Trainee Field Guide

Trainee Field Guide and Expectations

UICOMP Graduate Medical Education · Cohort beginning July 2027

1. What this is

You are not being asked to produce an invention. You are being asked to become someone who can find and solve real problems in healthcare — and to leave visible evidence that the capability now belongs to you.

The project is the vehicle. The change in you is the product.

Keith S. Hansen, MD, received a Stanford Biodesign Innovation Fellowship in 2021 and trained there in the needs-based process of identification, invention, and implementation. He and his team worked through 450 clinical needs before landing on the problem of intraoperative warm ischemia. What came of it was the kidney pod — a single-use device applied externally to a donor kidney just before sew-in — and the device company DIATIRO. But the part Hansen wants to teach is the method, not the device. He describes it as "less about jumping to solutions" and more about identifying the real clinical need and staying with what is actually wrong.

— Pullen B. Fostering and profiting from the physician innovator. American Journal of Transplantation. 2023;23:159–160.

Transformation is the goal. Commercialization is one possible route among many, and it is not the measure of success. A workflow redesign that removes ninety minutes of invisible nursing labor is a complete success. A device that never leaves the bench but taught you how to run a disconfirming experiment is a complete success. An idea you defended for eight months and then abandoned on evidence is a complete success.

The only failure in this track is an unlearned experiment. Data you collected and did not interpret. Feedback you received and did not let change you. A pivot you made for comfort rather than for evidence.

2. The progression you are working

Eight stages. A cycle, not a checklist. You will pass through it more than once, and each return should show better judgment than the last.

Mapped to the program's phases: PI-1 (typically PGY-2) lives mostly in SEE through CREATE and ends with an approved capstone plan. PI-2 (typically PGY-3) lives mostly in TEST through IMPLEMENT. Neither is clean. Expect to fall back to SEE from stage six. That is the design.

What constraint does for you

You will not have protected time, a budget, or a research team. Good. Scarcity is a magnifier — when not everything can be done, the consequential few become visible.

Three questions carry the weight:

• DISCERN — Of everything I can observe, what matters?
• PRIORITIZE — Of everything that matters, what warrants action?
• RESOURCE — Given what actually exists here, what can we responsibly do?

3. Where the work lives

Two surfaces. They do different jobs, and you should not blur them.

Notion — your thinking

Private to you and the faculty who review it. This is the messy, dated, chronological record.

Your workspace contains, at minimum:

• Field Log — dated observation entries. Raw. What you saw, heard, and felt in a clinical space. No solutions permitted in this database.
• Needs Inventory — a database of candidate needs, each one a row, each written as a need statement, each with a status (open / parked / retired / converged).
• Stakeholder Register — who you talked to, their role, what they told you, what surprised you, what it changed.
• Decision Log — every consequential choice. Date, decision, alternatives considered, evidence, what would make you reverse it.
• AI Ledger — see §6.
• Weekly Entry — one per week, twenty minutes minimum, written the same week. Not edited after the fact.

Observation does not keep business hours. If you are working nights or weekends, capture then — that is often when the most revealing things happen. Thin staffing, handoffs, improvised workarounds, the things nobody would do at ten in the morning with an audience. A two-line entry written at three in the morning is worth more than a polished paragraph reconstructed the following week.

Notion on your personal phone is the recommended way to do this. Open the field log, type what you just saw, close it. Clean it up later if you want, but do not wait to write it well. The same privacy rule applies wherever you are writing from: nothing identifiable, ever, on any device.

Field Log and Weekly Entries are append-only in spirit. If your thinking changes, write a new entry that says so. Do not go back and make yourself look smarter than you were. The false starts are the evidence.

GitHub — your artifacts and your public narrative

One repository per trainee, named pi-<lastname>-2027, with GitHub Pages enabled on /docs.

