 in Sep (last 20 Sep).
- Source: https://huggingface.co/api/models?author=dendriteholdings&expand[]=createdAt (verified=True)

| date | value |
|---|---|
| 2026-06-24 | 8 |
| 2026-06-25 | 10 |
| 2026-06-30 | 41 |
| 2026-07-31 | 97 |
| 2026-08-26 | 116 |
| 2026-08-27 | 118 |
| 2026-08-28 | 119 |
| 2026-08-29 | 121 |
| 2026-09-01 | 122 |
| 2026-09-05 | 123 |
| 2026-09-06 | 124 |
| 2026-09-11 | 125 |
| 2026-09-14 | 126 |
| 2026-09-20 | 127 |

### albedo-kings-per-month — Albedo (SN97): Albedo coronations per month

- Unit: kings/month · kind: rate · source type: team · shape: **falling** · points: 4 (2026-06-30 -> 2026-09-24, 41 -> 6)
- Note: Same HF createdAt source; Sep partial (24 days). Fewer coronations can mean a harder-to-beat king (judge score rising, see albedo-judge-score) or fewer serious miners.
- Source: https://huggingface.co/api/models?author=dendriteholdings&expand[]=createdAt (verified=True)

| date | value |
|---|---|
| 2026-06-30 | 41 |
| 2026-07-31 | 56 |
| 2026-08-31 | 19 |
| 2026-09-24 | 6 |

### albedo-judge-score — Albedo (SN97): Albedo judge score of the winning challenger at coronation

- Unit: score (0-1) · kind: rate · source type: team · shape: **rising** · points: 3 (2026-09-11 -> 2026-09-20, 0.679 -> 0.762)
- Note: albedo.tech/data/dashboard.json coronation records: CXXV 0.635 -> 0.679 (11 Sep), CXXVI 0.666 -> 0.724 (14 Sep), CXXVII 0.725 -> 0.762 (20 Sep). Only the last three coronations carry the score; three points.
- Source: https://albedo.tech/data/dashboard.json (verified=True)

| date | value |
|---|---|
| 2026-09-11 | 0.679 |
| 2026-09-14 | 0.724 |
| 2026-09-20 | 0.762 |

### reliquary-checkpoint-pushes-per-day — Reliquary (SN81): Checkpoint commits per day to the live RL model repo (ReliquaryForge/qwen3-4b-base-dapo-v4)

- Unit: commits/day · kind: rate · source type: third-party · shape: **rising** · points: 38 (2026-08-18 -> 2026-09-24, 4 -> 51)
- Note: Counted from the HF commits API (2,161 commits, 18 Aug-24 Sep 2026). Each commit is a validator-published checkpoint after a window; 4/day -> ~120/day. 22 Sep (17) looks like an outage; 24 Sep partial. Team API today: window_n 46,859, checkpoint_n 3,020, 26 active hotkeys, 336 seats/window, 256 registered operators. Repo has 980,090 downloads all time (287K in 30 d), mostly miners pulling checkpoints.
- Source: https://huggingface.co/api/models/ReliquaryForge/qwen3-4b-base-dapo-v4/commits/main (verified=True)

| date | value |
|---|---|
| 2026-08-18 | 4 |
| 2026-08-20 | 16 |
| 2026-08-22 | 23 |
| 2026-08-24 | 26 |
| 2026-08-26 | 41 |
| 2026-08-28 | 76 |
| 2026-08-30 | 51 |
| 2026-09-01 | 45 |
| 2026-09-03 | 55 |
| 2026-09-05 | 36 |
| 2026-09-06 | 58 |
| 2026-09-08 | 38 |
| 2026-09-10 | 29 |
| 2026-09-12 | 62 |
| 2026-09-14 | 85 |
| 2026-09-16 | 101 |
| 2026-09-18 | 112 |
| 2026-09-20 | 100 |
| 2026-09-22 | 17 |
| 2026-09-24 | 51 |

(subsampled; all 38 points are in training.json)

