Insights & Observability

Use `spl insights` to surface proactive observations the system computes from its own activity log — trust drift detection and lingering open work — so problems find you before you go looking.

Overview

Your instance is constantly computing derived views over its own record log: trust scores, task state, calibration signals. Most of the time you ask for those views explicitly — spl trust, spl task list, spl status. spl insights flips that around. It surfaces the observations the system already noticed on its own, so the things worth your attention come to you instead of waiting to be queried.

spl insights

Insights:
  ⚠ Trust drift: did:sync:agent:dev in 'code' — a shift in recent verdicts
    was detected after 302 observation(s); review recent reviews.

If nothing needs attention, spl insights says so — no noise.

What it surfaces

Insights are folded from the same record log everything else derives from. Two kinds today:

Trust drift

The trust machinery watches each actor's stream of verdicts for a change in pattern — a run of acceptances that turns into a run of rejections, or vice versa. When it detects a statistically meaningful shift, it raises a drift signal naming the actor and domain.

A drift signal is a prompt to look, not a verdict. An agent whose recent work in code started getting rejected might be hitting a class of task it's weak at, working against a moved target, or running on a changed underlying model. The signal tells you where to look; the trust page explains how the scoring works.

Lingering open work

Insights also flags open work that's been sitting — threads and tasks that opened but never closed, well past the point where you'd expect them to. This is the same signal that feeds task triage, surfaced proactively: work that's quietly stalled, so you can re-engage, reassign, or close it.

When to run it

spl insights is cheap and read-only — run it whenever you want a pulse on what's drifting:

At the start of a session, alongside spl status, to see if anything moved while you were away.
After a burst of agent work, to catch trust drift early rather than discovering it three rejections later.
Periodically, the same way you'd glance at task triage — to keep stalled work and reputation shifts from accumulating silently.

How it relates to `spl status`

The two are complementary:

Command	Answers	Character
`spl status`	"Is the system healthy right now?"	A snapshot: uptime, record and thread counts, engine counters, recent activity. Plus, best-effort, the top task suggestions.
`spl insights`	"What has the system noticed that I should look at?"	A focused list of observations — drift and lingering work — distilled from the log.

spl status is the dashboard; spl insights is the alert feed. Status tells you the instance is up and how much it's done. Insights tells you which corner of that activity deserves a second look. Run both at the start of a session: status for the health check, insights for the watch-outs.

Insights pairs naturally with spl task suggest — insights tells you what's drifting, suggest tells you what to do about it.

What's next

Task suggestions & triage — turn the backlog into a next-move shortlist and a close-list.
Trust — how drift detection and verdict scoring work.
CLI reference — full command surface, including spl status.