{
 "baseline": {
  "blocks": [
   "0",
   "1",
   "2",
   "3"
  ],
  "commitment": "a36b4dfd92d41d82089f4b4035a7e9beac5562b0de5e2286420c085d8e57b840",
  "head_pass_on_kept": 0.6429236499068901,
  "kept_tasks": 179,
  "note": "baseline.blocks[\"0\"] = selection run (k=4 on all 1000), blocks 1-3 = p_i runs on the kept tasks; merged_per_task = blocks 1-3",
  "plan_seeds": [
   "c2b70911731c868224b1f368b73339c01380dfbb7fc0f178f314c2d200e5b64c",
   "aa271a630e74f4375bb21c51d7429403d3bd5853066ae7d05efca5ed6c631c94",
   "cc9b492a7efedf250e40a17a426a0cbf6c199adc560d1e8a84b7fce95c9f7775",
   "26eacb7850ba888f9c1b08b27bd2882dd6ad50209f09a41df8bc7d1f15b93c23"
  ],
  "rollouts_per_task": 12,
  "seed": "57bd382fb6ab2d4abbba55c6a72311d64da6d8b2c1122310c36d9cbfc7233779",
  "set_size": 1000,
  "settings": {
   "harness": {
    "enable_thinking": true,
    "max_context_tokens": 12288,
    "max_output_tokens": 8192,
    "policy": "autoregressive",
    "temperature": 1.0,
    "top_k": 20,
    "top_p": 0.95,
    "version": "text-tools-mimo-complete-group-v1"
   },
   "request_seed_rule": "seed(task t, sample j) = int(sha256(P + \":\" + env + \":\" + t + \":\" + j)[:12], 16); P = plan seed of the run",
   "runtime": [
    "16f69a4b06f427c3",
    "770812395998dd40",
    "511142f1d0ebe8c5",
    "1fbc4f47e06e1e58",
    "vllm serve \"$CK\" --host 127.0.0.1 --port \"$PORT\" --served-model-name probe \\",
    "  --dtype bfloat16 --max-model-len 12288 --gpu-memory-utilization \"$GMU\" --enable-prefix-caching \\",
    "  --limit-mm-per-prompt '{\"image\":0,\"video\":0}' --max-num-seqs ${MAX_SEQS:-512} \"$@\"",
    "      8 NVIDIA B200"
   ],
   "sampling_contract": {
    "calibration": {
     "backend_profile_sha256": "38778ba9ee364d5c199a06be470b25484957f32c4ba55765e6c095ac881915a0",
     "cdf_abs_error": 0.0,
     "checkpoint": "d1987e6a9dc40082c8dc8c40e6027ffd3e751247a682a8c7dc3868b46a7eda6a",
     "harness_sha256": "9245ecb73e860e09fc3b6d0579ff5d03da51f370919fe5050f1bb350740de059",
     "logprob_atol": 0.0,
     "model_runtime_revision": "qwen3_5-mimo-v1",
     "report_sha256": "e57c0f986cc56da6779d8e4c8a14266b0aae5964544dde2d2e3118ab35686b58",
     "toploc_exp_mismatches": 0,
     "toploc_mant_err_mean": 0.0,
     "toploc_mant_err_median": 0.0,
     "version": "cached-prefill-calibration-v1"
    },
    "generation": "cached-eager-inverse-cdf",
    "randomness": "f8d9097c24091f6aeeb10df51f653385a33de5c08ced0e91b5cd26c58bc69cfd",
    "support_adjudication": "exact-cached-replay-v1",
    "verification": "prefill-cdf-calibrated",
    "version": "forced-inverse-cdf-complete-group-mimo-v1"
   }
  },
  "settings_hash": "a8dcaf15c54271d67e3622728666b98ccaf95d13ef57a32c5ef8089bf5cd5566",
  "url": "/gate/heldout/r20261031T123358Z-ca64c1f1.json"
 },
 "decision": {
  "bar_z": 1.0,
  "c_cap_rate": 0.10049999999999999,
  "c_cap_rate_kept": 0.2011173184357542,
  "c_pass": 0.7276536312849162,
  "checks": {
   "commitment_matches": true,
   "decision_matches": true,
