Insights

Insights answers the questions status meetings exist for, without the meeting. It’s one page of plain counts, and every number opens the exact issues behind it. Numbers. No spin.

Before you start

Anyone in the workspace can open Insights. It’s read-only, so there’s nothing to break. Who can do what.
Insights uses the workspace’s planning timezone for weeks and dates. Until an Admin confirms one under Settings › Calendar, it uses UTC and says so. See Confirm the planning calendar.
The Cycles section and the Current cycle window need cycles turned on. Everything else works without them.

Steps

Set the scope

Open Insights in the sidebar. Two pickers at the top scope everything above the Cycles section:

Window: Last 7 days, Last 4 weeks (default), Last 12 weeks or Current cycle. Last 7 days is today and the six days before it, each from midnight in the workspace timezone.
Who: Everyone, one person or agent, or Unassigned.

The line underneath shows the dates, the timezone and when the numbers were taken. The page refreshes itself every minute while you’re looking at it.

Read the tiles

Four tiles, each a count you can click for the list of issues:

Tile	Counts
In progress	Issues in progress right now, and how many people, agents or nobody hold them.
Waiting for review	Issues in a review stage right now, and how many have waited over 3 days.
Blocked	Open issues with a direct blocker that isn’t Done or Canceled, and how many blockers.
Delivered	Issues that reached Done in the window, how many were captured, and whether the backlog grew, shrank or held.

The first three are right now. Delivered is the window. Parent issues that group sub-issues are never counted in the tiles.

Read In versus out

In versus out compares issues captured with issues delivered, week by week, as paired bars. Weeks start on Monday. Beside the chart:

N in, N out, and which way it’s going, with a streak when it’s lasted two weeks or more. Both numbers open their issues.
Delivered per week, averaged over the complete weeks in the window. The current week is “so far” and never drags the rate down.
Median from capture to Done: calendar days from creation to first Done, across the issues delivered in the window, with how many that is.

The chart waits for 2 complete weeks of history rather than drawing a line through one dot. Same numbers as a table is there if you’d rather not read bars.

Last 7 days works the same way, one day at a time: seven daily bars from midnight in the workspace timezone, with today “so far”. The rate is delivered per day over the complete days, the streak counts days, and the chart waits for 2 complete days of history.

Read Activity

Activity shows who is doing the work, people or agents, over the same window and timezone as the rest of the page. The line at the top gives the number of actions and the share agents took, and how that share moved against the same stretch just before (Last 7 days, Last 4 weeks and Current cycle; Last 12 weeks has nothing that far back to compare with). Four charts follow:

Actions per day (per week on Last 12 weeks): people solid, agents hatched. Stack by: Crew splits the bars by crew instead, the three busiest crews and everyone else, each with its own pattern.
Busiest hours: a grid of weekdays by hour, darker for more. Hover a cell for its count, how many of that weekday it covers and the agent share.
Work by crew: one bar per person, their own work solid and the agents they ran hatched, with each agent named.
Agent share: the share of actions agents took, per day on Last 7 days and per week otherwise.

Every chart has Show as table, which swaps it for the same numbers in place. Changes by the scheduler and the GitHub integration are left out of the charts and counted in one line under them.

Who narrows Activity too. A person narrows it to their crew: their own work and every agent action they ran. An agent narrows it to that agent’s actions, whoever ran it, and Work by crew becomes Who ran it. Unassigned has no Activity: unassigned issues don’t take actions.

The charts wait for 2 full days of history, so one busy morning doesn’t pass for a habit. Until then you get the count so far. What counts as an action is under How activity is counted.

Read Where it waits

Where it waits shows how long the issues delivered in the window spent in each stage, so you can see where the queue forms. There’s one row per Ready, In Progress or review stage they went through, in your workflow’s order, each with its median and how many issues that is. The line above names the longest stage.

Every stay counts. An issue sent back from review and reworked adds up all its laps, so rework isn’t hidden.
It’s calendar time. Snoozed and blocked time stays in the stage the issue was in.
Backlog and Iced aren’t shown. Capture to Done already covers the wait before work starts.
A stage with fewer than 5 issues gets a striped bar and “too few to trust”.

Review wait adds up every stage that behaves as review for each issue, the same stages the Waiting for review tile counts, and gives the median. It also says how long the longest current review has been waiting; Needs a decision lists the ones waiting over 3 days. If an environment has approvers, there’s a line for it too: the median time from an issue reaching it to an approver’s first answer (approve, hold or reject; changing their mind later doesn’t count again), and the longest wait there now.

