IM-4 — Mentor-Session Facilitation & Guest Lectures

ASSIP 2026 runs a layered-mentorship model — one faculty mentor (Prof. Gao) plus four PhD/RA sub-leaders running four pods of three or four students each, fourteen students in total. This section is about how the two mentoring voices the camp leans on — Prof. Gao’s Monday all-hands session and the sub-leaders’ daily drop-in office hours — actually get run, and how guest-lecture slots drop into the cohort sync without disrupting the daily rhythm. IM-1 (pacing) tells you where in the week each of these lands; this section tells you how to run them so they earn their place. Read it alongside IM-2 (rubrics), because every mentor session feeds the research-design and capstone rubrics, and IM-3 (pitfalls), because the mentor sessions are the single best opportunity to surface a week’s predictable pitfall before it migrates into the next chapter.

A standing principle for all three: an outside (or senior) voice is most valuable when it is concrete and slightly uncomfortable — a real result that did not replicate, a regulator’s question the data could not answer, a deadline that forced a shortcut. Curate for honesty over polish.

Part A — Prof. Gao’s Monday all-hands mentor session (9:00–10:00 ET, every Monday)

The defining synchronous fixture of the ASSIP week is the Monday morning all-hands: Prof. Gao walks the full 14-student cohort through one of his papers, framed as the week’s central identification or design lesson, in sixty minutes of Socratic discussion. The eight sessions are sequenced to mirror the curriculum (M1 = “What is a finding?” feeds Week 1; M4 = “Detecting discrimination with a clean design” feeds the Week-4 DiD work, etc.; see IM-1 §3 for the full sequence). This is the only block of the week when all 14 students are in the same room with the faculty mentor; everything about how it is run should respect that scarcity.

Each Monday session ships with the same five parts and they are designed to be run in order: a one-page pre-read packet circulated by Friday afternoon, three Socratic warm-up questions students answer on paper before the session, a five-slide speak-from deck, three stretch questions tied to that week’s Gao paper, and a post-session reflection prompt assigned in the last five minutes. The Week 4 packet (mentor4-detecting-discrimination.md) is the template to study before running any of them; it shows the intended register exactly. Below are the facilitation notes that turn those materials into a good sixty minutes for a 14-person remote room.

Use the pre-read packet as a gate, not a suggestion. The packet is one page and is meant to be read over the weekend, slowly, before the 9:00 am Monday session. The single highest-leverage move is enforcing that it was read — not by quizzing, but by opening the session with the warm-ups, which are unanswerable by a student who skipped them. Have the four pod sub-leaders collect their pod’s written warm-up answers by 8:45 am Monday in a shared doc (one row per student); Prof. Gao reads them in the five minutes before the session opens, and now knows where every pod is before saying a word. Tell students explicitly, in Week 1, that the packets are short because they are dense and that skimming them defeats the design.

Run the warm-ups Socratically — they are framed to have no clean answer. The three warm-ups are written, per the Week 4 example, as questions with “honest answers and dishonest ones, not right and wrong ones.” Resist the instinct to confirm a correct answer and move on. The move is to take a student’s sentence and press it: what would have to be true for that to fail? When a student gives the textbook answer (“control for creditworthiness, read the residual”), Prof. Gao’s job is to be the unobserved confounder — to name the soft variable the underwriter saw that the regression did not. Aim to spend ten to twelve minutes here and to have surfaced the session’s central tension before any slide appears. The deck then resolves a tension the warm-ups opened, which is the right pedagogical order.

Time the five-slide deck to leave room for the stretch questions. The deck is a speak-from scaffold, not a read-aloud — each slide is bullets Prof. Gao talks around. Budget roughly four minutes a slide, twenty minutes total, and protect that ceiling ruthlessly. If the deck runs long, cut from the deck, never from the stretch questions or the discussion. A practical rhythm for the sixty minutes: pre-read assumed done (0 min), warm-ups and discussion (12 min), deck (20 min), stretch questions worked in pods then surfaced to the room (22 min), reflection prompt assigned (6 min). The shift from “pairs” to “pods” for the stretch-question work is the load-bearing adaptation for a 14-student cohort — pods are already formed, they already trust each other from the daily standup, and four pod-reports to the whole room is more legible than seven pair-reports.

