Back to the proof
Receipts
Every session, as it ran.
The calculator build, exported from OpenCode’s session API: two main chats and 18 role sessions. Nothing is edited except one redacted email address. Read the moments below first, or any session in full.
DeepSeek V4.1 Flash (max thinking) in OpenCode 2.0.18
Start here
Ten moments, in order.
- 01Runner · spec“First question, the scope one:”read
- 02Runner · plan“Does this plan look right?”read
- 03Runner · card 1.1“Card started. Handing it to the tester now (a subagent, so its work stays separate from mine):”being testedread
- 04Runner · card 1.1“Failing state confirmed exactly as reported … Handing card 1.1 to the implementer:”being builtread
- 05Reviewer · card 1.1“Verdict: PASS — card 1.1 set to awaiting-approval (not done; waiting for your approval)”in reviewread
- 06Reviewer · card 1.1“Live browser check — attempted, not possible”read
- 07Runner · card 1.1“Card 1.1 — Number entry is built and waiting for your approval.”waiting for youread
- 08Runner · card 1.1“Approved and committed: 78565bd — [1.1] Number entry”doneread
- 09Runner · phase 1 retro“L-001 · Hand each role its own status rules”read
- 10New chat · phase 2“Where things stand: Phase 1 (arithmetic engine) is approved and the retro is done. Next is phase 2, card 2.1 Page and keypad”read
Phase 1 · cards 1.1–1.4
One main chat, twelve role sessions.
The main chat also exported these transcripts, so its peak and cost include that. Every role session starts near 9k tokens.
| Session | Calls | Starts at | Peak | Cost | |
|---|---|---|---|---|---|
| — Runner (main chat) | 120 | 9,157 | 280,280 | $0.1856 | transcript |
| 1.1 Tester | 14 | 9,257 | 32,126 | $0.0142 | transcript |
| 1.1 Implementer | 10 | 9,264 | 25,760 | $0.0099 | transcript |
| 1.1 Reviewer | 14 | 8,851 | 74,174 | $0.0352 | transcript |
| 1.2 Tester | 18 | 9,222 | 36,675 | $0.0147 | transcript |
| 1.2 Implementer | 14 | 9,201 | 34,943 | $0.0142 | transcript |
| 1.2 Reviewer | 17 | 8,789 | 38,892 | $0.0160 | transcript |
| 1.3 Tester | 18 | 9,211 | 34,378 | $0.0126 | transcript |
| 1.3 Implementer | 19 | 9,232 | 25,389 | $0.0095 | transcript |
| 1.3 Reviewer | 19 | 8,826 | 41,030 | $0.0174 | transcript |
| 1.4 Tester | 17 | 9,248 | 35,369 | $0.0129 | transcript |
| 1.4 Implementer | 19 | 9,317 | 43,162 | $0.0197 | transcript |
| 1.4 Reviewer | 17 | 8,851 | 56,751 | $0.0273 | transcript |
Phase 2 · cards 2.1–2.2
A new main chat, six role sessions.
The new chat picked up from the handoff file at 9.2k tokens. It also deployed the app to GitHub Pages.
| Session | Calls | Starts at | Peak | Cost | |
|---|---|---|---|---|---|
| — Runner (main chat) | 61 | 9,192 | 120,058 | $0.0663 | transcript |
| 2.1 Tester | 39 | 9,330 | 134,817 | $0.0751 | transcript |
| 2.1 Implementer | 17 | 9,652 | 61,198 | $0.0251 | transcript |
| 2.1 Reviewer | 21 | 9,001 | 67,445 | $0.0277 | transcript |
| 2.2 Tester | 22 | 9,469 | 85,044 | $0.0409 | transcript |
| 2.2 Implementer | 17 | 9,497 | 60,093 | $0.0184 | transcript |
| 2.2 Reviewer | 25 | 8,982 | 67,482 | $0.0250 | transcript |
Numbers from checks/session-costs.mjs on the exported sessions. Absolute Windows paths appear as they were in the sessions.