TRIPLE::PASS LABS  /  TallyLock  /  Registry and how to submit

Receipts Board

Every entry below was re-checked from its recorded runs when this page was built. The score is the average over completed runs with its 95% range, not the best run, and failed runs are counted. Tap an entry to see every run.

Entries
4
Runs recorded
20
Failed runs counted
0
Benchmarks
1

score (average of runs)95% rangebest single run

Benchmark 1

Colony Zero sealed set 4 (EXP-CZ-003a, revealed), board subset: 100 lowest task_ids per family

Harness colony-zero cz.exe probe (Season 0 build), llama.cpp b11539 Vulkan, RX 9070 XTPrompt template aaae0e27e713Dataset 8cfbd91929fbmax_tokens 512min_p 0.05samples 1seed attempt seed via --sample-seed-basetemperature 0.7template P3top_k 40top_p 0.95
  1. 1Quentin (Qwen3 0.6B, Q8_0)Qwen__Qwen3-0.6B-q8_00.294± 0.022best 0.312

    Quentin (Qwen3 0.6B, Q8_0) on Colony Zero sealed set 4, board subset

    Every run

    Score
    0.294 ± 0.022
    Best run
    0.312 (not the score)
    Runs
    5 recorded, 0 failed
    Model
    Qwen__Qwen3-0.6B-q8_0
    Trust
    anchored
    Source
    tallylock, by TriplePass Labs (Colony Zero founders)

    Record

    1. Plan lockedCommit 94e8139, command ./eval.sh quentin {seed}. Changing either starts a new plan.
    2. 5 runs recorded5 completed, 0 failed, 4 min 14 s in all. Every run is in the ledger.
    3. Instant timestamps10 of 10 verified from 2 public services from Nov 8, 2026 04:08 UTC to Nov 8, 2026 04:12 UTC.
    4. Bitcoin timestampEvery ledger update is in Bitcoin block 975,020, 975,023, 975,026. Check the blocks with tallylock check --online.

    Against the others here

    Ahead of Grace (Granite 4.0 H 350M, Q8_0) beyond noiseAhead of Quinn (Qwen3.5 0.8B, Q8_0) beyond noiseAhead of Graham (Granite 4.0 350M, Q8_0) beyond noise

    Claim line

    0.312 best of 5 runs; adjusted 0.294 ± 0.022 (95% CI, 5 completed, 0 failed); plan 40a62f35 (2 of 4 in repo); ledger 57cc2acf; tallylock check

    Plan 2 in its repo. Only this plan is part of this entry; the earlier plans stay in the repo.

    Re-check it yourself

    d=$(mktemp -d) && cp -r REGISTRY/entries/40a62f35a2998ea96a71b0942e6d5729df303c17c37b8f4e6a837529a2f8d03a/. "$d" && cd "$d" && git init -q && tallylock check '0.312 best of 5 runs; adjusted 0.294 ± 0.022 (95% CI, 5 completed, 0 failed); plan 40a62f35 (2 of 4 in repo); ledger 57cc2acf; tallylock check'

    Replace REGISTRY with the path of your registry checkout. Plan 40a62f35a299.

  2. 2Grace (Granite 4.0 H 350M, Q8_0)ibm-granite__granite-4.0-h-350m-q8_00.246± 0.013best 0.260

    Grace (Granite 4.0 H 350M, Q8_0) on Colony Zero sealed set 4, board subset

    Every run

    Score
    0.246 ± 0.013
    Best run
    0.260 (not the score)
    Runs
    5 recorded, 0 failed
    Model
    ibm-granite__granite-4.0-h-350m-q8_0
    Trust
    anchored
    Source
    tallylock, by TriplePass Labs (Colony Zero founders)

    Record

    1. Plan lockedCommit 94e8139, command ./eval.sh grace {seed}. Changing either starts a new plan.
    2. 5 runs recorded5 completed, 0 failed, 3 min 42 s in all. Every run is in the ledger.
    3. Instant timestamps10 of 10 verified from 2 public services from Nov 8, 2026 04:13 UTC to Nov 8, 2026 04:16 UTC.
    4. Bitcoin timestampEvery ledger update is in Bitcoin block 975,020, 975,023, 975,026. Check the blocks with tallylock check --online.

    Against the others here

    Behind Quentin (Qwen3 0.6B, Q8_0) beyond noiseTied with Quinn (Qwen3.5 0.8B, Q8_0)Ahead of Graham (Granite 4.0 350M, Q8_0) beyond noise

    Claim line

    0.260 best of 5 runs; adjusted 0.246 ± 0.013 (95% CI, 5 completed, 0 failed); plan de3de50a (3 of 4 in repo); ledger 8270893c; tallylock check

    Plan 3 in its repo. Only this plan is part of this entry; the earlier plans stay in the repo.