Work the stretch questions in pods, then surface to the room. The three stretch questions take the week’s identification ideas and point them at one of Prof. Gao’s papers, deliberately reason-about-method-only — the Week 4 set says plainly, “Do not invent or recite any specific reported magnitudes from the paper.” Honor that instruction when facilitating; if a student reaches for a number, redirect to the judgment (“never mind what they found — what comparison makes the finding believable?”). Have pods work the questions in their pod breakout for six or seven minutes first, with the sub-leader holding the room. Then each pod’s sub-leader nominates one student to report that pod’s best answer to the full cohort. Pod-work-then-report is the mechanism that gets quiet students talking — a student who will not address fourteen people will tell their pod what they think, and the sub-leader who knows their pod can route the report to the student who needs the practice.

Draw out quiet students by structure, not by cold-calling. Cold-calling a quiet student in a 14-person Zoom room on a hard identification question usually produces silence and embarrassment, not learning. Three structural moves work better and are designed for the remote, layered format. First, pod-and-report (above): the sub-leader has the relationship and routes the question. Second, the written-first habit — because the warm-ups and stretch questions ask for written sentences, every student has something on paper to read from, so “read me what you wrote for the second warm-up” is always a fair ask. Third, route by interest: the recurring student cast (Maya–fair lending, Devon–crypto, Priya–climate/insurance, Sam–markets, Leah–patents/text) maps onto the sessions, and a quiet student is far more likely to speak when the question is steered toward their hook. The Week 4 fair-lending session is Maya’s; the Week 6 text session is Leah’s; use that. Keep a private tally of who has spoken each week (the four sub-leaders can each tally their pod) and close the gap deliberately over the eight weeks. By Week 6 every student in every pod has been nominated to report at least twice.

Connect every Monday session explicitly to the capstone. This is the thread that makes the eight sessions a sequence rather than eight disconnected talks. Every session should end with one sentence naming what this week’s identification lesson will demand of the student’s own capstone. The materials already do this — the Week 4 deck’s final slide says outright, “Your Lab 4 and Capstone 1 are exactly this.” Say it out loud every week: Week 2’s SE discipline is the wild-cluster-bootstrap row of the capstone rubric; Week 3’s selection-on-observables failure is why the capstone needs a design, not a longer control list; Week 5’s “anatomy of a paper” is the structure of the JSSR abstract they will write; Week 7’s “how I pick a project — and kill one” is the literal task of their Week 7 deliverable. The reflection prompt is the hand-off: it asks students to take an outside claim and run it through the week’s discipline, which is rehearsal for running it on their own result. Read a few reflections aloud at the start of the following Monday — the materials invite this — so students see that the prompt was real work, not a formality, and so the sessions chain.

Recording, captioning, and the equity guarantee. Every Monday all-hands is recorded with captions and posted to the cohort wiki by Monday noon. A student who missed the live session for an accommodation, a bandwidth failure, or a documented health reason watches the recording before Tuesday’s curriculum block and posts their warm-up answers to the wiki — the asynchronous participation path. The recording is not a substitute for synchronous attendance (see IM-3 ASSIP-R.5); it is the documented exception for the genuine case.

Part B — Sub-leaders and the daily drop-in office hours (1:00–4:00 ET, every weekday)

Prof. Gao does not run daily mentoring of 14 students remotely; the four pod sub-leaders do. Each weekday afternoon during the mentor research block (1:00–4:00 ET), each sub-leader holds an open Zoom room dedicated to their pod’s three or four students. Students drop in for ten or twenty minutes when they hit a blocker — a stata error, a WRDS query that times out, an identification memo that they cannot get to commit to a single primary spec. The afternoon is not a lecture; it is the working-research-group hour the textbook’s chapters do not provide.

The sub-leader’s three duties. Each sub-leader runs three concrete things:

The 4:00 pm pod standup (ten minutes, every weekday). Each pod member reports what they built today, what is blocked, what they will do tomorrow. The sub-leader summarizes in two sentences to Prof. Gao by 4:30 in a shared doc — pod by pod, the cohort’s daily state of play. This is the surveillance mechanism that catches ASSIP-R.1 (the silent dropout, IM-3) early.
Drop-in office hours in the afternoon (three hours, every weekday). Zoom room open from 1:00 to 4:00. Students arrive when they need help; the sub-leader is on camera, working on their own code or reading, available the moment a student arrives. No appointment system — appointments create friction and friction makes students not ask for help, which is the exact failure mode of ASSIP-R.1.
Daily problem-set grading for their pod. Each sub-leader grades their pod’s daily problem set against IM-2 §2 and posts feedback to the wiki by the next morning’s 9:00 kickoff, so the day’s chapter lands on students aware of yesterday’s highest-signal pitfall. The four sub-leaders norm their grading on Sunday afternoon (IM-5 Part E) so problem-set scores are consistent across pods.