   "decision_recomputed": "green",
   "indices_match_seed": true,
   "z_published": 3.8804607275736713,
   "z_recomputed": 3.8804607275736713
  },
  "code": "/gate/heldout/heldout.py score_ivw",
  "decided_at_cr": "2026-10-31 07:39:26 CR (UTC-6)",
  "decision": "green",
  "gain": 0.08360008536064877,
  "gains": 27,
  "h_cap_rate_kept": 0.308659217877095,
  "h_cap_rate_set": 0.11975,
  "h_pass": 0.6429236499068903,
  "kept": 179,
  "losses": 10,
  "math": "w_t = 1/(q(1-q)), q = (c_sel+1)/(n_sel+2); d_t = candidate rate - head rate; gain = sum w d / sum w; se_task = sqrt(sum w^2 d^2) / sum w; se = sqrt(se_task^2 + sigma_run^2 (1 + 1/n_p_blocks)); z = gain / se; GREEN iff gain > 0, z > bar_z, estimated whole-set cap-hit < 0.20, >= 100 kept",
  "n": 179,
  "n_p_blocks": 3,
  "reasons": [],
  "recomputed": {
   "decision": "green",
   "matches": true,
   "z": 3.8804607275736713
  },
  "rule": "heldout.score_ivw(selection block 0, merged p_i blocks, candidate cells, kept, batch_tasks, bar_z, z_inclusive, cap_max, min_kept, set_size); https://affine.io/gate/heldout/heldout.py",
  "se": 0.021543855544421717,
  "se_task": 0.021339161841684072,
  "sigma_run": 0.0025658161878812694,
  "weighting": "1/(q(1-q)), q=(c_sel+1)/(n_sel+2)",
  "z": 3.8804607275736713,
  "z_inclusive": false
 },
 "epoch": 354,
 "head_after": "ce3971dc21a17a964c03273b68a969984dd43503094937e56bee8cf1280b1853",
 "head_before": "87fa740811e75e4c221258ce6cf2d8cb1db1519d69aa0a65b42cb5fddea5a499",
 "miner": {
  "file": {
   "bytes": 381381,
   "note": "The exact file copied from the miner object at the turn (before the batch-cap trim).",
   "published": true,
   "sha256": "e1def164a7048cf25bf89d9fb0459e45c49cec1ae546b8d2125c1239e22098f8",
   "url": "/k1/turns/354/batch.json"
  },
  "format": "affine-batch-v1",
  "hotkey": "5C4jMYe9YuupMqKdecwJ4mhwSSL9WpoL46y88EBZwppD235z",
  "train": {
   "kl": 0.02,
   "lr": 5e-07,
   "source": "trainer report (turn before per-turn train params: fixed lr / kl)"
  },
  "train_given": null,
  "train_params": null,
  "uid": 260
 },
 "probe": {
  "candidate": "ce3971dc21a17a964c03273b68a969984dd43503094937e56bee8cf1280b1853",
  "candidate_k": 4,
  "candidate_plan_seed": "aa271a630e74f4375bb21c51d7429403d3bd5853066ae7d05efca5ed6c631c94",
  "cells": "head_per_task (merged blocks 1-3) and candidate_per_task, per task {n, c, w, u, b, x, tok, L}",
  "head": "87fa740811e75e4c221258ce6cf2d8cb1db1519d69aa0a65b42cb5fddea5a499",
  "tasks_scored": 179,
  "url": "/gate/reveal/354.json"
 },
 "published_at": 1793473342.6,
 "round_id": "r20261031T123358Z-ca64c1f1",
 "rule": "Published for GREEN turns only, after the held-out round is closed. RED turns and open rounds publish no baseline, probe results, tasks or files.",
 "schema": "sn120-k1-transparency-v1-green",
 "tasks": {
  "count": 1000,
  "kept": 179,
  "sha256": "399360872f7952389b8f06a57427ff3913bcb0046dd5bd1299fc9c5a21e3c51b",
  "url": "/k1/turns/354/tasks.json"
 },
 "turn": "/k1/turns/354/turn.json"
}