Metrics on the Projects pages
Your organization's metrics live in one place: the Projects pages of the Portal, in the top navigation, at /<org>/projects. You don't need to install anything: open the Portal and click Projects. (Older /<org>/insights links still work and send you here.)
Who can see it
Every role of the organization: Owner, Admin, Member and Viewer. BAs, POs, and PMs (Viewers) are the main audience. You only see the organization in the address, and you can't see another organization's data, even if you belong to both. Platform staff get no extra access to a customer organization.
The numbers come from Ryoma, which collects AI sessions and task progress for each organization. The Portal asks Ryoma on your behalf, so the numbers are always the ones for the organization you have open.
What is measured
Only two frameworks:
- The Flow Framework: flow velocity, flow time, flow efficiency, flow load (work in progress), and flow distribution by work type.
- The AI measurement framework: AI usage, adoption, cost, and the share of work that used AI.
There are no DORA metrics and no surveys. Everything is measured per project and per workflow, never per person. Compare a project with itself over time, not with other projects.
Time range
At the top of the Projects list and of every project page you pick a time range (last 7, 30, or 90 days, or a custom range). It is kept in the address, so you can share a link with exactly the view you're looking at, and it carries over when you open a project.
Charts group work by week, starting Monday. The range is rounded to whole weeks, and the longest range is 51 weeks. Task counts by status, forecast cost, and at-risk tasks always show the state right now, so the range doesn't apply to them.
The Projects list
/<org>/projects is the view for the whole organization:
- Four headline numbers, each compared with the previous week: items done, flow time (median), items with AI, and AI cost (with the cost per done item underneath).
- Two charts: items done per week by work type, and AI cost per week.
- One table with a row per project: repositories, tasks by status, items done, flow time, flow efficiency, work in progress, items with AI, AI cost, and at-risk tasks. Click a column to sort and a project name to open its page. Repositories that belong to no project are mentioned in a note under the table.
A project page
/<org>/projects/<key> shows one project, with the same time range:
- Header. The project's repositories and links (Bandum board), and a control to link or edit repositories if you're allowed to.
- Release approvers. Who can approve releases of the project in Bandum. Organization owners always can. With no list, owners and admins can; with a list, only the people on it and owners can (an admin who is not on the list cannot). Owners and admins change it with Change.
- Headline numbers. The four above plus flow efficiency and work in progress.
- Flow. Items done per week by work type, flow time (median and p85), flow efficiency, work in progress per day, and the work mix. A warning appears when too few finished items carry a work type label.
- AI. Items with AI, AI cost per week (all of it and the part tied to tasks), cost coverage, and cost per done item by work type. A note appears when little of the cost is tied to tasks.
- Before and after AI. The weeks before the project started using AI next to the same number of weeks after it: tasks finished per week, flow time, flow efficiency, work in progress, first-time approval, and rework. See below.
- Monthly report. Download a month as a one-page PDF or as CSV. See below.
- Releases. The latest releases of the project, each with its release wait time and its AI cost. Release wait time runs from the moment the release was ready (its last task approved or closed) to the moment the release was created. It is one part of the flow time of those tasks, measured per release. AI cost is the AI cost of the release's tasks added together. A release made outside Bandum has no approval record, so it shows dashes. The section is empty until a release is approved in Bandum.
- Tasks. Tasks by status, first-pass approval, rework rate, average time in each status, and forecast cost.
- At-risk tasks. The unfinished tasks that need attention, and why.
Before and after AI
Did the project get faster after it started using AI? The Before and after AI panel of a project page puts two periods side by side: the 8 weeks before the project started using AI and the 8 weeks after. Both come from the synced task history, with the same Flow definitions as the charts. A Change column shows how much each number moved.
- The start date. By default it is the first day Ryoma received telemetry from the project's repositories, and the panel says so. An organization owner or admin can change it with Change start date (for example when a team started earlier on another tool), and go back with Use first telemetry day. Other roles can read it.
- Not enough tasks. When fewer than 10 tasks finished in a period, the panel warns that the comparison is not reliable. Read it as a hint, not as proof.
- Recent start. When the project started using AI less than 8 weeks ago, the second period is shorter. Compare the per-week numbers, not the totals.
- No start date yet. Until telemetry arrives and no date is set, the panel only tells you that.
It is a correlation, not proof of cause, and it is measured for the project, never for a person. There is no AI cost for the period before, because there was no AI to measure.
