Glossary

Work log

The record that makes an unwatched run reviewable afterwards.

Definition

A work log is a per-job record of what was done to complete a unit of work: the steps taken, the sources consulted, the artefacts produced, and the time attributed to each. In agent systems a work log is the primary audit surface, because nobody watched the run, and it is the evidence a reviewer or a payer checks a claim against.

Granularity: One entry per job
Primary purpose: Audit and billing
Written by: The runtime, not the model

General usage and Polaris usage

In general usage, a work log is any chronological record of work performed: the timesheet, the engineering logbook, the case notes. Professions that bill by the hour have kept them for a century, for exactly one reason: a client who cannot see what was done has no way to check what was charged.

Polaris uses the term in that older billing sense rather than as a generic activity feed. Each AI worker has a work log listing every job it completed, what the run consisted of, and the human-equivalent hours attributed to it, which is what the bill is computed from.

What belongs in an agent work log

The observable effort
How many searches were run, how much finished prose was written, how many checklist items were ticked, how many comments posted, how many files produced. Countable things, not the model's own account of its diligence.
The artefacts
Every file the run produced, attached and retrievable, so a reader can check the claim against the output.
The attributed time
The estimate the billing formula produced from that effort, shown per job rather than rolled up into a monthly total.
Failures and retries
Runs that errored, and how many attempts were made. A log showing only successes is a marketing artefact.

Commonly confused with

Term	What it records	The difference
Activity feed	Discrete events: assigned, status changed, run started	Event-shaped and system-wide. A work log is job-shaped and effort-shaped.
Chat transcript	Every message in a conversation	Complete but unstructured, and far too long to audit routinely.
Audit log	Security-relevant actions and who performed them	Answers who touched what. A work log answers what was produced and what it cost.
Timesheet	Hours a person claims to have spent	Self-reported. An agent work log is computed from observable effort.

Why it decides whether usage billing is trustworthy

Any usage-priced product faces one question from a buyer: how do I know you are not inflating the meter? The only durable answer is a record detailed enough that a customer can recompute the number, and a formula published clearly enough that they can check the arithmetic.

That is why Polaris shows the human-hours estimate job by job on the worker's work log and publishes the formula behind it. Any line on the bill can be challenged from the log entry that produced it.

Human-equivalent hours

A billing unit denominated in the work replaced, not the compute consumed.

Usage-based pricing

The bill follows consumption, which cuts both ways for buyer and vendor.

Delivery comment

Work arrives where the task already lives, attributed and reviewable, and the task stays open.

Headless agent

Nobody is typing at it, so its output is actions and artefacts rather than replies.

The audit trail behind every agent session

If you cannot reconstruct what happened three weeks later, you do not have an audit trail. You have a feeling.

What an AI worker costs, and how the hours are counted

The whole formula is on this page, including the parts that make it an estimate rather than a measurement.

Questions people ask

+Who writes the work log, the AI or the system?

The runtime writes it, from counters it maintains while the job runs. Letting the model report its own effort would make the log an assertion rather than a measurement, and the whole point of the artefact is that it is not the agent's opinion of its work.

+How long should work logs be kept?

At least as long as the billing period they support, and in practice longer, because disputes and quality reviews arrive after the fact. Teams using agent output in regulated work usually align retention with whatever their record-keeping obligations already require.

+Can a work log be challenged?

In Polaris, yes: each job's hour estimate is shown on the worker's work log next to the effort it was computed from, and the formula is published. A line you disagree with can be disputed from the entry itself rather than from an aggregate invoice.