### connito-baseline-loss — ConnitoAI (SN102): Baseline (reference model) training loss on the Connito dashboard

- Unit: loss (lower is better) · kind: rate · source type: team · shape: **U** · points: 10 (2026-09-05 -> 2026-09-24, 1.71 -> 2.17)
- Note: baseline_loss_history from the team leaderboard API (256 points, 5-24 Sep 2026; fetched in-browser because Cloudflare blocks curl). Fell 1.71 -> 1.30 by 14 Sep, then jumped to 3.95 on 16 Sep (dataset switched to TxT360) and is 2.17 today, higher than a week ago. Not improving on the current data. Dashboard cycle #17,440 (524-block cycles); 256 miners, 5 validators (2 online); 167 successful commits (65.2%). The homepage "15781" is a static sample number.
- Source: https://dashboard-api-v2.connito.ai/api/v3/leaderboard (verified=True)

| date | value |
|---|---|
| 2026-09-05 | 1.71 |
| 2026-09-08 | 1.42 |
| 2026-09-12 | 1.37 |
| 2026-09-14 | 1.3 |
| 2026-09-16 | 3.95 |
| 2026-09-17 | 2.06 |
| 2026-09-18 | 2.1 |
| 2026-09-21 | 1.81 |
| 2026-09-23 | 2.04 |
| 2026-09-24 | 2.17 |

### cascade-participants-per-round — cascade (SN91): Miners participating per round (daily mean over rounds)

- Unit: miners · kind: rate · source type: team · shape: **rising** · points: 65 (2026-07-20 -> 2026-09-24, 28 -> 235)
- Note: Hippius receipts index (361 rounds, 20 Jul-24 Sep 2026; dated by receipt published_at). 28 on day one -> ~220 by 28 Jul -> plateau 215-235 with a dip to 172 on 2 Sep. Rising in the first week, flat since; near the 256-UID cap.
- Source: https://s3.hippius.com/cascade-manifests/receipts/index.json (verified=True)

| date | value |
|---|---|
| 2026-07-20 | 28 |
| 2026-07-24 | 131 |
| 2026-07-28 | 218 |
| 2026-08-01 | 214 |
| 2026-08-04 | 210 |
| 2026-08-08 | 230 |
| 2026-08-11 | 214 |
| 2026-08-15 | 223 |
| 2026-08-18 | 222 |
| 2026-08-21 | 212 |
| 2026-08-25 | 202 |
| 2026-08-28 | 222 |
| 2026-08-31 | 191 |
| 2026-09-04 | 182 |
| 2026-09-07 | 210 |
| 2026-09-11 | 194 |
| 2026-09-14 | 213 |
| 2026-09-17 | 205 |
| 2026-09-21 | 217 |
| 2026-09-24 | 235 |

(subsampled; all 65 points are in training.json)

### cascade-king-error — cascade (SN91): King model geometric-mean forecast error (lower is better)

- Unit: geomean error · kind: rate · source type: team · shape: **U** · points: 64 (2026-07-20 -> 2026-09-24, 0.3206 -> 0.3236)
- Note: Same receipts. 0.32 (genesis baseline) -> 0.05 on 23 Jul -> back to 0.28-0.34 after the eval task mix changed (28 Jul, 5 Sep). Today 0.3236, i.e. no better than the genesis baseline on the current mix. Not improving.
- Source: https://s3.hippius.com/cascade-manifests/receipts/index.json (verified=True)

| date | value |
|---|---|
| 2026-07-20 | 0.3206 |
| 2026-07-24 | 0.0548 |
| 2026-07-30 | 0.2512 |
| 2026-08-02 | 0.1759 |
| 2026-08-05 | 0.2452 |
| 2026-08-09 | 0.2468 |
| 2026-08-12 | 0.235 |
| 2026-08-15 | 0.1775 |
| 2026-08-19 | 0.1774 |
| 2026-08-22 | 0.2406 |
| 2026-08-25 | 0.2418 |
| 2026-08-28 | 0.2473 |
| 2026-09-01 | 0.2163 |
| 2026-09-04 | 0.3326 |
| 2026-09-07 | 0.3378 |
| 2026-09-11 | 0.2993 |
| 2026-09-14 | 0.3087 |
| 2026-09-17 | 0.3276 |
| 2026-09-21 | 0.3322 |
| 2026-09-24 | 0.3236 |