/README.md            One-paragraph statement of the need. Updated as it changes,
                      with prior versions preserved below a horizontal rule.
/docs/                GitHub Pages site — the public narrative of the journey
/needs/               Retired and converged need statements, with the reasoning
/observations/        De-identified synthesis (never raw field notes — those stay in Notion)
/prototypes/          Paper prototypes photographed, sketches, storyboards, mockups,
                      code, protocols. One folder per iteration, numbered.
/experiments/         One file per test: hypothesis, method, what would disconfirm,
                      result, interpretation, what changed.
/decisions/           Copies of the consequential entries from your Decision Log
/value-case/          See §9

Commit small and often. A repo with fourteen commits over a year tells me you did the work the night before review. A repo with two hundred small commits tells me where you actually were on the seventeenth of March.

Paper prototypes are expected. Photograph them, commit the photograph, and write two sentences about what the physical object made you notice that a description would not have.

Privacy — non-negotiable

Nothing that could identify a patient, a family member, or a named colleague goes into a public repository. Ever. Not in a commit message, not in an image, not in a filename, not in a photograph of a whiteboard, not in the EXIF data of a phone photo. Assume every commit is permanent and public, because it is — deleting a file does not remove it from history.

Raw field notes stay in Notion. Synthesis goes to the repo. If you are unsure, it stays in Notion and you ask.

Before any photography in a clinical space, you clear it. Before any structured data collection from patients or staff, we determine together whether it is QI or human subjects research, and if research, it goes to IRB before it starts, not after.

4. The cadence

The standing rhythm

Once the track opens in July 2027, this repeats until you graduate.

The continuity hour is one hour every two weeks, held inside your program's existing protected didactic time — for Internal Medicine, flexible within the 11:30–1:00 window. It alternates between two uses:

• Cohort hour (weeks 1, 5, 9…) — whole cohort plus faculty, sometimes an SME guest. Bring a live problem, not a progress report. This is where you offer your thinking for correction.
• Individual review (weeks 3, 7, 11…) — roughly twenty minutes each inside the same hour. You set the agenda and send it 24 hours ahead. If you arrive without one, we spend the twenty minutes on why.

No scheduled obligation falls on an evening or a weekend, and the concentrated work happens inside elective blocks rather than stacked on top of a full clinical schedule. Your own observation is a different matter — if you are on nights, write on nights. See §3.

The calendar

Focused-block dates are set with your program director well ahead of the academic year they fall in. Ask early; lead time is the only thing that makes these blocks good rather than improvised.

5. Standing expectations

These are starting numbers, calibrated for a resident with full clinical duties. They are dials, not laws — we will tune them together in your first month. But they are the default, and moving them requires a reason on the record.

July – September 2027 · the first 90 days (SEE and UNDERSTAND):

• 40 logged observations. No proposed solutions in any of them.
• 20 candidate needs written as need statements.
• 10 stakeholder conversations across at least 4 different role types — and physicians may be at most 3 of the 10. Talk to registrars, environmental services, schedulers, medical assistants, patients, billing, transport.
• Zero prototypes. This is deliberate. You are not permitted to build in the first 90 days.

October 2027 – June 2028 · rest of PI-1 (FRAME and CREATE):

• Converge to 3 needs, then to 1, with the reasoning written down for each retirement. A retired need is a documented asset, not a discard.
• The surviving need statement gets defended in front of faculty and revised at least twice.
• At least 5 distinct solution concepts before any one gets developed. Generating five is easy; generating five you would actually defend is the exercise.
• At least 3 paper prototypes, tested on a human being who is not in this program.

July 2028 – June 2029 · PI-2 (TEST through IMPLEMENT):

• Minimum four experiments, each with a pre-registered disconfirming condition written before the test. "What result would tell me I am wrong?" If you cannot answer that, it is not an experiment.
• One value case (§9).
• One documented pivot decision — persist, pivot, or stop — made explicitly and on evidence.

6. AI

You may use Claude or any other model. I want you to. But it occupies one seat and not the other.

AI is your partner-organizer. It is not your creative.

The distinction is not about tool use. It is about where perception, judgment, and decision live. Those must remain with you, and you must be able to demonstrate that they did.

The AI Ledger

Every substantive AI-assisted work session gets one row in Notion:

This is not surveillance and it does not count against you. An honest ledger showing heavy AI use and clear retained judgment scores higher than a thin ledger with a suspiciously polished artifact. The ledger is how you prove ownership. Use it.

