Training
Training · Lead 8

Knowing it's working

Whether the system is compounding or quietly going stale is readable in three signals you can check yourself, once a month.

Written in your agent’s voice: this is how it talks to you in your workspace.

The system is supposed to be worth more every month, and that claim is testable in real things: knowledge files added, feedback closed, skills shipped.

When those three are moving, the compounding is running.

When one flattens, it names the move that fixes it.

Signal 1 — Knowledge files are being added

Each domain carries its own knowledge files — what the domain has learned, saved as it works.

That body should grow as the work gets done, not sit frozen at the day the domain shipped.

Good looks like new knowledge files and new decisions every month.

Stale looks like the same files that shipped, no new decisions — the team is using the system as a static tool, and the compounding never started.

To pull the read on a domain, ask the agent in a session: "What’s been added to this domain’s knowledge and decisions in the last month?"

Signal 2 — Feedback is being closed

When an end user hits something wrong, they say "turn this into feedback" — and it lands in that domain’s feedback log through the /feedback pipeline.

Logging it is half the loop; the signal that matters is whether the owner is acting on it.

Good looks like feedback coming in and the owner closing it — corrected, pruned, folded into the knowledge.

Stale looks like feedback piling up with no one acting on it, or a domain with no named owner to act.

The owner who closes feedback is the difference between a domain that sharpens and one that just gets longer.

When this signal is flat, ask the owner straight: "What feedback have you closed in your domain this month?"

Signal 3 — Skills are shipping

A skill is a repeatable job the domain’s agent can run on its own — the method, saved so it runs the same way every time.

New skills shipping means owners are turning what piled up into capability the whole team gets, not just a bigger pile of notes.

Good looks like an owner shipping a new skill, or sharpening one that already runs.

Stale looks like no new skills since delivery — accumulation with nothing built on it.

To check, ask the owner: "What can your agent do now that it couldn’t a quarter ago?"

The payoff: your agent is better month over month

The three signals produce one result you can feel.

An agent that has absorbed a quarter of added knowledge, closed feedback, and shipped skills holds context a new hire would take years to build.

Good looks like less correction on the work it does often, and more of your context held without re-explaining.

Your owners and heavy users feel this before any signal shows it — so ask them: "Is your agent better at your work than it was a month ago?"

Read the three together

Each signal alone can mislead, so read them as a set.

Knowledge piling up with no feedback closed is hoarding; feedback closed with no skills shipping is a method that never reaches the team.

The healthy pattern is all three moving at once: knowledge added, feedback closed, skills shipped.

When one flattens, it names the move — a domain needs an owner, an owner needs time, or a team needs reminding to feed what they use.

The one thing to remember

The compounding is readable in three real things, checked monthly: knowledge files added, feedback closed, skills shipped.

Read them as a set, and act on whichever one goes flat.

Next: the appendix — reference modules, on demand

Your agents. Your OS. Deployed in 60 days.