(subsampled; all 64 points are in training.json)

### cascade-dethronements-cumulative — cascade (SN91): Cumulative king dethronements

- Unit: dethronements · kind: cumulative · source type: team · shape: **rising** · points: 65 (2026-07-20 -> 2026-09-24, 1 -> 38)
- Note: Same receipts; count of rounds with dethroned=true. Rising by construction; 38 in 66 days, most in late Jul-Aug.
- Source: https://s3.hippius.com/cascade-manifests/receipts/index.json (verified=True)

| date | value |
|---|---|
| 2026-07-20 | 1 |
| 2026-07-24 | 3 |
| 2026-07-28 | 5 |
| 2026-08-01 | 9 |
| 2026-08-04 | 15 |
| 2026-08-08 | 16 |
| 2026-08-11 | 18 |
| 2026-08-15 | 20 |
| 2026-08-18 | 20 |
| 2026-08-21 | 20 |
| 2026-08-25 | 23 |
| 2026-08-28 | 26 |
| 2026-08-31 | 26 |
| 2026-09-04 | 26 |
| 2026-09-07 | 29 |
| 2026-09-11 | 32 |
| 2026-09-14 | 35 |
| 2026-09-17 | 35 |
| 2026-09-21 | 35 |
| 2026-09-24 | 38 |

(subsampled; all 65 points are in training.json)

### capcomp-run-counter — Capcomp (SN103): Highest run number published to Hugging Face by date

- Unit: run id · kind: cumulative · source type: third-party · shape: **rising** · points: 6 (2026-09-03 -> 2026-09-08, 421 -> 426)
- Note: capcomp/sn103-run-<n> datasets, HF createdAt. Runs 417-426 published 3-8 Sep 2026 (about one per day), nothing since 8 Sep. 856 downloads all time across them; one adapter-pool model with 0 downloads. Very small subnet (mcap ~$209K).
- Source: https://huggingface.co/api/datasets?author=capcomp&expand[]=createdAt (verified=True)

| date | value |
|---|---|
| 2026-09-03 | 421 |
| 2026-09-04 | 422 |
| 2026-09-05 | 423 |
| 2026-09-06 | 424 |
| 2026-09-07 | 425 |
| 2026-09-08 | 426 |