Who and Window apply here. With Who set, the issues are the ones that person held when they moved to Done.

Clear the Needs a decision list

Needs a decision lists issues that have been in a review stage for more than 3 days, and open issues that are holding up other work, oldest first. When it’s empty: “Nothing is waiting on you. Enjoy it; it won’t last.”

See who’s holding what

Who’s holding what has one row per assignee: how many issues they have in progress, in review, ready and blocked. The Unassigned row shows what’s ready and in the backlog with nobody on it. Open as issue list takes you to the same work as a list.

It’s issue counts, not hours. There’s no ranking, no score and no leaderboard. It shows where the work is sitting, not who’s winning.

Read the Cycles section

With cycles on, the bottom section shows:

Velocity: the range of issues delivered per closed working cycle, over the last closed cycles, with how many that is. “A range, not a forecast.”
The running cycle’s burnup. See Cycles.
A table of each cycle’s Committed, Delivered (with anything added mid-cycle) and Carried over, plus the running cycle so far.

This section covers the whole workspace on each cycle’s own dates. Who and Window don’t apply down here.

Check the rules

How it’s counted, in the line under the heading, opens the exact definitions. The same rules apply in the app and for agents over MCP.

What you should see

A page you can take in at a glance. If the workspace uses sizes, size points appear beside the counts: on the tiles, per assignee, in the per-week or per-day rate and in the cycles table. Unsized issues are left out of every sum and counted beside it. T-shirt sizes count XS 1 to XXL 6; Fibonacci sizes count as their own number. Sizes are ordinal, so a sum is a hint, not a measurement. See Labels, sizes and priority.

What Insights deliberately doesn’t do: forecast, estimate capacity, rank people or compare them against each other. If you want a burndown to wave at stakeholders, this isn’t the place.

How activity is counted

Activity counts what the workspace activity stream shows, one event at a time: filed, moved, refined, shipped (an environment move, an approval decision or a release), commented, edited, linked, joined, and everything else as other. Agents read the same numbers over MCP (get_activity_stats).

Everyone sees the same numbers. Nobody’s own work is left out, and Members and Admins get identical counts.
People, agents and system. Work in the browser is a person’s. A connection that acts as a named agent counts as that agent, as it was named when it acted. A connection not shown as a named agent counts under its person as Other agents; its own name never appears. GitHub counts as system.
Crews and lanes. A person’s crew is their own work plus every agent action they ran, shared agents included. Lanes split a crew into the person’s own work, each named agent they ran and their other agents.
Owners. An agent’s owner is its owner today. Hand an agent to someone else and its past work moves with it.
Normal. The same stretch of time one to four weeks earlier, in the workspace timezone, and the middle of those weeks. Days compare with the same weekday; hours with the same hour of the week. Up to 90 days back. Against normal, activity reads Dry under 10%, Trickle to 59%, Steady to 129%, Surging to 249% and Burst pipe from 250%.
Where the work stands. Beside the activity: how many issues sit in each stage right now, and how many reached Done for the first time in the window, counted as Insights counts delivered.
History stays as it happened. Work on an issue that’s in the trash still counts. Comment and description text never enters a count.
Hours and odd timezones. Hours are whole UTC hours. Older activity is kept by the hour, so a last-activity time from it is accurate to the hour. In a timezone 30 or 45 minutes off the hour, an hour that crosses midnight counts in the day it starts in.

If it goes sideways

“(no workspace timezone yet; confirm one in Settings › Calendar)” Weeks and dates are in UTC until an Admin confirms the planning calendar under Settings › Calendar.
Current cycle is greyed out. Cycles are off. An Admin can turn them on in Settings › Calendar.
Activity says it’s waiting. It needs 2 full days of history in the workspace. Give it a couple of days.
The In versus out chart is empty. It needs 2 complete weeks of history, or 2 complete days on Last 7 days. Come back next week, or tomorrow on Last 7 days.
Delivered is lower than you expected. Delivered counts an issue’s first move to Done inside the window. Reopening and finishing again doesn’t count twice, and an issue created already Done or Canceled isn’t delivered.
A person’s delivered count looks off. Delivered credits whoever held the issue when it reached Done. The right-now counts use the current assignee.
No points anywhere. The workspace’s sizing is Off, or no issue has a size yet. A cycle that closed before sizes were recorded shows none.
“The refresh didn’t come back.” You’re seeing the last snapshot; nothing was replaced with a zero. Click Retry, or check status.hydrant.dev (opens in a new tab).