The weekly one-on-one (mandatory, ten minutes, scheduled). Once every five weekdays, each sub-leader holds a private ten-minute one-on-one with every member of their pod. Camera-on if the student’s accommodations allow; phone if not. The conversation is open-ended: “how is the week going, what is hardest, what do you need.” This is the single highest-signal intervention against the silent-dropout pitfall (IM-3 ASSIP-R.1). Sub-leaders log a one-sentence summary to a private shared doc (Prof. Gao + sub-leaders only) immediately after each one-on-one. A student whose summary two weeks in a row reads “seems disengaged” is escalated to Prof. Gao the same day.

The escalation protocol. A sub-leader escalates to Prof. Gao on the first of any of these: missed problem set; two consecutive standup absences; a wiki page that has not been touched in a week; a one-on-one that surfaces personal crisis or harassment; an academic-integrity concern. Early escalation is cheap; late escalation is sometimes too late. The cohort’s culture is set in Week 1 by how the first escalation is handled.

Pod composition and reshuffle. Pods are formed by Prof. Gao before orientation, balanced on (a) declared topic interest from the application essay (fair-lending, markets, fintech, macro), (b) prior coding background, and (c) demographic diversity. The pods reshuffle exactly once, at the start of Week 4. The reshuffle is not punitive — it breaks early cliques, exposes students to fresh mentorship, and balances the second-half cohort around their now-clearer capstone topics. The post-reshuffle pod composition is the one that runs through the Symposium; do not reshuffle again during Weeks 4–8, because the cost of relationship rebuilding mid-capstone outweighs any benefit.

A note on register. Prof. Gao’s Monday materials are written in the first person and trade on the authority of someone who took a question “all the way from a dataset to Congressional testimony.” Sub-leaders do not impersonate that voice in their daily office hours. The sub-leader is a senior peer — a PhD student or RA two or three years ahead — and their authority comes from having recently been in the student’s seat, not from war stories. Present a paper as a worked case from the literature, keep the Socratic structure exactly, and be honest that you are relaying the design lesson rather than the war story. The structure carries the learning; the personal stakes are a bonus, not the load-bearing element.

Part C — Guest lectures and how to slot them

You do not need a guest every week, and you should not try; four guest lectures across the eight weeks is plenty, and each should land where the chapters have just made the visitor’s world legible. Guest lectures drop into the Wednesday 4:00–5:00 cohort-sync slot — sub-leaders pause the standup that day; the guest takes the full hour. The four profiles below are chosen so that, together, they cover the arc from regulation to practice to product to public communication — the full life of an empirical finding once it leaves the notebook. Treat the week assignments as defaults; the binding constraint is always “what have the students just learned enough to interrogate this person about?”

A fair-lending regulator or SEC economist — slot Wednesday of Week 4. This is the most important guest of the four because it lands the week the students study identification for discrimination detection, the same week Mentor Session 4 anchors on Sun & Gao (2019) and ch4 builds DiD on HMDA. A practicing economist from a fair-lending unit, the CFPB, HUD, or an SEC division can do something a chapter cannot: explain what it means for a coefficient to become evidence in a proceeding — the documentation a regulator demands, the adversarial scrutiny a number survives in litigation, why “residual disparity after controls” is necessary but not sufficient in a courtroom exactly as the Monday mentor pre-read argues it is not sufficient in a journal. Brief the guest to bring one real case where the design, not the size of the gap, was what decided whether the agency could act. This guest also de-abstracts the entire fair-lending thread that runs Weeks 4 through 6.

A quant practitioner — slot Wednesday of Week 2 or Week 5. Someone from a hedge fund, an asset manager, or a bank’s modeling desk grounds the standard-error and replication material in money. Week 2 (the OLS engine, the SE flavors) is the natural home if you want them to drive home Mentor Session 2’s thesis that “your standard error is the whole ballgame” — a practitioner who has watched a strategy die because its backtest used the wrong inference is the most persuasive possible witness for clustered and HAC errors. Week 5 (reading the classics — Fama–French, Jegadeesh–Titman momentum) works if you want them to talk about how a published anomaly does or does not survive contact with transaction costs and capacity. Either way, ask them to bring one failed idea: students hear plenty about successes, and the failure is where the inference lesson lives.