## Single numbers

| subnet | netuid | metric | value | date | source | type | note |
|---|---|---|---|---|---|---|---|
| Affine | 120 | Alpha price | 0.04644 tao | 2026-09-24 | https://affine.io/api/v1/snapshot | team | snapshot.market; TAO $293.4; miners_tao_per_day 137.1 |
| Affine | 120 | Miner emissions | 137.1 tao/day | 2026-09-24 | https://affine.io/api/v1/snapshot | team | miners_tao_per_day from team snapshot |
| Affine | 120 | Paid crowns in current 72-hour payout window | 1 crowns | 2026-09-24 | https://affine.io/api/v1/snapshot | team | Only reign 21 currently paid; payout rule is a 72-hour window |
| Affine | 120 | Advertised king payout | 50,000 USD/week | 2026-09-24 | https://x.com/affine_io/status/2091498650119901215 | team | Team post 23 Aug 2026: "$50K/week per slot". Not verifiable on-chain from this sweep |
| Affine | 120 | Subnet market cap | 58,600,000 USD | 2026-09-24 | https://subnetradar.com/subnet/120 | third-party | emission share 6.49%; 252 miners / 4 validators; 500 GitHub commits in 30 d |
| Affine | 120 | GitHub stars, AffineFoundation/affine | 2 stars | 2026-09-24 | https://api.github.com/repos/AffineFoundation/affine | third-party |  |
| Affine | 120 | Hugging Face downloads, AffineFoundation datasets (all time) | 33,674 downloads | 2026-09-24 | https://huggingface.co/api/datasets?author=AffineFoundation&expand[]=downloadsAllTime | third-party | Models under the org: 0 downloads all time. Miner org vera6: 217 models, 24,512 downloads all time. |
| Teutonic | 3 | Active registrations | 246 miners | 2026-09-24 | https://teutonic.ai/dashboard.json | team | dashboard.json stats |
| Teutonic | 3 | King payout rate | 24.6 alpha/hour | 2026-09-24 | https://teutonic.ai/dashboard.json | team | dashboard.json king_payout; about $221/h at alpha 0.0305 tao and TAO ~$293 |
| Teutonic | 3 | Teutonic-I duels (run 1, 2 Jun-10 Aug 2026) | 2,163 duels | 2026-08-10 | https://teutonic.ai/paper | team | Teutonic-I paper: 2,163 duels, 203 coronations, 73 coldkeys, 22 namespaces, acceptance 9.3%, best King #191, 94 benchmarked checkpoints, 62.28% 11-task mean, 640 GPU-hours eval, 4.24T eval tokens |
| Teutonic | 3 | Teutonic-I coronations | 203 coronations | 2026-08-10 | https://teutonic.ai/paper | team | Same paper; run 2 (Teutonic-II) is at 28 coronations after 33 days |
| Teutonic | 3 | Hugging Face downloads, dendriteholdings teutonic repos (all time) | 9,840 downloads | 2026-09-24 | https://huggingface.co/api/models?author=dendriteholdings&expand[]=downloadsAllTime | third-party | 155 teutonic repos; genesis 5,378; Teutonic-I final 354. Data source org IFM: 1.04M model downloads (not a subnet metric). |
| Teutonic | 3 | Subnet market cap | 54,200,000 USD | 2026-09-24 | https://subnetradar.com/subnet/3 | third-party | emission share 4.16%; 249 miners; 658 GitHub commits/30 d; unarbos/teutonic 12 stars |
| IOTA | 9 | Active miners across the 9 Train-at-Home runs | 742 miners | 2026-09-24 | https://iota-web.api.macrocosmos.ai/mainnet/v1/runs_occupancy | team | 73+23+120+120+69+73+120+120+24; several runs are at the 120 cap |
| IOTA | 9 | Orion-16B v9 active nodes | 134 nodes | 2026-09-24 | https://iota.macrocosmos.ai/api/data/research/orion-16b-v9/ov_config.json | team | RTX 4090/5090 class consumer GPUs |
| IOTA | 9 | Countries with Train-at-Home participants | 16 countries | 2026-06-30 | https://x.com/IOTA_SN9/status/2071985168966184998 | team | Team map post; chart saved |
| IOTA | 9 | Hugging Face downloads, macrocosm-os datasets (all time) | 390,075 downloads | 2026-09-24 | https://huggingface.co/api/datasets?author=macrocosm-os&expand[]=downloadsAllTime | third-party | Dominated by code-parrot-github-code mirror (336,856); no IOTA model checkpoints on the org |
| IOTA | 9 | Subnet market cap | 45,300,000 USD | 2026-09-24 | https://subnetradar.com/subnet/9 | third-party | emission share 1.42%; 244 miners; realwork 52%; macrocosm-os/iota 34 stars |
| Gradients | 56 | Successful training jobs, all time | 25,145 jobs | 2026-09-24 | https://api.gradients.io/v1/network/status | team | number_of_jobs_success; no dated history exposed |
| Gradients | 56 | Tournament participants (14 Sep 2026 round) | 42 miners | 2026-09-14 | https://api.gradients.io/tournament/latest/details | team | text 18, image 12, environment 12; entry fees 0.7/0.4/0.6 TAO |
| Gradients | 56 | Paying users (team via third party) | 3,000 users | 2026-09-24 | https://simplytao.ai/subnets/56 | third-party | "over 3,000 paying users" - not verifiable; pricing page needs login |
| Gradients | 56 | Champion payout, first week | 24,500 USD | 2026-06-30 | https://x.com/gradients_ai/status/2071970881790677261 | team | Team post; chart saved |
| Gradients | 56 | Hugging Face downloads, rayonlabs org (all time) | 28,922 downloads | 2026-09-24 | https://huggingface.co/api/models?author=rayonlabs&expand[]=downloadsAllTime | third-party | 2,219 models |
| Gradients | 56 | Subnet market cap | 28,600,000 USD | 2026-09-24 | https://subnetradar.com/subnet/56 | third-party | emission share 1.96%; 242 miners; rayonlabs/G.O.D 37 stars |