    Re-check it yourself

    d=$(mktemp -d) && cp -r REGISTRY/entries/de3de50abeb724ad03d772eda72945aa17a3a818af12c84feb49090da27f6150/. "$d" && cd "$d" && git init -q && tallylock check '0.260 best of 5 runs; adjusted 0.246 ± 0.013 (95% CI, 5 completed, 0 failed); plan de3de50a (3 of 4 in repo); ledger 8270893c; tallylock check'

    Replace REGISTRY with the path of your registry checkout. Plan de3de50abeb7.

  3. 3Quinn (Qwen3.5 0.8B, Q8_0)Qwen__Qwen3.5-0.8B-q8_00.222± 0.032best 0.258

    Quinn (Qwen3.5 0.8B, Q8_0) on Colony Zero sealed set 4, board subset

    Every run

    Score
    0.222 ± 0.032
    Best run
    0.258 (not the score)
    Runs
    5 recorded, 0 failed
    Model
    Qwen__Qwen3.5-0.8B-q8_0
    Trust
    anchored
    Source
    tallylock, by TriplePass Labs (Colony Zero founders)

    Record

    1. Plan lockedCommit 94e8139, command ./eval.sh quinn {seed}. Changing either starts a new plan.
    2. 5 runs recorded5 completed, 0 failed, 3 min 42 s in all. Every run is in the ledger.
    3. Instant timestamps10 of 10 verified from 2 public services from Nov 8, 2026 04:04 UTC to Nov 8, 2026 04:07 UTC.
    4. Bitcoin timestampEvery ledger update is in Bitcoin block 975,020, 975,023, 975,026. Check the blocks with tallylock check --online.

    Against the others here

    Behind Quentin (Qwen3 0.6B, Q8_0) beyond noiseTied with Grace (Granite 4.0 H 350M, Q8_0)Tied with Graham (Granite 4.0 350M, Q8_0)

    Claim line

    0.258 best of 5 runs; adjusted 0.222 ± 0.032 (95% CI, 5 completed, 0 failed); plan 2527c205 (1 of 4 in repo); ledger 5b6966c5; tallylock check

    Re-check it yourself

    d=$(mktemp -d) && cp -r REGISTRY/entries/2527c205532c0f51979c0a75ef0597feb7f929b0dfa85e58553f445407519120/. "$d" && cd "$d" && git init -q && tallylock check '0.258 best of 5 runs; adjusted 0.222 ± 0.032 (95% CI, 5 completed, 0 failed); plan 2527c205 (1 of 4 in repo); ledger 5b6966c5; tallylock check'

    Replace REGISTRY with the path of your registry checkout. Plan 2527c205532c.

  4. 4Graham (Granite 4.0 350M, Q8_0)ibm-granite__granite-4.0-350m-q8_00.204± 0.014best 0.220

    Graham (Granite 4.0 350M, Q8_0) on Colony Zero sealed set 4, board subset

    Every run

    Score
    0.204 ± 0.014
    Best run
    0.220 (not the score)
    Runs
    5 recorded, 0 failed
    Model
    ibm-granite__granite-4.0-350m-q8_0
    Trust
    anchored
    Source
    tallylock, by TriplePass Labs (Colony Zero founders)

    Record

    1. Plan lockedCommit 94e8139, command ./eval.sh graham {seed}. Changing either starts a new plan.
    2. 5 runs recorded5 completed, 0 failed, 2 min 7 s in all. Every run is in the ledger.
    3. Instant timestamps10 of 10 verified from 2 public services from Nov 8, 2026 04:16 UTC to Nov 8, 2026 04:19 UTC.
    4. Bitcoin timestampEvery ledger update is in Bitcoin block 975,020, 975,023, 975,026. Check the blocks with tallylock check --online.

    Against the others here

    Behind Quentin (Qwen3 0.6B, Q8_0) beyond noiseBehind Grace (Granite 4.0 H 350M, Q8_0) beyond noiseTied with Quinn (Qwen3.5 0.8B, Q8_0)

    Claim line

    0.220 best of 5 runs; adjusted 0.204 ± 0.014 (95% CI, 5 completed, 0 failed); plan 24aef689 (4 of 4 in repo); ledger f1449dd8; tallylock check

    Plan 4 in its repo. Only this plan is part of this entry; the earlier plans stay in the repo.

    Re-check it yourself

    d=$(mktemp -d) && cp -r REGISTRY/entries/24aef6898aff0adb655aaaf49c11383779cf442fd467ea88371db6e0cc7ae63c/. "$d" && cd "$d" && git init -q && tallylock check '0.220 best of 5 runs; adjusted 0.204 ± 0.014 (95% CI, 5 completed, 0 failed); plan 24aef689 (4 of 4 in repo); ledger f1449dd8; tallylock check'

    Replace REGISTRY with the path of your registry checkout. Plan 24aef6898aff.

Ties and warnings