predictor · Domain-agnostic trend-detection pipelinepredictor.explorer · Cytoscape-based living map of a field — provenance on every edgepredictor.ingest-pipeline · Daily curated ingestion + LLM extraction behind quality gates, batch-AP…predictor.daily-timers · Three domains read themselves every morning — film 06predictor.two-lens-model · Landscape vs Moverspredictor.quality-gates · Zero-token CPU checks on every LLM outputpredictor.shadow-mode · A cheaper understudy model silently re-runs primary-model jobs; disagre…predictor.calibration-reports · Rolling 7-day gate pass-rate + quality-drift reporting; thresholds tune…predictor.epoch-boundary · Scores never compare across a restartpredictor.domain-agnostic-boundary · Framework/domain separation enforced by a grep audit test — domains are…predictor.domain-template · New domain onboarding from scaffoldpredictor.chatter-source-policy · Ingest-only mode for short-form sources — count mentions, skip extracti…predictor.dev-prod-split · Two clones, one data planepredictor.playwright-harness · Self-contained web testspredictor.ontology-page · Specialist-facing taxonomy reference generated from domain profilesagent-platform · Multi-agent platformagent-platform.director · Director agent — coordinates the crew, human approval in the loop, per-…agent-platform.director.telegram · Telegram-native command surfaceagent-platform.director.morning-brief · An ambient 07agent-platform.sysadmin · Reads the server's own health every morning at 06agent-platform.scout · Wanders the box unprompted at 05agent-platform.scout.session-ingest · Twelve thousand turns of agent session log, stripped to query/progress/…agent-platform.scout.leads-ledger · An append-only story queue the Scout may add to but never edit — 324 le…agent-platform.writer · Drafts field notes from a claimed lead, in a voice loaded from bottled…agent-platform.wire-editor · Triage desk for inbound wire items, with a chief's shadow pass scoring…agent-platform.ask-assistant · One assistant per solution page, each given the same facts that built t…agent-platform.blog-pipeline · Content + marketer agents drafting and shipping posts to blog.uzelhub.c…agent-platform.decision-trace · Every agent writes structured what-ran/what-it-decided/what-it-cost row…agent-platform.health-monitor · The platform checks its own pulse hourly and pages a human when it stopsagent-platform.ops-calendar · Ops calendar as code — .ics in the repo carrying drills, kickoffs, pric…agent-platform.ops-desk · Operator surfaces behind one capability URLagent-platform.n8n-stack · n8n as the automation substrate — workflows, webhook surface via ngrok,…ops.backup-system · Layered backupsops.notify-telegram · One shared Telegram channel for all agents — [AGENT]-prefixed, outbound…ops.safe-reboot · Reboot as orchestrationops.truth-map-practice · _host/README.md as cross-project source of truth with dated corrections…ops.security-architecture · The box's exposure written down and then audited against itself on a da…ops.docker-hygiene · Disk-bounded Dockerweb.studio · Editorial site — 'a manifesto disguised as a portfolio'; literary regis…web.marketing-generator · Decision-free page productionweb.field-notes · True stories from operating the systems, drafted from the lead queue, r…web.apex-sandbox · A throwaway rebuild of the whole apex, every page in reading order with…web.agents-page · The crew introduced in plain language, each with what it does, a receip…web.sister-site-doctrine · Two sites, same truths, different attention budgets — parity without ap…web.translation-patterns · Editorial→marketing translation as a studied craftweb.coverage-tripwire · Dropped-ideas tripwireweb.promotion-inventory · The box documents itselfweb.register-gradient · Five-step register gradientweb.voice-ledger · The site's voice recorded as dictated rather than inferred — one entry…web.publishing-drip · One or two URLs a week, real dates never backdated, one promotion momen…web.cycle-canvas · Hand-tuned canvas animation on the editorial home (orbit glow, delivere…method.4cs · Converse, Constrain, Construct, Curate — the working method behind ever…method.human-ai-collaboration · 100% human-AI collaboration, zero manual coding/tests, zero hardcoded d…method.dated-amendments · Docs never silently rewritten — corrections carry dates and keep the wr…

Agents/Analytics deep dive

Analytics deep dive

Every agent run on this server, and what it costs

Every agent writes down a row each time it runs: the model it used, how much it read, how much it wrote, and what that cost. This page counts four months of those rows.

1,408 Runs logged every agent run, one row each
12 Kinds of step agents and their named sub-steps
5 Models used across four model families
$66.93 Total cost everything, 2026-04-25 to 2026-08-09
4.27M Tokens read text paid at full price, not the cache
1.90M Tokens written model output

The same agent, on two different models

The Scout does two jobs in one run. First it reads widely and cheaply, looking for anything worth a second look. Then it looks for a connection between two things nobody told it to compare. Those are different jobs, so they run on different models. The cheap model reads about seventy times more text for about two per cent of the money. Paying for the expensive model only where it is needed is the whole design, and the table below is the proof.