| Albedo | 97 | Models evaluated | 1,499 models | 2026-09-24 | https://albedo.tech/data/dashboard.json | team | stats.evaluated; benchmark job counts: 404 jobs, 121 models, 39,904 task results |
| Albedo | 97 | Terminal-Bench 2.1, king CXXVII vs genesis | 39.3 percent | 2026-09-24 | https://albedo.tech/results/v1/manifest.json | team | genesis 44.9 -> CXXVII 39.3 (down); SWE-rebench CXXVII 30.0 (GLM-5.2 53.6); Multilingual 49.0 -> 41.7; Aider 49.8 -> 42.2. Only SWE-bench Verified improved. |
| Albedo | 97 | Training data claimed | 12,000,000 trajectories | 2026-07-13 | https://tao.media | third-party | 12M trajectories, 122,000 PRs, 3,000 repos per tao.media 13 Jul 2026 interview; not verifiable |
| Albedo | 97 | Hugging Face downloads, albedo king repos (all time) | 330,388 downloads | 2026-09-24 | https://huggingface.co/api/models?author=dendriteholdings&expand[]=downloadsAllTime | third-party | 122 repos; king CXX alone 51,023; dataset dendriteholdings/albedo 6,523 |
| Albedo | 97 | Subnet market cap | 9,400,000 USD | 2026-09-24 | https://subnetradar.com/subnet/97 | third-party | emission share 2.26%; 253 miners; unarbos/albedo 3 stars |
| Reliquary | 81 | Training windows completed | 46,859 windows | 2026-09-24 | https://reliqua.ai/api/status | team | window_n; the ecosystem doc had 29,522 earlier in Sep 2026 (undated) so roughly +17K in the month |
| Reliquary | 81 | Checkpoints published | 3,020 checkpoints | 2026-09-24 | https://reliqua.ai/api/status | team | checkpoint_n |
| Reliquary | 81 | Hugging Face downloads, live RL repo (all time) | 980,090 downloads | 2026-09-24 | https://huggingface.co/api/models/ReliquaryForge/qwen3-4b-base-dapo-v4?expand[]=downloadsAllTime | third-party | qwen3.5-4b-reliquary-v4 340,040; Reliquary-4B release 17 Sep 2026 |
| Reliquary | 81 | Subnet market cap | 9,600,000 USD | 2026-09-24 | https://subnetradar.com/subnet/81 | third-party | emission share shows 0.00% on subnetradar; 451 GitHub commits/30 d |
| ConnitoAI | 102 | Training cycles completed | 17,440 cycles | 2026-09-24 | https://dashboard-api-v2.connito.ai/api/v3/leaderboard | team | cycle counter; 524 blocks per cycle |
| ConnitoAI | 102 | Model parameters (compressed architecture) | 3.11B parameters | 2026-09-24 | https://dashboard-api-v2.connito.ai/api/v3/leaderboard | team | architecture 15.71B -> 3.11B; connito HF org: 2 models, 0 downloads |
| ConnitoAI | 102 | Subnet market cap | 2,500,000 USD | 2026-09-24 | https://subnetradar.com/subnet/102 | third-party | 143 GitHub commits/30 d |
| cascade | 91 | Subnet market cap | 2,900,000 USD | 2026-09-24 | https://subnetradar.com/subnet/91 | third-party | 756 GitHub commits/30 d |
| Capcomp | 103 | Hugging Face downloads, capcomp datasets (all time) | 856 downloads | 2026-09-24 | https://huggingface.co/api/datasets?author=capcomp&expand[]=downloadsAllTime | third-party | 10 run datasets |
| Capcomp | 103 | Subnet market cap | 209,000 USD | 2026-09-24 | https://subnetradar.com/subnet/103 | third-party |  |
| Refinery | 125 | Optimizer rounds published | 19 rounds | 2026-09-24 | https://refinery125.com/dashboard.json | team | 20 submissions, 16 scored, 7 active miners; fee 0.723 TAO/submission, 15.337 TAO deposited by 9 coldkeys; 41,328 alpha paid, 14,760 alpha burned |
| Refinery | 125 | Best submitted optimizer loss vs AdamW baseline | 2.568 loss | 2026-09-24 | https://refinery125.com/dashboard.json | team | AdamW baseline 2.7764 on FW2_smol_360M, 188k steps on a B200: 0.19 nats better, team-confirmed. Only one dated point. |
| Ralph | 40 | Public quantization rounds | 7 rounds | 2026-09-24 | https://ralphlabs.ai | team | 4 crown files for Qwen3-8B: 1-bit 1.16 GB, ternary 2.48 GB, 2.3-bpw 2.94 GB, 4-bit 4.61 GB; 3 judges. No dated series. |
| Quasar | 24 | Hugging Face downloads, silx-ai Quasar models (all time) | 60,236 downloads | 2026-09-24 | https://huggingface.co/api/models?author=silx-ai&expand[]=downloadsAllTime | third-party | Quasar-10B 39,860; Quasar-3B-A1B 12,787. Subnet emission 0% on subnetradar. |
| Feval | 47 | Hugging Face downloads, feval-center (all time) | 241 downloads | 2026-09-24 | https://huggingface.co/api/datasets?author=feval-center&expand[]=downloadsAllTime | third-party | Renamed GPUForge (github forgenet47/gpuforge); 1 model, 1 dataset |
| ChronoLLM | 38 | Subnet market cap | 10,800,000 USD | 2026-09-24 | https://subnetradar.com/subnet/38 | third-party | emission share 1.92%; price +71% on the week; chronollm.crunchdao.com is marketing only, no dashboard |
| Hone | 5 | Registered UIDs | 256 UIDs | 2026-09-24 | https://honedashboard.com/api/aggregates?validator=&window_minutes=1440 | team | Dashboard aggregates showed 0 attempts in the 24 h and 7 d windows for the tao.bot validator; no usable output metric |
| Occamy (reference model) | 0 | Hugging Face downloads, Accio-Lab/occamy-1.0-GGUF (all time) | 34,778 downloads | 2026-09-24 | https://huggingface.co/api/models/Accio-Lab/occamy-1.0-GGUF?expand[]=downloadsAllTime | third-party | Referenced in the Affine matrix as a comparison model; not a subnet |