A FinTech founder — slot Wednesday of Week 6. Week 6 is text-as-data and the AI co-pilot module, and it is where the camp’s subject most clearly touches the world the students are about to enter. A founder of a lending, payments, or alt-data startup can speak to algorithmic underwriting from the inside — which ties straight to the Bartlett et al. (2022) fair-lending-in-the-FinTech-era reader’s guide and to ch6.5’s warnings about look-ahead leakage and validating model outputs. Brief them to be candid about the tension between moving fast and the disparate-impact exposure the students just learned to measure; that tension is the lecture.

A data journalist — slot Wednesday of Week 7 or Week 8. As students turn their own analysis into a Symposium poster, a JSSR abstract, and an 8-minute talk, a journalist who covers finance or does data-driven investigations teaches the skill the chapters undersell: communicating a quantitative finding to someone who will not read the appendix. They can critique a student’s headline sentence, show how a chart either clarifies or misleads, and explain how they themselves decide a result is solid enough to publish — a real-world echo of the capstone’s “would this survive a hostile referee” standard, extended to “and a skeptical editor and a Mason News press reporter at the Symposium.” This guest pairs naturally with ch8.3 (writing) and ch8.5 (the talk).

Logistics that make a guest worth the disruption. Send the guest a one-paragraph brief naming the exact chapters the students have finished and the two or three things you want them to not re-teach (the students already have the mechanics; you want the war stories). Ask for forty to forty-five minutes of talk and a hard fifteen of questions, and seed the Q&A: assign each student to arrive with one written question tied to a specific chapter, collected to the wiki beforehand so a quiet Zoom never stalls. Record with permission and post to the wiki by end-of-day; the recording is the equity guarantee for any student who missed live. If a profile is unavailable, a recorded talk or a published interview plus a live student-led discussion is a fully acceptable substitute — the goal is the outside perspective, not the logistics of a calendar.

Protect the chapter framing. The one failure mode to guard against is the guest who, in good faith, contradicts something the chapters established — tells the room that “stars are all that matter” the week after Mentor Session 2 argued the standard error is the whole ballgame, or waves away identification with “we just control for everything.” This is the confusion IM-3 catalogs, and the fix is a thirty-second framing in your introduction and a one-sentence debrief at the next Monday’s all-hands: name explicitly where the practitioner’s shortcut is a reasonable working compromise under a deadline versus where it is the exact mistake the camp is teaching students to avoid. A guest who disagrees with a chapter is a teaching opportunity, not a problem, provided you surface the disagreement rather than leaving students to silently resolve it the wrong way.

Tie each guest forward. Like the Monday mentor sessions, a guest lecture should close with one sentence connecting the visit to the students’ own work: the regulator’s documentation standard is the threats-table discipline of their capstone; the quant’s failed backtest is why their robustness section must include the check that failed; the founder’s disparate-impact exposure is what the fair-lending capstone topic measures; the journalist’s headline test is the 8-minute Symposium talk. The guest gives the war story; you draw the line back to the deliverable.

Part D — How the layers compose across a typical week

A single representative week (W3, the Causal-I week) illustrates how the three voices interact:

Monday 9–10: Prof. Gao all-hands, “Natural experiments” (Deng, Gao & Kim 2020). Warm-ups collected by the four sub-leaders at 8:45; reflection assigned at 9:55.
Mon–Fri 1:00–4:00: Sub-leaders run pod drop-ins in their Zoom rooms. Pod 1 has two students working on a Bartik shift-share replication; Pod 2 a fair-lending IV memo; etc. Prof. Gao floats through one pod’s afternoon by rotation (this week: Pod 3 on Wednesday).
Wednesday 4:00–5:00: No guest this week. Cohort sync runs as the rotating skill-clinic slot — the four sub-leaders co-teach a thirty-minute clinic on the first-stage F (the W3 high-signal topic) followed by a thirty-minute pair-debug round.
Friday 4:30–4:50: Show-and-tell. One randomly drawn student demos their W3 weak-IV / Anderson–Rubin coverage task. Two students from other pods ask questions.
Monday following 9–9:05: Prof. Gao reads aloud one student’s W3 reflection prompt — the hand-off into W4.

The three voices — Prof. Gao on Mondays, sub-leaders daily, guests on selected Wednesdays — do not compete; they layer. Prof. Gao sets the week’s intellectual frame; sub-leaders carry it into the daily work; guests, when they appear, anchor it in a world outside the cohort. The discipline that keeps the layering clean is that every voice ties back to the same place: the threats table, the PAP, the rebuild, the eight-minute talk. That is the camp’s gravitational center, and every layer of mentorship in IM-4 is doing one job: pulling each student’s work toward it.

