# Cost log (English experiments)

One line per script run. Token prices are approximate list prices (see PRICE in common.py); ASR is priced per audio hour.

| date | script | calls | prompt tok | completion tok | est. $ | note |
|---|---|---|---|---|---|---|
| 2026-10-14 | 2_8_skills/make_tasks.py draft | 8 | 10217 | 8554 | 0.0397 | task drafting with gemini-3.8-flash |
| 2026-10-14 | 2_6_overgeneralize/pilot_homonym.py amazon shell python target | 144 | 356452 | 83982 | 0.3150 |  |
| 2026-10-14 | 2_6_overgeneralize/pilot_homonym.py amazon_b amazon_c shell_terminal | 120 | 302993 | 51070 | 0.2092 |  |
| 2026-10-14 | 2_8_skills/make_tasks.py draft | 127 | 385407 | 260767 | 1.2669 | task drafting with gemini-3.8-flash |
| 2026-10-14 | corpus/transcribe.py grok grok_terms compat | 0 | 0 | 0 | 0.0883 | ASR only, 9.5 audio min per variant |
| 2026-10-14 | ad-hoc probes (gateway 400 bisect, ~25 tiny gemini calls; 2 ASR calls on W01) | ~27 | ~1500 | ~8000 | 0.0250 | logged by hand, upper-bound estimate |
| 2026-10-14 | corpus redo W05/W14/W20 ASR (Niamh->Xochitl) | 9 calls | - | ~$0.01 |
| 2026-10-14 | 2_1_stateless/run.py gemini-3.8-flash | 9 | 6207 | 2403 | 0.0103 |  |
| 2026-10-14 | 2_1_stateless/run.py deepseek-v4.1-flash | 9 | 18852 | 1263 | 0.0045 |  |
| 2026-10-14 | 2_1_stateless probe deepseek turn-1 usage x3 | 3 | 30167 | 2138 | 0.0072 |  |
| 2026-10-14 | 2_1_stateless probe gateway overhead (hi, zh turn1, en short) | 8 | 10297 | 2285 | 0.0073 |  |
| 2026-10-14 | 2_9_isolation/run.py --variant grok --runs 1 | 58 | 179251 | 43132 | 0.1537 |  |
| 2026-10-14 | 2_4_skill/run.py | 112 | 596330 | 57755 | 0.2883 | 28 W+L memos x 2 conditions x 2 models |
| 2026-10-14 | 2_8_skills/make_tasks.py baseline | 779 | 448279 | 660830 | 4.3857 | no-skill baseline on all candidates, 3 models + cross-model judge (judge cost included) |
| 2026-10-14 | 2_5_perspectives/run.py | 80 | 518385 | 36627 | 0.2019 | skill_only vs with_perspectives, 2 writers x 2 conds x 10, cross-model judge |
| 2026-10-14 | 2_1_workspace/run_chat.py gemini-3.8-flash A1 | 1 | 59 | 1925 | 0.0058 |  |
| 2026-10-14 | 2_1_workspace/run_chat.py gemini-3.8-flash A2 | 1 | 59 | 2312 | 0.0070 |  |
| 2026-10-14 | 2_1_workspace/run_chat.py gemini-3.8-flash A3 | 1 | 59 | 1781 | 0.0054 |  |
| 2026-10-14 | 2_8_skills ad-hoc gateway probes (tool calling, gemini 400 'now' trigger bisect, ~45 tiny calls) | ~45 | ~3000 | ~20000 | 0.0500 | logged by hand, upper bound |
| 2026-10-14 | 2_8_skills/run.py grok-4-fast none full_1 index index_rep2 | 137 | 554735 | 18647 | 0.7400 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_5_perspectives/run_hedge.py | 160 | 663588 | 114954 | 0.4410 | hedge pilot I01-I10: default x2, perspective (v2), attack_v1; cross-model judge |
| 2026-10-14 | 2_6_overgeneralize/run.py | 86 | 417061 | 42592 | 0.2042 | steps 1-3, 5 runs per distillation, 11 memos x 3 rules downstream |
| 2026-10-14 | 2_9_isolation/run.py --variant grok --runs 3 | 116 | 358612 | 73564 | 0.2692 |  |
| 2026-10-14 | 2_8_skills/run.py deepseek-v4.1-flash none full_1 index index_rep2 | 170 | 2146440 | 102434 | 0.9934 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_5_perspectives/run_hedge.py attack_rep | 40 | 81472 | 10591 | 0.0472 | second run of attack_v1 wording, I01-I10 |
| 2026-10-14 | 2_8_skills/run.py gemini-3.8-flash none full_1 index index_rep2 | 182 | 1034232 | 115354 | 1.2083 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py deepseek-v4.1-flash full_all | 1 | 336477 | 323 | 0.0507 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py gemini-3.8-flash full_all | 1 | 349905 | 637 | 0.2648 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py grok-4-fast full_all | 1 | 318141 | 105 | 0.3979 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_5_perspectives/run.py rep | 80 | 597949 | 42363 | 0.2359 | skill_only vs with_perspectives, 2 writers x 2 conds x 10, cross-model judge |
| 2026-10-14 | 2_8_skills/run.py deepseek-v4.1-flash full_all | 29 | 10814898 | 15468 | 1.6615 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py gemini-3.8-flash full_all | 29 | 10147296 | 18385 | 7.6794 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py grok-4-fast full_all | 29 | 9226077 | 5720 | 11.5469 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_3_names/run.py deepseek-v4.1-flash gemini-3.8-flash | 152 | 304271 | 123116 | 0.4217 |  |
| 2026-10-14 | 2_8_skills/grade.py | 363 | 0 | 0 | 0.7899 | judge calls; token counts not tracked here, $ from gateway trace |
| 2026-10-14 | 2_8_skills/grade.py | 87 | 0 | 0 | 0.1638 | judge calls; token counts not tracked here, $ from gateway trace |
| 2026-10-14 | 2_1_workspace counterfactual_url.py + run_agent.py B1-B3 (ASR only) | 0 | 0 | 0 | 0.0320 | compat ASR 3x82 s (0.36/h) + agent grok ASR <=9x82 s (0.10/h), upper bound; agy model tokens are on the Antigravity subscription, not AIB |
| 2026-10-14 | 2_8_skills/fix_descriptions.py rewrite | 21 | 60037 | 24005 | 0.1350 | writer gemini-3.8-flash |
| 2026-10-14 | 2_8_skills/run.py grok-4-fast index_fixed | 41 | 257758 | 5783 | 0.3367 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py deepseek-v4.1-flash index_fixed | 28 | 476409 | 19143 | 0.2329 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/run.py gemini-3.8-flash index_fixed | 20 | 146213 | 7342 | 0.1372 | gateway estimated_cost_usd per call (trace) |
| 2026-10-14 | 2_8_skills/grade.py | 56 | 0 | 0 | 0.0967 | judge calls; token counts not tracked here, $ from gateway trace |
| 2026-10-14 | 2_9_isolation/run.py --variant grok --runs 3 | 84 | 20046 | 71188 | 0.2236 | --conds batch1 (control) |
| 2026-10-14 | 2_9_isolation/run.py --conds batch1 (first attempt, crashed on a gemini JSON parse before logging) | ~100 | - | - | 0.0500 | logged by hand, upper-bound estimate |
| 2026-10-14 | 2_8_skills/grade.py secondary | 506 | 0 | 0 | 2.2795 | judge calls; token counts not tracked here, $ from gateway trace |