7. Put the artifact away

Four times a year, at the progression gate, we close the repo, close Notion, and you reconstruct the journey out loud.

Five questions:

1. What did you personally experience?
2. What did you personally notice or understand because of that experience?
3. What did AI contribute?
4. What judgment or decision remained yours?
5. How are you different now because you went through this?

The fifth is the one that matters.

What an owned answer sounds like:

I thought this. Then I saw this. I talked to this person — that made me question my assumption. I considered these alternatives. I chose this because of these constraints. We tried it. This happened. That surprised me, so I changed this. I am still uncertain about that. Therefore, this is what I would do next.

Unpolished by nature. Surprise, reversal, and residual uncertainty are the markers of real passage. Fluent, confident, seamless narration with no changed minds in it is the thing we are watching for — if the reasoning disappears once the artifact is gone, the competence may be borrowed.

There is no penalty for a weak gate early. There is a real consequence for a gate where nothing has changed since the last one.

8. The six-month checkpoint

Twice a year, the gate carries a decision attached to it.

The December and June gates are checkpoint gates. Everything else about them is the same — artifacts away, five questions — but they end with a written determination about whether you continue in the track.

Three possible determinations:

• Continue. The default and the expected outcome.
• Continue with conditions. Two or three specific, written, dated conditions and a review point eight weeks out. Conditions are always about documented work or engagement — never about the project succeeding.
• Withdraw from the track. Rare, narrow, and described below.

What the checkpoint is not about

Not the quality of your prototype. Not whether the project is working. Not whether anyone will fund it, publish it, or adopt it. Not how polished the repo looks.

A failed project is never grounds for removal. Neither is a pivot, an abandoned need, a refused pilot, a negative result, or eight months of work you killed on evidence. Those are the outcomes the track is designed to produce. If you are worried that stopping a project will cost you your place here, you have the incentive exactly backwards — and you should say so out loud at the next continuity hour.

What it is about

Whether the learning is happening and whether the evidence of it exists.

• Are the field entries there, at roughly the agreed rhythm?
• Does the repo show work distributed across weeks rather than assembled before a review?
• Has your thinking moved across the eight stages, or has it stalled at the same place across two gates?
• Do you own your answers at the gate, or does the reasoning disappear when the artifacts do?
• Is the AI ledger honest? Heavy use with clear retained judgment scores well. A thin ledger under a polished artifact does not.
• Does feedback change anything? Receiving it graciously and acting on none of it is the pattern we are actually watching for.

Withdrawal from the track

Narrow. Four circumstances:

1. Sustained absence with no communication — after the stall protocol has been used and a direct conversation has been had.
2. Borrowed competence at two consecutive gates, after it has been named in writing at the first.
3. A breach of patient privacy, professionalism, or clinical safety.
4. Your clinical training needs the time. This one is not a judgment about you at all, and it is the most likely of the four.

How this is handled

• No surprises. Nothing appears at a checkpoint that has not already been raised with you in writing at least once. If a concern shows up cold, say so — that is a process failure and it counts in your favour.
• You can respond. The written determination reaches you within seven days. You may add a written response, which goes in the file alongside it.
• Your program director is informed of any determination other than "continue," so scheduling can be adjusted.
• This is not an adverse academic action. The track is elective and additive. Withdrawal returns your protected and elective time to your home program. It is not reportable, it does not follow you, and it is not a residency performance finding.
• You may leave at any time, for any reason, without explaining yourself. Tell us and it is done. There is no version of this where staying in the track under strain is the right answer.

9. The value case

Somewhere in PI-2, every project produces a one-page value case. Not a business plan. A stewardship argument.

• Who bears this burden today, and what does it cost them in time, money, or harm?
• What is the smallest version of this that could run on resources that already exist here?
• Who would have to say yes, and what would they need to see?
• What does it displace? What breaks if it works?
• What would tell you in 90 days that it is working — or that it is not?

The bias is toward bootstrapping: solutions that pay for themselves inside the system that hosts them, that survive on existing infrastructure, that do not require a funding event to become real. That constraint tends to produce better ideas, not smaller ones.