# IM-4 — Mentor-Session Facilitation & Guest Lectures ASSIP 2026 runs a **layered-mentorship** model — one faculty mentor (Prof. Gao) plus four PhD/RA sub-leaders running four pods of three or four students each, fourteen students in total. This section is about how the two mentoring voices the camp leans on — **Prof. Gao's Monday all-hands session** and **the sub-leaders' daily drop-in office hours** — actually get run, and how **guest-lecture slots** drop into the cohort sync without disrupting the daily rhythm. IM-1 (pacing) tells you *where* in the week each of these lands; this section tells you how to run them so they earn their place. Read it alongside IM-2 (rubrics), because every mentor session feeds the research-design and capstone rubrics, and IM-3 (pitfalls), because the mentor sessions are the single best opportunity to surface a week's predictable pitfall before it migrates into the next chapter. A standing principle for all three: an outside (or senior) voice is most valuable when it is *concrete and slightly uncomfortable* — a real result that did not replicate, a regulator's question the data could not answer, a deadline that forced a shortcut. Curate for honesty over polish. --- ## Part A — Prof. Gao's Monday all-hands mentor session (9:00–10:00 ET, every Monday) The defining synchronous fixture of the ASSIP week is the **Monday morning all-hands**: Prof. Gao walks the full 14-student cohort through one of his papers, framed as the week's central identification or design lesson, in sixty minutes of Socratic discussion. The eight sessions are sequenced to mirror the curriculum (M1 = "What is a finding?" feeds Week 1; M4 = "Detecting discrimination with a clean design" feeds the Week-4 DiD work, etc.; see IM-1 §3 for the full sequence). This is the only block of the week when all 14 students are in the same room with the faculty mentor; everything about how it is run should respect that scarcity. Each Monday session ships with the same five parts and they are designed to be run in order: a **one-page pre-read packet** circulated by Friday afternoon, **three Socratic warm-up questions** students answer on paper before the session, a **five-slide speak-from deck**, **three stretch questions** tied to that week's Gao paper, and a **post-session reflection prompt** assigned in the last five minutes. The Week 4 packet (`mentor4-detecting-discrimination.md`) is the template to study before running any of them; it shows the intended register exactly. Below are the facilitation notes that turn those materials into a good sixty minutes for a 14-person remote room. **Use the pre-read packet as a gate, not a suggestion.** The packet is one page and is meant to be read *over the weekend*, slowly, before the 9:00 am Monday session. The single highest-leverage move is enforcing that it was read — not by quizzing, but by opening the session with the warm-ups, which are unanswerable by a student who skipped them. Have the four pod sub-leaders collect their pod's written warm-up answers by 8:45 am Monday in a shared doc (one row per student); Prof. Gao reads them in the five minutes before the session opens, and now knows where every pod is before saying a word. Tell students explicitly, in Week 1, that the packets are short *because* they are dense and that skimming them defeats the design. **Run the warm-ups Socratically — they are framed to have no clean answer.** The three warm-ups are written, per the Week 4 example, as questions with "honest answers and dishonest ones, not right and wrong ones." Resist the instinct to confirm a correct answer and move on. The move is to take a student's sentence and press it: *what would have to be true for that to fail?* When a student gives the textbook answer ("control for creditworthiness, read the residual"), Prof. Gao's job is to be the unobserved confounder — to name the soft variable the underwriter saw that the regression did not. Aim to spend ten to twelve minutes here and to have surfaced the session's central tension before any slide appears. The deck then *resolves* a tension the warm-ups *opened*, which is the right pedagogical order. **Time the five-slide deck to leave room for the stretch questions.** The deck is a speak-from scaffold, not a read-aloud — each slide is bullets Prof. Gao talks *around*. Budget roughly four minutes a slide, twenty minutes total, and protect that ceiling ruthlessly. If the deck runs long, cut from the deck, never from the stretch questions or the discussion. A practical rhythm for the sixty minutes: pre-read assumed done (0 min), warm-ups and discussion (12 min), deck (20 min), stretch questions worked in **pods** then surfaced to the room (22 min), reflection prompt assigned (6 min). The shift from "pairs" to "pods" for the stretch-question work is the load-bearing adaptation for a 14-student cohort — pods are already formed, they already trust each other from the daily standup, and four pod-reports to the whole room is more legible than seven pair-reports. **Work the stretch questions in pods, then surface to the room.