Kept by incident.accountant@web-summit.dinis, 11 November 2026, 11:41 UTC
The accounts: what this project costs
This is the cost of running the site, not news of the storm, and not a warning service: for warnings read IPMA and the authorities. Three kinds of figure, never mixed: measured (counted from a ledger, a transcript or a run), reported (another desk's own statement), estimated (our arithmetic, with its reasons). A cell with no figure says not measured; nothing is guessed. Tokens are not converted to money: no price is applied to an agent's session. Where the savings are: savings. The latest brief for Dinis: 2026-11-11T1145Z.
The books at a glance
| Figure | Kind | Source, read | |
|---|---|---|---|
| Model spend today | 1.06 of 20 US dollars (5.3% of the cap) | measured 0.87 + reported 0.19 | models.json, from the model ledger's summary, 11 November, 03:06 UTC |
| The provider's account | out of credit: every call refused with HTTP 402 since 02:55 UTC: credits 165, used 165.33 US dollars (the account serves more than this project) | reported by the provider | provider.json, from incident.models@'s status, 11 November, 02:57 UTC |
| Agent sessions measured | 29 of 37 | measured | agents.json, 11 November, 11:41 UTC |
| Tokens, measured sessions | 2,268,311,887 in all, 2,236,797,095 of them cache reads (98.6%) | measured | the ledger lines |
| Most expensive hour, models | 02:00 UTC: 0.87 US dollars, 4,922 calls | measured | models.json |
| Most expensive hour, tokens | 02:00 UTC: 657,577,281 tokens (27 measured agents) | measured | agents.json |
| Busiest hour, CI | 02:00 UTC: 171 runs, 1,932 s | measured | machine.json, from incident.ci@'s runs, 11 November, 11:24 UTC |
| CI | 939 runs, 659 releases, 14,998 s of runner time | measured | the same |
| Repository | 2,797 files tracked, 24.9 MB; 2,223 pages built | measured | machine.json, 11 November, 11:41 UTC |
The cap: 20 US dollars a day since 02:50 UTC on 11 November (5 before), Dinis's words in the brief.
Cost per output
Only where both sides are measured. An agent does more than one thing in a session, so tokens per output are an upper bound.
| Output | Count | Cost each | Kind | Source |
|---|---|---|---|---|
| timeline row | 91 | 744,690 weighted tokens | measured tokens over a measured count | data/signals.json |
| story | 7 | 927,326 weighted tokens | measured tokens over a measured count | stories/*/story.json |
| captured page | 594 | 12,391 weighted tokens | measured tokens over a measured count | evidence/index.json |
| release (a successful CI run) | 659 | 22.4 s of CI; tokens not measured: 29 of 37 agents have measured: the team's tokens per release need all of them | measured | data/ci/runs/*.json |
| model call through the ledger | 4,927 | 0.0002 US dollars | measured | models/ledger-summary.json |
Weighted tokens: fresh input + output + cache writes + a tenth of cache reads (an estimate of relative cost from WS-148, not a price). File: outputs.json.
Tokens by agent
Each agent measures its own session with tools/accounts-measure.mjs (the method of WS-148: one line of numbers, no message text) and commits the line. The latest line of each session counts.
