Two providers · two host environments · bounded page

Machine-wide LLM observatory

The price of anything is the amount of life you exchange for it
What this observatory measures.

Agent Telemetry is this operator’s passive, anonymized and aggregated observatory for Claude Code and Codex CLI activity across WSL and Windows. It publishes metadata rather than prompts, messages, code, paths, or private project mappings. Dollars are exact-model API-list-price equivalents where available, not subscription bills or invoices; collection targets 30 minutes, publication is daily, and live data age appears at right.

Loading bounded telemetry…

01 / Headline telemetry

Machine-wide overview

Checking totals
Exact UTC window
Preparing exact window…

02 / Windowed economics

Activity & cost

Exact selected-window totals; plotted series use at most 48 consecutive UTC buckets.

03 / Attribution

Where usage lands

Anonymous-or-approved identities · top six plus an exact other rollup
Inspect the bounded project ranking

04 / Human attention

Attention economics

Recorded attention and API-equivalent cost remain separate units. No productivity or project-worth score is produced.

Checking the selected window for explicitly recorded attention…

Five-mode composition

Explicit timer classifications
Derived

Resource ledger

Top six projects plus an exact other rollup
Derived
ObservedDirectly captured vendor values or operator-started timer intervals; “recorded” attention is incomplete when the timer is not used.
DerivedDeterministic arithmetic over observed rows, without combining unlike units.
Self-reportedValues the operator deliberately enters in this browser.
ScenarioCounterfactual results that depend on those assumptions and are never observations.
Unknowable herePrompt content, cognitive workload, output quality, causal productivity, and actual time saved are not observed.
Counterfactual scenario
ScenarioSelf-reported inputsBlank by default. Inputs remain only in this open page and are not stored or sent.
The aggregate other row cannot be modeled as one project.
How many hours the same project would require in the counterfactual.
Must be positive; no wage or value is assumed.
API-equivalent dollars are not an invoice and never combine with actual cash.
A nonempty browser-only name is required.
From 0 through 100 percent; no interruption penalty is assumed.
May be zero; no value is inferred.
Complete every assumption to see scenario results. Nothing is stored or sent.
What this cannot prove
  • Session span is not human attention.
  • Agent elapsed time is not time saved.
  • No recorded attention does not mean no work or no progress.
  • API-equivalent cost is not an invoice.
  • Recorded project transitions are counts, not a fixed time or cognitive penalty.
  • Scenario results depend on explicit user assumptions.

05 / Outcome depth

Feature outcomes

Governed-loop evidence only · UTC completion window
Inspect the bounded feature-cost ranking

06 / Continuity

Stability & coverage

Checking system
Inspect current doctor checks and provider roots

07 / Flagship evidence

Loop evidence

Checking joins
Inspect six recent feature records

Metric catalog

Metric definition