** The three stretch questions take the week's identification ideas and point them at one of Prof. Gao's papers, deliberately *reason-about-method-only* — the Week 4 set says plainly, "Do not invent or recite any specific reported magnitudes from the paper." Honor that instruction when facilitating; if a student reaches for a number, redirect to the judgment ("never mind what they found — what comparison makes the finding believable?"). Have pods work the questions in their pod breakout for six or seven minutes first, with the sub-leader holding the room. Then each pod's sub-leader nominates one student to report that pod's best answer to the full cohort. Pod-work-then-report is the mechanism that gets quiet students talking — a student who will not address fourteen people will tell their pod what they think, and the sub-leader who knows their pod can route the report to the student who needs the practice. **Draw out quiet students by structure, not by cold-calling.** Cold-calling a quiet student in a 14-person Zoom room on a hard identification question usually produces silence and embarrassment, not learning. Three structural moves work better and are designed for the remote, layered format. First, **pod-and-report** (above): the sub-leader has the relationship and routes the question. Second, the **written-first** habit — because the warm-ups and stretch questions ask for written sentences, every student has something on paper to read from, so "read me what you wrote for the second warm-up" is always a fair ask. Third, **route by interest**: the recurring student cast (Maya–fair lending, Devon–crypto, Priya–climate/insurance, Sam–markets, Leah–patents/text) maps onto the sessions, and a quiet student is far more likely to speak when the question is steered toward their hook. The Week 4 fair-lending session is Maya's; the Week 6 text session is Leah's; use that. Keep a private tally of who has spoken each week (the four sub-leaders can each tally their pod) and close the gap deliberately over the eight weeks. By Week 6 every student in every pod has been nominated to report at least twice. **Connect every Monday session explicitly to the capstone.** This is the thread that makes the eight sessions a sequence rather than eight disconnected talks. Every session should end with one sentence naming what this week's identification lesson will demand of the student's own capstone. The materials already do this — the Week 4 deck's final slide says outright, "Your Lab 4 and Capstone 1 are exactly this." Say it out loud every week: Week 2's SE discipline is the wild-cluster-bootstrap row of the capstone rubric; Week 3's selection-on-observables failure is why the capstone needs a *design*, not a longer control list; Week 5's "anatomy of a paper" is the structure of the JSSR abstract they will write; Week 7's "how I pick a project — and kill one" is the literal task of their Week 7 deliverable. The reflection prompt is the hand-off: it asks students to take an outside claim and run it through the week's discipline, which is rehearsal for running it on their own result. Read a few reflections aloud at the start of the *following* Monday — the materials invite this — so students see that the prompt was real work, not a formality, and so the sessions chain. **Recording, captioning, and the equity guarantee.** Every Monday all-hands is recorded with captions and posted to the cohort wiki by Monday noon. A student who missed the live session for an accommodation, a bandwidth failure, or a documented health reason watches the recording before Tuesday's curriculum block and posts their warm-up answers to the wiki — the asynchronous participation path. The recording is not a substitute for synchronous attendance (see IM-3 ASSIP-R.5); it is the documented exception for the genuine case. --- ## Part B — Sub-leaders and the daily drop-in office hours (1:00–4:00 ET, every weekday) Prof. Gao does not run daily mentoring of 14 students remotely; the four pod sub-leaders do. Each weekday afternoon during the **mentor research block (1:00–4:00 ET)**, each sub-leader holds an open Zoom room dedicated to their pod's three or four students. Students drop in for ten or twenty minutes when they hit a blocker — a stata error, a WRDS query that times out, an identification memo that they cannot get to commit to a single primary spec. The afternoon is not a lecture; it is the working-research-group hour the textbook's chapters do not provide. **The sub-leader's three duties.** Each sub-leader runs three concrete things: 1. **The 4:00 pm pod standup (ten minutes, every weekday).** Each pod member reports what they built today, what is blocked, what they will do tomorrow. The sub-leader summarizes in two sentences to Prof. Gao by 4:30 in a shared doc — pod by pod, the cohort's daily state of play. This is the surveillance mechanism that catches ASSIP-R.1 (the silent dropout, IM-3) early. 2. **Drop-in office hours in the afternoon (three hours, every weekday).** Zoom room open from 1:00 to 4:00. Students arrive when they need help; the sub-leader is on camera, working on their own code or reading, available the moment a student arrives. No appointment system — appointments create friction and friction makes students not ask for help, which is the exact failure mode of ASSIP-R.1. 3. **Daily problem-set grading for their pod.