## Not rising

Series above whose honest shape is flat, peaked, U or falling (falling is the good direction only where the note says "lower is better"):

- `affine-king-total` — U: genesis 56.3 -> 43.3 (15 Sep) -> 67.4 (19 Sep) -> 63.2 now; +7 over genesis after a month, below the 68.8 teacher
- `affine-king-bench` — U: 64.6 -> 43.6 -> 70.3 -> 67.8
- `affine-king-env` — U: 50.6 -> 43.2 -> 65.5 -> 60.3 (the one group clearly above genesis)
- `affine-king-terminal` — U: 78.6 -> 57.6 -> 91.6; noisy king to king
- `affine-king-math` — peaked: 44 -> 93.5 on the first visible king, oscillating 69-92 since
- `affine-king-tool-use` — U: 65.5 -> 24.9 -> 80.7 -> 72.3
- `affine-king-swebench-verified` — flat: 79.5 genesis, 73.7 now, one crash to 48.2
- `affine-king-terminal-bench-2` — flat: 36.4 genesis, 37.9 now
- `affine-verdicts-per-week` — peaked: 91 -> 151 -> 172 -> 123 -> 55 (partial week)
- `affine-registrations-per-month` — flat: ~90/month flat since June 2026
- `teutonic-ii-king-loss` — falling: 12.5 -> 0.93, but the eval mix changed nine times; falling = improving
- `teutonic-ii-gsm8k` — peaked: 74.3 peak on 9 Sep, 63.7 now after the eval-mix change
- `teutonic-ii-duels-per-day` — peaked: peaked 61 on 10 Sep, 1-3/day for the last week
- `iota-completed-experiments` — falling: 730 -> 798 -> 218; counter reset/redefined, unusable
- `iota-orion-16b-v9-loss` — falling: 12.2 -> 2.80, flattening at 2.8-2.9; falling = improving
- `iota-orion-model-scale` — peaked: 100B demo (Jun) then 16B sustained runs; not a rising scale story
- `gradients-hf-models-per-month` — peaked: 477 in Oct 2025 then 0; outputs left the public org
- `albedo-swebench-verified-by-king` — U: 58.2 -> 2.4 -> 61.2; +3 over genesis after three months
- `albedo-kings-per-month` — falling: 41 -> 56 -> 19 -> 6
- `connito-baseline-loss` — U: 1.71 -> 1.30 -> 3.95 (dataset switch) -> 2.17; higher than a week ago
- `cascade-king-error` — U: 0.32 -> 0.05 -> 0.32; back at the genesis baseline after eval changes