Monthly report
On a project page, and at the top of the Projects list for the whole organization, Monthly report lets you pick a month and download:
- PDF, one A4 page you can hand to someone who is not on the team. It names the period and the date it was generated, then gives a few plain sentences, then the numbers of the month next to the period before AI: AI cost per approved task, how many tasks were approved the first time, how many were sent back (rework), and flow time. It ends with data notes, for example missing labels or machines that sent no telemetry.
- CSV, the weekly numbers: one row per project and per week that touches the month, with the Flow numbers and the AI cost of the week.
Every role of the organization can download it. The current month is partial and runs up to today. The files are built from the same figures the Projects pages read, and they never contain numbers about individual people.
Guardrails counts
How often Guardrails blocked, asked about, or masked something is not a project metric, so it lives with the policy: Settings → Guardrails policy, section What guardrails did (last 30 days). It shows four counts (Block, Ask, Mask, False positives) and a table by project and rule. It shows counts only, never names, and everyone who can open that settings page sees it.
What each metric means
| Metric | What it tells you |
|---|---|
| Items done (flow velocity) | How many work items were finished in the period. |
| Flow time | How long an item takes from the moment someone starts it until it is done. The median means half of the items finish faster. p85 means 85% finish within that time. |
| Flow efficiency | The share of an item's time spent actually being worked on, versus waiting for testing, fixes, or merge. A low number means work waits more than it moves. |
| Work in progress (flow load) | How many items are started but not finished, on average per day. Many items at once usually means each one takes longer. |
| Work mix (flow distribution) | The share of finished items by type: feature, bug, debt, or risk. It shows where effort goes. |
| Work type coverage | The share of finished items that carry a work type label. Below 80% the mix is not reliable yet. |
| AI users | How many different people used AI in the project's repositories. Only counted, never named. |
| Items with AI | The share of finished items that had at least one AI session tagged with the task. |
| Cost coverage | How much of the AI cost can be tied to a task. Low means many AI sessions don't record which task they were for. |
| AI cost | The converted cost of AI usage at API prices. It is an estimate, not an invoice. "Attributed" is the part tied to tasks. |
| AI cost per done item | The AI cost of the items finished in the period, divided by how many were finished. |
| Release wait time | How long a finished release waited, from the moment its last task was approved or closed to the moment the release was created. One part of an item's flow time, measured per release. |
| AI cost per release | The AI cost of all the tasks in a release added together, at API prices (an estimate). A dash means none of its tasks had an AI session. |
| First-pass approval | The share of tested tasks approved on the first test run. |
| Rework rate | The share of tested tasks that were sent back at least once. |
| At-risk tasks | Unfinished tasks that wait too long for testing, stay too long in progress, were sent back many times, or cost much more AI than usual. |
| Forecast cost | The estimated AI cost of finishing the remaining work, based on similar finished tasks. |
| Guardrails decisions | How often Guardrails blocked, asked about, or masked something. Counts only, by project and rule, never names. |
| False positives | Blocks that people reported as wrong. Many reports on one rule mean the rule needs tuning. |
A metric with no data shows a dash (—), never a made-up zero.
When a page is empty
A new organization has nothing to show yet, and that's normal. Each empty state tells you the next step:
- No projects. Connect your GitHub repos and link them to a project. See Connect GitHub and Organizations.
- No AI numbers. Install Phillip on a developer machine and run a task, so telemetry starts to arrive. See Get started for developers and testers.
- No Flow numbers. Tasks come from GitHub through Bandum. Run tasks with
bandum startso they move through the statuses. - Work mix looks wrong. Add a label
type: feature,type: bug,type: debt, ortype: riskto each task.
New numbers can take a few minutes to appear. The Portal keeps results for 45 seconds.
Use the same numbers in the terminal
ph insights # overview, one row per project
ph insights flow --days 90
ph insights ai --project shop-web
ph insights health --json
ph insights flow --compare --project shop-web # before and after AI
ph insights releases --project shop-web # release wait time and AI cost per release
ph insights report --month 2026-12 --format pdf --project shop-web
ph insights report --month 2026-12 --lang vi # the PDF in Vietnamese (en is the default)
ph project release-approvers shop-web set maya.chen bo # owner or admin: who can approve releases
ph project adoption shop-web --set 2026-09-01 # owner or admin: change the start dateThe form is ph insights [flow|ai|health|releases|report] [--project key] [--days N] [--from YYYY-MM-DD --to YYYY-MM-DD] [--json]. It prints the same numbers as the Projects pages, and every role can use it. --json prints the raw data. ph insights report saves the PDF or CSV of a month to a file (-o); --lang vi writes the PDF labels, the summary and the data notes in Vietnamese (the Portal download follows the language you use there), and ph project adoption <key> shows, sets or resets the date a project started using AI. See ph insights.