** Each sub-leader grades their pod's daily problem set against IM-2 §2 and posts feedback to the wiki by the next morning's 9:00 kickoff, so the day's chapter lands on students aware of yesterday's highest-signal pitfall. The four sub-leaders norm their grading on Sunday afternoon (IM-5 Part E) so problem-set scores are consistent across pods. **The weekly one-on-one (mandatory, ten minutes, scheduled).** Once every five weekdays, each sub-leader holds a private ten-minute one-on-one with every member of their pod. Camera-on if the student's accommodations allow; phone if not. The conversation is open-ended: "how is the week going, what is hardest, what do you need." This is the single highest-signal intervention against the silent-dropout pitfall (IM-3 ASSIP-R.1). Sub-leaders log a one-sentence summary to a private shared doc (Prof. Gao + sub-leaders only) immediately after each one-on-one. A student whose summary two weeks in a row reads "seems disengaged" is escalated to Prof. Gao the same day. **The escalation protocol.** A sub-leader escalates to Prof. Gao on the *first* of any of these: missed problem set; two consecutive standup absences; a wiki page that has not been touched in a week; a one-on-one that surfaces personal crisis or harassment; an academic-integrity concern. Early escalation is cheap; late escalation is sometimes too late. The cohort's culture is set in Week 1 by how the first escalation is handled. **Pod composition and reshuffle.** Pods are formed by Prof. Gao before orientation, balanced on (a) declared topic interest from the application essay (fair-lending, markets, fintech, macro), (b) prior coding background, and (c) demographic diversity. **The pods reshuffle exactly once, at the start of Week 4.** The reshuffle is not punitive — it breaks early cliques, exposes students to fresh mentorship, and balances the second-half cohort around their now-clearer capstone topics. The post-reshuffle pod composition is the one that runs through the Symposium; do not reshuffle again during Weeks 4–8, because the cost of relationship rebuilding mid-capstone outweighs any benefit. **A note on register.** Prof. Gao's Monday materials are written in the first person and trade on the authority of someone who took a question "all the way from a dataset to Congressional testimony." Sub-leaders do **not** impersonate that voice in their daily office hours. The sub-leader is a senior peer — a PhD student or RA two or three years ahead — and their authority comes from having recently been in the student's seat, not from war stories. Present a paper as a worked case from the literature, keep the Socratic structure exactly, and be honest that you are relaying the design lesson rather than the war story. The structure carries the learning; the personal stakes are a bonus, not the load-bearing element. --- ## Part C — Guest lectures and how to slot them You do not need a guest every week, and you should not try; **four guest lectures across the eight weeks** is plenty, and each should land where the chapters have just made the visitor's world legible. Guest lectures drop into the **Wednesday 4:00–5:00 cohort-sync slot** — sub-leaders pause the standup that day; the guest takes the full hour. The four profiles below are chosen so that, together, they cover the arc from regulation to practice to product to public communication — the full life of an empirical finding once it leaves the notebook. Treat the week assignments as defaults; the binding constraint is always "what have the students just learned enough to *interrogate* this person about?" **A fair-lending regulator or SEC economist — slot Wednesday of Week 4.** This is the most important guest of the four because it lands the week the students study identification for discrimination detection, the same week Mentor Session 4 anchors on Sun & Gao (2019) and ch4 builds DiD on HMDA. A practicing economist from a fair-lending unit, the CFPB, HUD, or an SEC division can do something a chapter cannot: explain what it means for a coefficient to become *evidence in a proceeding* — the documentation a regulator demands, the adversarial scrutiny a number survives in litigation, why "residual disparity after controls" is necessary but not sufficient in a courtroom exactly as the Monday mentor pre-read argues it is not sufficient in a journal. Brief the guest to bring one real case where the *design*, not the size of the gap, was what decided whether the agency could act. This guest also de-abstracts the entire fair-lending thread that runs Weeks 4 through 6. **A quant practitioner — slot Wednesday of Week 2 or Week 5.** Someone from a hedge fund, an asset manager, or a bank's modeling desk grounds the standard-error and replication material in money. Week 2 (the OLS engine, the SE flavors) is the natural home if you want them to drive home Mentor Session 2's thesis that "your standard error is the whole ballgame" — a practitioner who has watched a strategy die because its backtest used the wrong inference is the most persuasive possible witness for clustered and HAC errors. Week 5 (reading the classics — Fama–French, Jegadeesh–Titman momentum) works if you want them to talk about how a published anomaly does or does not survive contact with transaction costs and capacity. Either way, ask them to bring one *failed* idea: students hear plenty about successes, and the failure is where the inference lesson lives. **A FinTech founder — slot Wednesday of Week 6.