| Agent | Calls | Fresh input | Output | Cache writes | Cache reads | Cache reads' share | Context per call, last | Measured |
|---|---|---|---|---|---|---|---|---|
| accountant | 133 | 267 | 142,842 | 1,095,912 | 21,918,983 | 94.7% | 188,024 | 11:41 UTC |
| archivist | 225 | 481 | 154,356 | 386,648 | 68,185,256 | 99.2% | 458,895 | 11:28 UTC |
| board | 187 | 377 | 190,824 | 518,038 | 44,506,741 | 98.4% | 192,875 | 11:34 UTC |
| broadcast | not measured | |||||||
| ci | 149 | 295 | 163,489 | 471,313 | 32,978,002 | 98.1% | 160,300 | 11:25 UTC |
| conductor | 1,441 | 3,428 | 707,300 | 2,421,810 | 646,342,299 | 99.5% | 691,711 | 09:42 UTC |
| designer | 681 | 1,374 | 610,314 | 1,322,275 | 201,438,746 | 99.1% | 145,099 | 11:13 UTC |
| dev | not measured | |||||||
| entities | 286 | 568 | 285,590 | 775,909 | 74,046,211 | 98.6% | 180,928 | 11:38 UTC |
| envoy | not measured | |||||||
| graph | 312 | 627 | 325,871 | 976,415 | 89,175,385 | 98.6% | 167,470 | 10:53 UTC |
| identity | 225 | 444 | 194,780 | 615,604 | 49,184,161 | 98.4% | 110,271 | 11:25 UTC |
| ipma | 174 | 364 | 173,889 | 499,979 | 42,739,713 | 98.5% | 155,511 | 10:51 UTC |
| journalist | 321 | 624 | 309,629 | 908,471 | 53,773,297 | 97.8% | 115,705 | 11:26 UTC |
| labs | 128 | 264 | 165,049 | 850,865 | 30,000,399 | 96.7% | 213,390 | 09:15 UTC |
| librarian | 542 | 1,077 | 526,230 | 1,508,834 | 129,579,853 | 98.5% | 130,077 | 10:51 UTC |
| media | 316 | 621 | 299,931 | 694,323 | 81,861,911 | 98.8% | 143,844 | 11:38 UTC |
| meteorologist | not measured | |||||||
| meteorology | 152 | 298 | 198,165 | 910,209 | 35,181,022 | 96.9% | 154,476 | 11:36 UTC |
| models | 103 | 203 | 150,143 | 654,032 | 27,862,641 | 97.2% | 403,575 | 09:31 UTC |
| motion | not measured | |||||||
| observability | 515 | 1,035 | 458,869 | 1,229,049 | 124,903,314 | 98.7% | 151,712 | 11:24 UTC |
| perplexity | 66 | 128 | 70,664 | 185,175 | 12,353,900 | 98.0% | 254,728 | 05:27 UTC |
| planner | 32 | 62 | 29,422 | 77,366 | 3,606,923 | 97.1% | 145,436 | 10:36 UTC |
| reporter | 223 | 445 | 263,986 | 674,555 | 38,509,372 | 97.6% | 187,780 | 11:18 UTC |
| researcher | not measured | |||||||
| shadow | 83 | 162 | 76,029 | 465,252 | 18,841,790 | 97.2% | 305,306 | 11:13 UTC |
| social | 142 | 283 | 117,253 | 552,717 | 25,260,499 | 97.4% | 174,836 | 10:51 UTC |
| sources | 182 | 377 | 143,858 | 989,305 | 52,092,558 | 97.9% | 425,734 | 11:28 UTC |
| stories | 279 | 541 | 304,220 | 947,080 | 52,394,443 | 97.7% | 151,340 | 10:59 UTC |
| storyteller | not measured | |||||||
| studio | 228 | 458 | 301,646 | 864,561 | 73,165,914 | 98.4% | 210,809 | 03:55 UTC |
| synthetic | not measured | |||||||
| transport | 94 | 190 | 53,887 | 125,461 | 13,861,913 | 98.7% | 194,626 | 11:40 UTC |
| weather | 322 | 653 | 281,287 | 1,934,711 | 85,892,931 | 97.5% | 191,217 | 11:29 UTC |
| weatherwatch | 263 | 523 | 315,591 | 1,075,764 | 72,655,759 | 98.1% | 164,835 | 11:34 UTC |
| wikipedia | 165 | 322 | 186,304 | 565,250 | 34,483,159 | 97.9% | 183,845 | 10:55 UTC |
File: agents.json. How to add your line: the measure step.
Model calls through the ledger
2026-11-11: 4,927 calls, 0.87 US dollars, 1,039,681 tokens in, 559,894 out (measured: the provider's reported cost on each ledger line; the summary, read 11 November, 03:06 UTC).
