Get the question right.
Start with what you are being asked to decide, explain, improve, or revisit before choosing the instrument or the metric.
EVOLVRS Research Office
You need to know what the evidence can actually support.
When the board asks how you know, a principal says the district view does not match the school, or you need to revisit a decision months later, "the data says" is not enough. Our Research Office works on the questions underneath those moments: what was measured, who is represented, what changed, and what still cannot be claimed.
Before the cabinetKnow what deserves a question.
Before the boardKnow what you can stand behind.
After you actKnow what changed and what remains uncertain.
The question behind the questionYour view
Before cabinet
Is this important enough to take into the room?
Not every difference deserves executive attention. We study how to surface patterns that matter without converting every signal into an alert or a diagnosis.
What stands out?Is the pattern large, persistent, widespread, or consequential enough to deserve a closer look?
What is still uncertain?What would you need to ask before turning the reading into a conclusion?
Before the board
What can you say when someone asks, "How do we know?"
We study what makes a statement defensible when the audience is not looking at the methodology appendix and the question is coming in real time.
Who is represented?Participation, coverage, reporting thresholds, and missing groups matter.
What is the limit?A credible answer includes what the evidence does not establish.
After leadership acts
What changed, and what can we reasonably connect to the decision?
Repeated readings can show movement. We study how to preserve the baseline, the intervention, and the later read without turning movement into a causal claim.
Keep the baseline.Remember what the district saw before the change.
Keep cause separate.A later difference does not prove what produced it.
When AI enters the work
What is changing for the people doing the job?
A tool can perform as promised while workload, judgment, skill, expectations, and decision-making change around it. We study those changes because adoption alone does not tell you whether the district got better.
What got easier?Look beyond use rates to the work itself.
What got weaker or harder?Watch judgment, learning, workload, and capability as the system changes.
What the Research Office does
Turn the question on your desk into evidence you can actually use.
The research begins with the decision, question, or change you are trying to understand. From there, we work backward: what would need to be known, what could support an answer, and what should remain unresolved.
Build a reading that can carry the decision.
Use evidence that is strong enough for the claim you may need to make and clear enough to be used outside the research team.
Show where confidence stops.
Keep coverage, participation, reporting thresholds, timing, and uncertainty visible instead of cleaning them away.
Put the rules into Signal.
Bring the research into the interface so the evidence, the limit, and the next question are present when you need them.
Current research agenda
Questions we're studying because you're already dealing with them.
The agenda follows the work: limited attention, district variation, board accountability, leadership transitions, and technology changing faster than most districts can study its effects.
Defensible attention
What should make it onto your radar?
A superintendent cannot chase every difference across a district. We are studying what makes a pattern consequential enough to surface, while keeping the system from turning normal variation into a stream of alerts.
QuestionWhat deserves a closer look?
Consider size, spread, persistence, context, and consequence rather than a single threshold.
BoundaryWhat is not a diagnosis?
Surface the signal without claiming the system already knows the cause.
UseWhat should you ask next?
Turn uncertainty into a better leadership question instead of a false answer.
Change over time
What can you learn after the district acts?
When a new initiative, professional learning strategy, process, or technology rolls out, we want the next reading to help you learn without making a causal claim it cannot support.
BeforeWhat was true at baseline?
Preserve the original question, method, participation, and organizational context.
BetweenWhat did leadership actually do?
Keep the intervention, decision, and timing connected to the read.
AfterWhat moved?
Compare like with like, then keep movement separate from explanation.
Leadership continuity
What should the next leader not have to reconstruct?
District leadership changes, but the reasoning behind consequential decisions should not disappear. We are studying what is useful enough to preserve and trustworthy enough for the next leader to use.
QuestionsWhat was leadership watching?
Preserve the questions and conditions that mattered at the time.
DecisionsWhat did the district choose?
Keep the evidence and assumptions attached to the decision.
LaterWhat happened afterward?
Return to the original read without relying on institutional memory alone.
AI + human capability
What changes in the work after AI arrives?
We are interested in what happens to judgment, reasoning, learning, workload, skill, and expectations when AI becomes part of district work, not just whether people adopted the tool.
WorkWhat moved in the workflow?
Look at time, handoffs, rework, and where human effort shifts.
JudgmentWhat decisions changed?
Watch where people rely on, check, override, or defer to AI.
CapabilityWhat are people learning or losing?
Study whether the technology strengthens capability or quietly weakens it over time.
Board-ready evidence
What makes an evidence statement hold up in public?