Also worth flagging: `affine-crowns-cumulative`, `teutonic-ii-coronations-cumulative`, `albedo-kings-cumulative` and `cascade-dethronements-cumulative` rise by construction; their rates (crowns per week/month) are falling or peaked in every case except Affine (9 crowns in the week of 14 Sep).

## Could not find

- Hone SN5: any output/usage metric (honedashboard.com aggregates 0 attempts for the default validator over 60/1440 min; no other JSON routes in the bundle; no team posts with numbers found in fxtwitter/X within budget).
- ChronoLLM SN38: any metric beyond market data (chronollm.crunchdao.com is a landing page; no dashboard, no HF org, no GitHub activity found).
- TPN SN65: any metric (site only; 0% emission on subnetradar; no dashboard, HF or GitHub numbers).
- Feval SN47 / GPUForge: usage or evaluation counts (GitHub repo has no metrics; HF org nearly empty).
- Refinery SN125 and Ralph SN40: dated history (dashboards expose current totals only; Wayback has no snapshots of refinery125.com/dashboard.json).
- Gradients SN56: dated job counts or revenue (api.gradients.io network/status has no history; pricing and dashboards need login; Wayback has no API snapshots; "3,000 paying users" is unverifiable).
- IOTA SN9: archived values for the 20 Apr 2026 dashboard launch (homepage rendered client-side, values show as dashes); the 23K+ figure comes from the roster note, post id unverified.
- Affine SN120: scores of hidden kings 1-10 (matrix hides them) and any pre-27 Aug 2026 series (older affine.io snapshots in Wayback are marketing pages without numbers).
- Teutonic SN3: per-duel history for Teutonic-I (run 1); only the paper totals are public. Teutonic-I paper figures saved as chart instead.
- arXiv: papers 2507.17766 / 2604.11947 (IOTA) not retrievable through the arXiv API today (empty results), so no parameter-scale history from papers.
- Hugging Face: no download counts for Affine kings (models are under miner orgs, e.g. vera6) or for Connito/Refinery outputs (0 downloads).

## X reads used

25 of 30 (5 calls, `max_results=10`, logged in `/tmp/sweep/xreads-training.log`):

| returned | query |
|---|---|
| 7 | `from:IOTA_SN9 OR from:MacrocosmosAI ("Train at Home" OR TAH) (downloads OR downloaded)` |
| 2 | `(from:MacrocosmosAI OR from:IOTA_SN9) "Train at Home" (downloads OR downloaded OR installs) -is:reply` |
| 8 | `from:gradients_ai (paying OR users OR customers OR "tasks" OR trained OR revenue) -is:reply -is:retweet` |
| 2 | `(from:const_reborn OR from:DendriteHQ OR from:tony_dendrite) (Teutonic OR Albedo) (loss OR MMLU OR coldkeys OR duels OR coronations OR downloads OR tokens OR miners) -is:reply -is:retweet` |
| 6 | `(from:MacrocosmosAI OR from:IOTA_SN9 OR from:mccrinbc) Orion (tokens OR "billion" OR MFU OR epoch OR loss) -is:reply -is:retweet` |

Known post ids were fetched through fxtwitter (free, not counted): affine_io 2065461449166835963, 2069810780019491188, 2091498650119901215, 2093008057261609231; DendriteHQ 2087169458402189560, 2088353208267858385, 2089729165721551040; const_reborn 2062258366739550618, 2091896286035235027; IOTA_SN9 2026634822026715565, 2046261387437949259, 2071985168966184998, 2077790675932680220; MacrocosmosAI 2061493162582118695, 2081791889674772618, 2102808976442458457; arbos_born 2050596495485800658.