Scout · reading, then connecting
RunsInput readCost
The cheap read · Haikuclaude-haiku-4-5-20251001871.55M$0.61
The connection · Fable 5claude-fable-52821K$31.82
Haiku Fable 5

What each kind of model costs

Sorted by price, almost none of the money goes to the model that does most of the reading. Look at the first row: nearly three hundred steps used no model at all. Those are ordinary code, with no AI in them, and they are counted here on purpose. Leaving them out would suggest that everything on this server is a language model. That is not true, and it would be the more flattering story.

Every model, across the whole server
RunsInput readCostShare of spend
No model · ordinary code2940$00%
Haiku · the cheap read3641.95M$0.771%
Sonnet · the everyday model7122.22M$33.6550%
Opus · the heavy read1089K$0.691%
Fable 5 · the connecting step2821K$31.8248%
No model Haiku Sonnet Opus Fable 5

One run costs anywhere from a fraction of a cent to a dollar

The cheapest step costs about a hundred and forty times less than the most expensive one. That is not waste — the jobs are not the same size. Packaging a blog post is small and repetitive, and costs about what you would expect. Looking for connections is the most expensive thing that runs here. It has to hold a dozen threads at once and say something new, and it does that twenty-eight times.

Cost of a single run, step by step
RunsAverageMost expensive run
scout_synthesisclaude-fable-528$1.14$1.80
sysadminclaude-sonnet-516$0.61$0.99
wire_triageclaude-sonnet-54$0.44$0.58
chief_shadowclaude-sonnet-54$0.23$0.38
writer_draftclaude-sonnet-512$0.11$0.20
directorclaude-sonnet-5140$0.08$0.51
content_agentclaude-sonnet-4-6276$0.02$0.03
marketer_agent.packageclaude-sonnet-4-6270$0.0082$0.0099
scout_walkclaude-haiku-4-5-2025100187$0.0070$0.01
ask.explorerclaude-haiku-4-5-202510011$0.0015$0.0015
marketer_agent.extractclaude-haiku-4-5-20251001276$0.0006$0.0007
Haiku Sonnet Fable 5

Most of what the models read, they had read before

The agents read the same long set of instructions at the start of every run. That text is paid for once, then read again at a fraction of the price — the rows below call that the cache. It is the reason the bill runs to two figures instead of four. It also explains something odd in the totals at the top of this page. The reading figure there counts only text paid at full price. The true amount read is much higher.

4.3× more read from the cache than read at full price

Where the reading came from
Tokens
Read from cache18.34M
Written to cache4.85M
Read at full price4.27M
Haiku Sonnet Fable 5

Busy months and expensive months are not the same months

May was by far the busiest month and one of the cheapest. The blog agents were running in bulk, making lots of small, cheap calls. July and August are quieter and cost far more, because the work changed. There were fewer runs, and each one asked a better model to think harder. So how busy a month was tells you very little about what it cost. That is why both columns are here, and why each has its own scale.

By month · how busy, and what it cost
RunsCost
2026-0412$0.11
2026-051,086$8.80
2026-0641$0.98
2026-07200$33.61
2026-0869$23.42
Sonnet

What the names in the tables mean

The names in the tables are the steps each agent logs, not the agents themselves. Several agents log more than one kind of step, and one name on this list is not an agent at all.

The logged name, and who is behind it
The logged nameWho that isWhat the step does
ask.explorerAsk assistanta visitor asking a question on a solution page
chief_shadowWire Editorthe Wire Editor marking its own shortlist afterwards
content_agentContent agentthe blog agents drafting from a code change
directorDirectora turn of conversation, usually started by a message from me
ghost.create_draftnot on the rosternot an agent — the blog itself accepting a post
marketer_agent.extractMarketer agentpulling descriptors out of a finished draft
marketer_agent.packageMarketer agentturning those descriptors into titles and tags
scout_synthesisScoutthe same run looking for connections, on the expensive model
scout_walkScoutthe Scout's cheap first read, looking widely for anything worth a look
sysadminSysadminthe 06:20 health check, reading the server end to end
wire_triageWire Editorthe Wire Editor sorting inbound items
writer_draftWriterone attempt at a field note from a claimed lead
Agent Observability

The system that produced this page

Counted 2026-04-25 to 2026-08-09. Snapshot taken 2026-08-09 — the figures move only when the snapshot is re-run, never while you read.