A board-ready statement has to be clear without pretending the evidence is cleaner than it is. We are studying the information leaders need close at hand when the questions get specific.
SourceWhere did this come from?
Make the instrument, timing, participation, and scope easy to explain.
MeaningWhat does it actually show?
Keep the descriptive finding separate from the explanation.
LimitWhat should you not claim?
Make qualification part of the answer instead of a footnote that disappears.
Research standard
If you have to explain it publicly, you should know exactly what it means.
We want the evidence to be useful in the room where the decision happens, not just technically correct in a methodology document.
- 01
Show the pattern before telling the story.
Start with what was observed. The explanation can come next, and it may require additional evidence.
- 02
Keep disagreement visible.
When self, supervisor, peer, direct-report, school, or district readings differ, do not average the difference away just to make the result simpler.
- 03
Show the limits before the number travels.
Participation, coverage, small-group suppression, timing, and method stay close enough to the reading that they can still shape the decision.
- 04
Keep the decision with the people responsible for it.
Research can narrow the uncertainty and sharpen the question. It should not quietly turn interpretation into an automated decision.
- 05
Do not confuse movement with cause.
A later reading can show that something changed. By itself, it cannot establish what produced the change.
Research Office Lead
The person responsible for keeping the evidence honest.
Research only matters here if it changes what Signal can responsibly say, show, compare, or ask.
Research Office Lead
Esha Manduva
Esha leads the work that turns research into rules the product has to follow. She helps define what a reading can support, where uncertainty needs to remain visible, how district-defined expectations should be interpreted, and what should be protected when groups are too small to report responsibly.
She works directly with Product so those decisions show up in Signal when you are looking at the evidence, not later in a methodology appendix.
Research designEvidence standardsInterpretationResearch × Product
In the district
The method has to work in a real district, not just on a clean dataset.
Real implementation brings uneven participation, reporting boundaries, different roles, different schools, competing priorities, and decisions that cannot wait for perfect information. That is part of the research environment.
Public education / current environmentFall 2026
94leaders in the completed pilot
4,000employees in the districtwide scope
50+schools and sites in the planned baseline
What this lets us learn
Whether the same evidence rules still hold when the reading moves from a small leadership pilot to an entire district, across roles and sites, with real participation, privacy, reporting, and implementation constraints.
Questions still open
What we do not know yet matters too.
These are not gaps we are trying to hide. They are questions we need to answer before the product should behave as if the answer is settled.
01What will you actually trust at 7:00 a.m. before cabinet?
We are studying what has to be visible immediately for a superintendent to decide that a pattern deserves attention, without requiring a long analytical workflow first.
02What should survive when leadership changes?
Which questions, decisions, assumptions, interventions, and later readings are useful enough for a new superintendent or district leader to inherit and trust?
03What makes a Board Brief credible?
Which evidence statements can survive a specific follow-up question, what qualification belongs beside the statement, and what should never be summarized away?
04How do you know whether AI made the work better?
We need more than adoption and performance. We are studying workload, judgment, reasoning, learning, capability, expectations, and the changes people experience around the technology.
05What can a second reading really tell you?
We want the later reading to help you learn what moved while keeping the explanation of why it moved separate until the evidence can support it.
From the Research Office
For the questions that do not fit neatly inside a dashboard.
Our writing goes deeper on the problems behind Signal: missing evidence, responsible AI, organizational change, and the difference between a model performing well and a system actually getting better.
Position paper
The information gap.
What happens when the question you need answered is not in the data the district already has?
→ Research essayWhat if the data you need doesn't exist?
Why better AI still cannot recover evidence the district never collected.
→ Research essayThe model worked. Did the system?
Why strong model performance does not automatically mean the surrounding work improved.
→Research × Product
The research should not disappear when the screen gets simpler.
Signal should make the evidence easier to use without stripping away the things that make the reading responsible.
Show the source
Keep the method close enough to matter.
Cycle, participation, coverage, scope, and reporting limits should not disappear when a number reaches your view.
Keep views distinct
Do not average away disagreement.
Self and other perspectives can tell you different things. Signal should let you see that difference instead of hiding it inside one score.
Protect access
Give people the view their role requires.
You see the system. District leaders, principals, and teachers see what is relevant to the work they own, with small groups protected.
Return later
Make the next reading comparable when it can be.
Keep the original question, baseline, intervention, and later evidence connected so a second reading can teach the district something.
Start with the question
What are you trying to understand before you act?
A board question. A district priority. A new technology rollout. A pattern you cannot explain yet. Bring us the question, and we can help think through what evidence would actually help.