## Charts saved

Directory: `/cursor/stores/bc-96ff1b3e-00b6-4da6-8a0d-5e72965f7e86/media/charts-gallery/`

| file | source URL | what it shows |
|---|---|---|
| teutonic-i-duels-and-loss-staircase.png | https://teutonic.ai/paper-assets/fig_duels_web.png | Teutonic-I paper figure: 2,163 duels, king loss staircase, 2 Jun-10 Aug 2026 |
| teutonic-i-vs-covenant-72b-benchmarks.jpg | https://x.com/DendriteHQ/status/2088353208267858385 | Teutonic-I vs Covenant-72B benchmark bars |
| teutonic-ii-110b-announcement.jpg | https://x.com/const_reborn/status/2091896286035235027 | Teutonic-II 110B launch card |
| affine-xxix-vs-qwen3-32b-benchmarks.jpg | https://x.com/affine_io/status/2065461449166835963 | AFFINE-XXIX vs Qwen3-32B, 12 Jun 2026 (+10.5 SWE-REBENCH) |
| affine-35b-iii-env-scores-jun-2026.jpg | https://x.com/affine_io/status/2069810780019491188 | AFFINE-35B-III environment scores, 24 Jun 2026 |
| iota-train-at-home-week2-feb-2026.jpg | https://x.com/IOTA_SN9/status/2026634822026715565 | Train-at-Home week 2 (2,100+ downloads), 24 Feb 2026 |
| iota-orion-100b-efficiency.jpg | https://x.com/MacrocosmosAI/status/2061493162582118695 | Orion-100B efficiency chart, 1 Jun 2026 |
| iota-128-node-run-dashboard-jun-2026.jpg | https://x.com/const_reborn/status/2062258366739550618 | 128-node 3B run dashboard, 3 Jun 2026 |
| iota-train-at-home-tps-mfu-jun-2026.jpg | https://x.com/IOTA_SN9/status/2065447149270110225 | Train-at-Home tokens/s and MFU speedups, 12 Jun 2026 |
| iota-train-at-home-16-countries-map.jpg | https://x.com/IOTA_SN9/status/2071985168966184998 | Train-at-Home participants in 16 countries, 30 Jun 2026 |
| iota-autoresearch-sweeps-jul-2026.jpg | https://x.com/mccrinbc/status/2077830752130011640 | 70+ autoresearch sweeps, 17 Jul 2026 |
| iota-orion-update-jul-2026.jpg | https://x.com/IOTA_SN9/status/2077790675932680220 | Orion-16B launch update, 27 Jul 2026 |
| iota-orion-16b-final-throughput-sep-2026.jpg | https://x.com/MacrocosmosAI/status/2102808976442458457 | Orion-16B v8 final throughput (~80k tok/s, ~20% MFU), 23 Sep 2026 |
| gradients-rl-env-tournament-scores-mar-2026.jpg | https://x.com/gradients_ai/status/2031491225132609601 | RL environment tournament scores, Mar 2026 |
| gradients-champion-payout-jun-2026.jpg | https://x.com/gradients_ai/status/2071970881790677261 | Champion payout $24.5K first week, 30 Jun 2026 |

Not saved: Teutonic paper SVG figures (fig_participation.svg etc.) — headless Chrome SVG rendering timed out; source https://teutonic.ai/paper-assets/. No chart images are posted by Albedo, Reliquary, Connito, cascade, Refinery, Ralph, Quasar, Capcomp, Feval, Hone, ChronoLLM or TPN; their dashboards render from the JSON routes cited above.