Dashboard and analytics

Two surfaces, two jobs. The dashboard tells you what to do today. Analytics tells you whether what you've been doing is working.

7.0 The one thing to do next#

The top of the dashboard is a single card: what to do now, why, and either the page to do it on or the prompt that does it.

It is deliberately one thing rather than a list. It picks by what blocks the most — a queue you have to clear stops every downstream step for every record in it, so clearing it outranks starting new work. If you only ever read one part of this app, read that card.

Under it, a bar showing how much of your list has actually been through the pipeline: not researched, being researched, waiting on you, researched, matched, drafted, contacted, replied. Waiting on you takes precedence over everything else — a professor who is already matched but has new findings waiting is, for your purposes, blocked.

7.1 The dashboard, in order#

Everything on the dashboard is arranged by urgency, and it's meant to be read top to bottom.

Needs you. Cards for decisions you can act on right now, sorted by urgency and then by volume. Amber means time-bound — a follow-up due today, a deadline inside two weeks, an interview needing prep. Blue means waiting on you — drafts to review, strong matches with no draft. Plain grey is upkeep, like stale research. When there's nothing, it says so.

One card you will see often is Findings to publish. It means an assistant checked some research against its sources and it is sitting in Review, not yet on the record. Nothing downstream — matching, drafting — can cite it until you publish it, which is why it sits first in the deck.

Recent activity. What has happened lately, newest first, in the assistant's own words: Recorded lab — created, Checked findings — everything checked out, Attached evidence.

This exists because a research run happens somewhere else. You start it in Claude Code or Codex, it takes an hour, and the app has no other way to tell you it is alive. The feed refreshes itself every fifteen seconds, so you can leave the dashboard open on a second screen and watch a run land. Each line links to the record it changed.

Imports are left out — one spreadsheet writes twelve thousand events and would bury everything else. Repeated actions collapse into one line with a count.

Contact next. Your strongest matches you haven't written to, in score order. Click straight through to a professor's email tab.

Replied — your move and Waiting on a reply. Conversations in flight. The first is the one to work.

Upcoming · 14 days. Follow-ups, interviews, deadlines, offer responses.

Recently researched. What came out of your last batches, with each professor's score.

Where things stand. Twelve numbers in three groups — research, outreach, applications. Click any of them to see the records behind it.

Activity and Conversion funnel. Two charts: what you've been doing lately, and where people are dropping out.

Pipeline and Research batches. How far your professors have got, and your batch state for the week.

7.2 Clicking a number#

Many figures open the records behind them. Clicking one shows a drawer listing the exact records that produced it — the same ones, not a re-derived approximation. The chart or table you clicked stays on screen behind it, so checking what a number is made of doesn't cost you your place. From the drawer you can open any record directly, or jump to the full filtered list.

Which ones are clickable follows a rule worth knowing, because it saves you clicking hopefully at figures that will not respond:

• On the dashboard, all twelve metrics are. The whole strip is built to be opened.
• In Analytics, counts of records are — total universities, total professors, researched, stale research, high-match, emails sent, interviews scheduled, planned, submitted, acceptances.
• Rates and averages are not. Response rate, average match score, average days to reply, acceptance rate, average sources per professor. A percentage is not a set of records, so there is nothing coherent to list; the counts it was computed from usually are clickable instead.
• A few counts — publications indexed, sources collected, research jobs completed — are not wired up yet either.

This is worth using whenever a number surprises you. "Response rate 40%" means something quite different once you open Emails sent and find the denominator is five.

7.3 Analytics#

Full screen

Filters sit at the top: university, priority, degree type, and a date range. They apply to everything below. Clear filters resets them.

Reading rates honestly#

Percentages come with the sample they're drawn from — 18% (n=11) — because a rate over eleven emails is a hint and a rate over two hundred is a finding. Very small non-zero rates show as <1% rather than rounding down to a flat 0%, which would read as "none".

What you are working with#

The first section under the funnel describes the shape of your list rather than your progress through it, and it deliberately ignores the filters above — it answers "what am I working with", which is a question about the whole workspace. The section says so on its face, because a chart that quietly respects a filter set three screens up is how people misread their own data.

Researched over time — professors and universities, cumulative. Cumulative because the question is whether the pile is growing and how fast, which a per-day chart of ones and twos does not answer. Dated by when each record was last checked against its sources.

Public and private — who runs each university. Not established is its own bar and is deliberately grey: it is not a third kind of university, it is research not yet done. Nothing is guessed from the name; the field is filled in from a source like any other fact.

Roster sizes — how many universities fall in each faculty-count band, with the mean, the median, and the largest roster underneath. Both averages, because they answer different questions: one 900-person institution among a hundred 40-person ones drags the mean somewhere no university actually sits.

Pipeline — every professor by how far they have got, in pipeline order. Waiting on you is amber wherever it appears, because it is the one state where nothing moves until you act.

Match scores — scored professors in ten-point bands. Ten-point, because a fit score is a judgement to one or two significant figures and a single-point histogram implies a precision it does not have.

Roster coverage — faculty on file against how many of them have been researched, for the largest rosters. Two bars rather than a stack: researched is a subset of on-file, and stacking a part on top of its own whole would draw a total twice the truth.

Publication years — when the papers on file were published, so you can see how current the evidence you are citing actually is. This counts published papers only, matching the Publications indexed figure below it. Anything still in Review is reported as a count in the footer rather than drawn: it isn't evidence you can cite yet, but a chart that silently omits it would hide the fact that the publish gate is holding work up.

The sections#

Inventory — how much you've got and how much of it is real: universities, professors, researched professors, research completeness, stale records, publications, sources, documents.

Matching — average score, how many are high-match, the score distribution across the 0–100 range, where your strong fits are concentrated by university, and which dimensions score highest across your whole pool. That last one is quietly useful: if topic fit is consistently strong and active research direction consistently weak, you're looking at the right field but the wrong generation of work.

Outreach — emails sent, professors contacted, response rate, positive and negative rates, no-response rate, follow-up response rate, and average days to a reply. Plus how your replies landed, and formal versus informal performance side by side.

Interviews — scheduled, completed, and how often a positive reply turns into an actual conversation.

Applications — planned, submitted, checklist completion, interviews, acceptances, rejections, waitlists, funded offers, acceptance rate.

Research operations — how much research has run, what failed, average sources and publications captured per professor, and how long a professor takes to get through a batch.

The funnel#

The funnel is the fastest diagnosis in the app, because the stage where the drop is steepest tells you what to fix:

Click any funnel stage to see exactly who is sitting in it.

7.4 A weekly rhythm that works#

• Daily, 5 minutes. Dashboard. Clear Needs you. Answer anyone under Replied — your move.
• Twice a week, 30 minutes. Run a batch, review what came back, draft one or two emails.
• Weekly, 15 minutes. Tracking → Upcoming for the month ahead. Then Analytics: the funnel and the response rate. Change one thing, not five.

What's next#

Tune the parts that shape all of the above: settings →