If a project genuinely warrants commercialization, we will pursue it and bring the institution's technology and IP offices in early. But you should not aim at it, and you will not be graded on it.

10. Working with each other

You will be two or three residents. I cannot require you to build one thing together, and I will not.

What I do require:

• You show your work to each other at cohort sessions, and you critique honestly.
• If you use another trainee's observation, need statement, or method, you credit it in the repo. Explicitly, by name.
• You are permitted, encouraged, and never obligated to merge projects. If you do, you write a short collaboration note in both repos stating who owns what and how you will handle divergence later.
• Silent competition is corrosive here. If you are holding something back from the cohort, say so out loud and say why. That is an acceptable position. Concealing it is not.

11. How you will be assessed

Not on the polish of your artifact. On demonstrated capability across the eight stages, at four levels: Emerging → Developing → Proficient → Advanced.

A rough calibration:

• Emerging — Names the obvious. Jumps to solutions. Treats feedback as an obstacle to route around.
• Developing — Observes without solving, but stays at the surface. Sees process problems as individual failures. Changes course when told to, not when evidence says to.
• Proficient — Connects individual observations to system design. Runs experiments that could actually fail. Pivots on evidence and can say what the evidence was.
• Advanced — Sees structural and compounded burdens others do not name. Actively hunts disconfirmation. Holds an idea loosely enough to kill it and firmly enough to defend it under pressure. Can teach the move to someone else.

You are expected to graduate at Proficient or better across most of the eight. You are not expected to be there in month three, and a project that stays at Developing while the learner moves to Proficient is a good outcome.

12. What you can expect from faculty

The same document governs both sides. Faculty are held to this, and you may hold us to it:

• We will not solve it for you. Faculty return questions, not answers, at roughly three questions to one assertion. If a faculty member hands you the solution, that is a failure of the program, and you should name it.
• We respond. Written feedback within seven days of your individual review. At least one substantive comment in your repo or Notion every two weeks, so you know the work is being read between meetings.
• We are explicit about our own uncertainty. When we do not know, we say so, and we tell you who might.
• We do not punish pivots. A documented, evidence-based change of direction raises your assessment. It does not lower it.
• We do not punish honest AI disclosure. The ledger is used to evaluate ownership, never to penalize use.
• We tell you where you actually stand. At every gate, in plain language, including when it is not what you hoped.
• No surprises at a checkpoint. Nothing that affects your continuation will be raised for the first time at a checkpoint gate. If it happens anyway, name it — that is our failure, not yours, and it counts in your favour.
• If you go quiet, we come find you. Two missed weekly entries triggers a direct conversation, not a demerit. Residency is hard and this track sits on top of it. Say so when it is too much; we will adjust the dials.

Appendix A — Weekly entry template

## Week of YYYY-MM-DD

**Where I was:** (setting, role, hours)

**What I observed:** (raw — no interpretation yet)

**What I noticed because of it:** (interpretation, separate from above)

**Who I talked to, and what surprised me:**

**What I now believe that I did not believe last week:**

**What I am uncertain about:**

**AI used this week:** (link to ledger rows)

**Next week's single most important question:**

Appendix B — Need statement format

A way to [do what] for [whom, specifically] that [achieves what measurable outcome] without [what constraint or trade-off].

No solution may appear in a need statement. If yours contains a technology, a device, an app, or a workflow, it is a solution statement wearing a disguise. Rewrite it.

Appendix C — Experiment record

# Experiment NN — [short name]

**Date:**
**Stage:** TEST
**Hypothesis:**
**What result would tell me I am wrong:**   ← write this before running it
**Method:**
**Participants / setting:**
**QI or research determination:**
**Result:**
**Interpretation:**
**What changed as a consequence:**
**Persist / Pivot / Stop, and why:**

Appendix D — Decision log entry

**Date:**
**Decision:**
**Alternatives I actually considered:**
**Evidence that drove it:**
**Who I consulted:**
**What would make me reverse this:**
**AI's role in this decision, if any:**