** Week 6 is text-as-data and the AI co-pilot module, and it is where the camp's subject most clearly touches the world the students are about to enter. A founder of a lending, payments, or alt-data startup can speak to algorithmic underwriting from the inside — which ties straight to the Bartlett et al. (2022) fair-lending-in-the-FinTech-era reader's guide and to ch6.5's warnings about look-ahead leakage and validating model outputs. Brief them to be candid about the tension between moving fast and the disparate-impact exposure the students just learned to measure; that tension *is* the lecture. **A data journalist — slot Wednesday of Week 7 or Week 8.** As students turn their own analysis into a Symposium poster, a JSSR abstract, and an 8-minute talk, a journalist who covers finance or does data-driven investigations teaches the skill the chapters undersell: communicating a quantitative finding to someone who will not read the appendix. They can critique a student's headline sentence, show how a chart either clarifies or misleads, and explain how they themselves decide a result is solid enough to publish — a real-world echo of the capstone's "would this survive a hostile referee" standard, extended to "and a skeptical editor *and* a Mason News press reporter at the Symposium." This guest pairs naturally with ch8.3 (writing) and ch8.5 (the talk). **Logistics that make a guest worth the disruption.** Send the guest a one-paragraph brief naming the exact chapters the students have finished and the two or three things you want them to *not* re-teach (the students already have the mechanics; you want the war stories). Ask for forty to forty-five minutes of talk and a hard fifteen of questions, and seed the Q&A: assign each student to arrive with one written question tied to a specific chapter, collected to the wiki beforehand so a quiet Zoom never stalls. Record with permission and post to the wiki by end-of-day; the recording is the equity guarantee for any student who missed live. If a profile is unavailable, a recorded talk or a published interview plus a live student-led discussion is a fully acceptable substitute — the goal is the outside perspective, not the logistics of a calendar. **Protect the chapter framing.** The one failure mode to guard against is the guest who, in good faith, contradicts something the chapters established — tells the room that "stars are all that matter" the week after Mentor Session 2 argued the standard error is the whole ballgame, or waves away identification with "we just control for everything." This is the confusion IM-3 catalogs, and the fix is a thirty-second framing in your introduction and a one-sentence debrief at the next Monday's all-hands: name explicitly where the practitioner's shortcut is a *reasonable working compromise under a deadline* versus where it is the exact mistake the camp is teaching students to avoid. A guest who disagrees with a chapter is a teaching opportunity, not a problem, *provided you surface the disagreement* rather than leaving students to silently resolve it the wrong way. **Tie each guest forward.** Like the Monday mentor sessions, a guest lecture should close with one sentence connecting the visit to the students' own work: the regulator's documentation standard is the threats-table discipline of their capstone; the quant's failed backtest is why their robustness section must include the check that *failed*; the founder's disparate-impact exposure is what the fair-lending capstone topic measures; the journalist's headline test is the 8-minute Symposium talk. The guest gives the war story; you draw the line back to the deliverable. --- ## Part D — How the layers compose across a typical week A single representative week (W3, the Causal-I week) illustrates how the three voices interact: - **Monday 9–10:** Prof. Gao all-hands, *"Natural experiments"* (Deng, Gao & Kim 2020). Warm-ups collected by the four sub-leaders at 8:45; reflection assigned at 9:55. - **Mon–Fri 1:00–4:00:** Sub-leaders run pod drop-ins in their Zoom rooms. Pod 1 has two students working on a Bartik shift-share replication; Pod 2 a fair-lending IV memo; etc. Prof. Gao floats through one pod's afternoon by rotation (this week: Pod 3 on Wednesday). - **Wednesday 4:00–5:00:** No guest this week. Cohort sync runs as the rotating skill-clinic slot — the four sub-leaders co-teach a thirty-minute clinic on the first-stage F (the W3 high-signal topic) followed by a thirty-minute pair-debug round. - **Friday 4:30–4:50:** Show-and-tell. One randomly drawn student demos their W3 weak-IV / Anderson–Rubin coverage task. Two students from other pods ask questions. - **Monday following 9–9:05:** Prof. Gao reads aloud one student's W3 reflection prompt — the hand-off into W4. The three voices — Prof. Gao on Mondays, sub-leaders daily, guests on selected Wednesdays — do not compete; they layer. Prof. Gao sets the week's intellectual frame; sub-leaders carry it into the daily work; guests, when they appear, anchor it in a world outside the cohort. The discipline that keeps the layering clean is that **every voice ties back to the same place: the threats table, the PAP, the rebuild, the eight-minute talk**. That is the camp's gravitational center, and every layer of mentorship in IM-4 is doing one job: pulling each student's work toward it.