You is the fourth destination in the top bar, and the only personal one. It shows how you work with your AI tools, what that costs, and the results. The other three destinations cover your organization's playbook.

Outcomes in the top bar shows organization-wide measures. You / Outcomes shows the same measures for your own use.

The registry cannot collect this view. Every figure on it is read from your own machines, and the aggregate-never-individual rule that governs the rest of the dashboard is the reason: no one, including an administrator, can open your numbers from here.

It opens on Now. The last breadcrumb is a menu with the other three views. The period control at the top right sets the window for all of them.

Now — the state

This view shows:

• Allowance used — how much of your five-hour window is gone, and when it resets. Click it for the climb, the projection of when it runs out at the current rate, and your weekly and spend windows where your provider reports them.
• Open right now — sessions running on this machine.
• The period's sessions, turns, tool calls, retries, and corrections.

Work — mechanics and friction

This view shows turn times, errors, and successful tool calls. Work schedule shows which days and hours you worked and the length of each sitting. The days are UTC; the hours use your machine's clock, and the page states both.

Tool calls open their own page: which tools, which programs inside a shell command, which subagents and skills.

Cost — money and tokens

This view shows cost, input and output tokens, the prompt-cache split, and where work is handed off.

Two kinds of money appear here and they never add together. Where your client reports what a session cost, that figure is a measurement. Where it does not, the page estimates the cost from published token rates and gives the date of those rates. It omits models that have no published rate.

A cache read costs a fraction of fresh input, so the cache split can show where restarting sessions increases cost.

Outcomes — what came of it

This view shows the shown → adopted → helped funnel for the techniques you were offered. It uses the same layout as the organization's Outcomes view and places the registry's rate beside yours at each step. For example, an 88% adoption rate has different context when the organization rate is 68% than when it is 90%. Below the funnel are a per-technique breakdown and the techniques you stopped using.

Was it checked shows how many sessions with changes ran a test, build, linter, or type check, and what the last check reported. It reads the command and result during the session and keeps neither.

The panel breaks that down by model and client together, with cost, tokens, tool calls, working time and turns per changed session on the same rows. The page names both because the client controls the tools, prompts, and stop conditions, so the same model can yield different results in two clients.

A passing check means that a check ran, its output indicated a pass, and no edit followed it. It does not mean the work met your request because this view does not know the request. When an edit follows the last pass, the session reads stale. When a check ran but did not report a result, it reads unknown.

The view has four limits:

• It shows no confidence band. Your sessions are self-selected and their tasks change from one to the next, so an interval would assume trials nobody ran.
• It ranks nothing. Rows are ordered by how much evidence each rests on, which is not the same as how well each did.
• Under 30 changed sessions, a row is marked thin and shows counts instead of percentages. OpenTacit does not pool thin session types.
• It never guesses why something failed. A failed tool call is an operational fact and stays one; missed requirements, regressions and wrong files are verdicts that need a verifier this has not got.

A session joins the panel when an edit tool runs or when your status line reports changed lines. OpenTacit may miss a session that edits only through the shell, so the panel lists its data sources. It does not count a missed session as unchecked.

Code still present shows how many lines written by your agents remain in the branch. The page cannot measure this on its own because it stores project names but not paths. Run this command in the repository:

tacit usage --landed --publish

Run it inside a repository. It reads that directory and nothing else, writes nothing to it, and files the result for this page. The result does not refresh on its own. It includes the date when you ran the command and leaves the page after three months.

Where the numbers come from

Your machine keeps two files with mode 0600 and uploads neither. Full session records remain for 30 days in sessions.jsonl; after that they become one row per day per client, model, project and kind of session in session-days.jsonl, which is kept for 400 days. Neither holds a prompt, a completion, a command, a path, file contents, a transcript or an identity. Everything in them is a count, a duration, a tool name, a project basename or a model label, and the session key is hashed.

Check evidence adds four counted facts to a session record and nothing else: whether a change was observed, how many recognised checks ran, how many passed, and the one word the latest check left it on. The command and result are matched while the hook event is handled and then dropped. None of this reaches the registry, the organization's Outcomes, an export, or a federated peer. Unlike other dashboard data, it has no organization-wide aggregate.

Sessions recorded before OpenTacit added these counts have no check data. An older window may therefore show no evidence; OpenTacit does not count those sessions as unchecked.

Cost, lines written and cache figures come from your client's status line. tacit init offers to wire it. Every panel says how many sessions in the window reported these figures and which clients reported any data. This distinguishes a quiet week from a client that has not been set up.

If the registry runs on a different machine from the one you work on, your machines seal their usage and publish it here, and only your key opens it. Run tacit usage --key on a machine you work on and paste what it prints. The key stays in your browser and is never sent to the registry, which is what lets the registry hold your numbers without being able to read them.

For the same figures as text, run tacit usage. See the command reference.