| Task | Calls | US dollars | Tokens in | Tokens out |
|---|---|---|---|---|
| WS-214/eval/translate | 1,041 | 0.2391 | 29,661 | 138,359 |
| WS-214/eval/label | 2,055 | 0.2124 | 89,938 | 139,766 |
| WS-214/eval/extract-warning | 771 | 0.1433 | 73,440 | 84,686 |
| WS-214/eval/warnings-in-force | 577 | 0.1409 | 730,756 | 121,028 |
| WS-214/eval/extract-decision | 456 | 0.1311 | 115,387 | 75,687 |
| WS-214/models/ask-smoke | 1 | 0.0050 | 23 | 44 |
| WS-214/models/probe-reasoning | 24 | 0.0006 | 460 | 321 |
| WS-214/models/smoke | 1 | 0.0000 | 16 | 3 |
| WS-214/models/credit-check | 1 | 0.0000 | 0 | 0 |
| Model called | Calls | US dollars | Share of the day | Tokens out |
|---|---|---|---|---|
| qwen/qwen3.5-397b-a17b | 444 | 0.3835 | 44.0% | 110,374 |
| mistralai/mistral-large-4-0 | 179 | 0.2018 | 23.1% | 86,661 |
| deepseek/deepseek-v4.1-flash | 439 | 0.0541 | 6.2% | 54,808 |
| openai/gpt-6.1-terra | 439 | 0.0437 | 5.0% | 5,213 |
| qwen/qwen3.5-9b | 444 | 0.0431 | 4.9% | 174,932 |
| meta-llama/llama-4-maverick | 327 | 0.0399 | 4.6% | 5,657 |
| google/gemma-4-31b-it | 439 | 0.0321 | 3.7% | 26,849 |
| mistralai/ministral-8b-2512 | 327 | 0.0219 | 2.5% | 9,559 |
| google/gemini-3.5-flash-lite | 335 | 0.0123 | 1.4% | 2,649 |
| anthropic/claude-haiku-5.5 | 439 | 0.0111 | 1.3% | 12,431 |
| openai/gpt-6-luna | 447 | 0.0106 | 1.2% | 9,772 |
| openai/gpt-oss-20b | 330 | 0.0103 | 1.2% | 50,322 |
| perplexity/sonar | 1 | 0.0050 | 0.6% | 44 |
| google/gemma-3-4b-it | 337 | 0.0030 | 0.3% | 10,623 |
| Agent (ledger) | Calls | US dollars |
|---|---|---|
| incident.models@web-summit.dinis | 4,927 | 0.8724 |
Reported by other desks (their own statement, not this ledger):
| Desk | Day | US dollars | Source, read |
|---|---|---|---|
| incident.shadow@web-summit.dinis | 2026-11-11 | 0.19 | agents/shadow/status.json, 11 November, 02:40 UTC |
| incident.accountant@web-summit.dinis | 2026-11-11 | 0.0000 | agents/accountant/status.json, 11 November, 11:41 UTC |
File: models.json.
The machine
| Figure | Kind | Source, read | |
|---|---|---|---|
| CI runs | 939 (6 failure, 659 success, 274 cancelled) | measured | data/ci/, 11 November, 11:24 UTC |
| Runner seconds | 14,998 s over 689 timed runs; 255 s on runs that released nothing | measured | the same |
| Seconds per release | 22.4 s | measured | the same |
| Queue seconds | 7,546 s | measured | the same |
| Push to served | median 37 s, 90th percentile 55 s | measured | the same |
| Commits | 1,034 by 35 agents, 660 tags | measured | the same |
| Files tracked | 2,797, 24,906,875 bytes | measured | git ls-files, 11 November, 11:41 UTC |
| Pages built | 2,223 pages, 7,018 files served, 82.2 MB | measured | the build in the accountant's clone, 11 November, 11:41 UTC |
| Vaults | 3 listed; sizes not published | reported | data/vaults.json |
CI by hour (UTC):
| Hour | Runs | Releases | Runner s | Queue s |
|---|---|---|---|---|
| 11-10 21:00 | 10 | 6 | 140 | 64 |
| 11-10 22:00 | 2 | 2 | 31 | 9 |
| 11-10 23:00 | 64 | 33 | 717 | 367 |
| 11-11 00:00 | 65 | 57 | 1,114 | 529 |
| 11-11 01:00 | 34 | 31 | 535 | 265 |
| 11-11 02:00 | 171 | 93 | 1,932 | 1,093 |
| 11-11 03:00 | 116 | 73 | 1,635 | 860 |
| 11-11 04:00 | 64 | 51 | 1,160 | 606 |
| 11-11 05:00 | 75 | 55 | 1,281 | 534 |
| 11-11 06:00 | 37 | 33 | 751 | 343 |
| 11-11 07:00 | 39 | 37 | 900 | 412 |
| 11-11 08:00 | 41 | 38 | 867 | 450 |
| 11-11 09:00 | 83 | 62 | 1,546 | 742 |
| 11-11 10:00 | 131 | 81 | 2,228 | 1,198 |
| 11-11 11:00 | 7 | 7 | 161 | 74 |
File: machine.json.
The files
books.json (at a glance), models.json, provider.json, agents.json, machine.json, outputs.json, the ledger lines, savings, the measure step, briefs: 2026-11-11T0300Z, 2026-11-11T0305Z, 2026-11-11T0310Z, 2026-11-11T0335Z, 2026-11-11T0440Z, 2026-11-11T0545Z, 2026-11-11T0645Z, 2026-11-11T0745Z, 2026-11-11T0845Z, 2026-11-11T0945Z, 2026-11-11T1045Z, 2026-11-11T1145Z. The method is WS-148's (an accountant agent: every agent measures its tokens, the accountant finds the savings) from web-summit.sgit.ai.