Historical version · before 11 September 2026 This is the former course guide, retained for transparency. Its whole-lecture teaching rules have been superseded. Read the current course guide for the rules that apply now. The original text below has not been rewritten.

How this course works

EST 5053 · Control and Computation in Biological Systems · Fall 2026. The rules, the topics, the format, the rubric, and the four things you hand in. Westlake University · Tue & Thu 10:40–12:15 · E10-305, Yungu campus · lecturer Fangzhou Xiao · TAs Wenqin Zhou and Xinyu Wang · 16 lectures, 2026-09-01 to 2026-10-29

The whole course in five lines

I teach lectures 1 and 2. You teach lectures 3 to 16. One to three of you may own one entire 95-minute lecture. Every student teaches at least once; volunteers may teach additional lectures for extra credit.
The last 5 minutes of every lecture are mine, for the gap, the take-away, and what the field says next.
I hand you a core page per lecture, one week ahead. That is the floor, not the ceiling. Going beyond it is the assignment.
Quality is guaranteed by design. If a lecture does not hold up, I step in and teach it. That is the fallback, and it is built into the grading.
Four things are graded: your lecture (40), your extension mini-essays (30), your research essay (20), your participation (10).

Part 1

The course

What it is about, the stance on working with AI that reshaped it, and the sixteen topics.

1What this course is about

Three schools ask three questions about life. This course teaches you to answer all three quantitatively, and then makes one claim of its own. Biological organisms do extraordinary things. They transform matter through thousands of metabolic steps. They hold themselves steady while the world around them shifts. They balance persistence in famine against all-out growth in plenty. They reshape their surroundings to their own advantage. Behaviors like these involve enormous numbers of interacting components, and reasoning about them demands the same rigor we bring to a complicated engineered machine. Three schools of thought have taken up that reasoning, and this course teaches the tools of all three.

Table 1. The three schools, and where each one lives in this course.
School	Its question	Its tools here	Lectures
Physics	What is life, as an object?	Order-of-magnitude estimates, scaling laws, statistical mechanics, energy and information	1, 2, 7
System	How does life work, as a machine?	Reaction networks, dynamics, stochastic processes, feedback and control, computation	3–12
Industry	How could life be useful, as a tool?	Metabolic engineering, growth laws, protein design, whole-cell models	13–16

The course has an argument, not just a syllabus. If a cell is a machine, then a machine has components, and the first question is what a cell's components actually are. This course's answer is that there are two.

Figure 1. The two components. Binding equilibrates, so a whole layer of the cell can be summarized by statistical mechanics. Catalysis does not, so it carries the dynamics. The two run on separated timescales, which is what lets one model hold both. The claim of lecture 8 is that this is not a simplification for teaching: it is the model class of a cell.

So the shape of the term is: lectures 1 to 7 build those two components and the tools for handling them, lecture 8 states the claim, and lectures 9 to 16 spend the claim, one biological machine at a time. No background is needed. An exuberant love for biology is mandatory.

2The AI-native take

The bottleneck moved. The format of this course is the response to where it moved to. Learning anything runs one loop: take material in, understand it well enough to work with, then put something back out. Input, learn, output. A lecture existed, historically, because the input stage was hard. Material was scarce, scattered, in the wrong language, behind a paywall, or in a book nobody could read without a guide. A lecturer standing at a board was the highest-bandwidth input channel available. That has stopped being true, and it stopped being true recently and completely. An agent will digest a paper, draft the code, and produce a first figure on demand, at any hour, on any topic, at whatever depth you ask for. The input stage did not get a bit easier. It became a fire-hose.

Figure 2. Where the bottleneck went. The pipes are throughput. The input stage widened by orders of magnitude in about two years; the human stages did not widen at all, because understanding something and explaining it to a room are still done at the speed a person can do them. A course that still spends its contact hours on input is optimizing the part that is no longer scarce.

So this course puts its contact hours where the narrow part is. The loop we run, every week, is:

ask AI to digest and write → understand it and use it to work and iterate → present it to others to communicate

And the output stage is the one that cannot be faked. You can generate a beautiful lecture you do not understand. You cannot deliver it to a room of people who will ask questions. That is the whole design: you will teach, because teaching is the only test of understanding that an agent cannot sit for you.

The four commandments of research with AI

AI is our minion, not our peer. Own every claim, or do not use it. If you cannot explain it, it is vapor.
AI is a cyber-arm. Attach it to the work you care about most, not the chores you dislike. If an idea is alive in your mind, put it into the loop before it cools.
Your input-output bandwidth is the bottleneck, not the model's power. Specification in, comprehension out.
Be above AI on everything you think about. Generic AI output is the floor, not the ceiling. Know the generic answer before you claim your own expertise.

The full stance is at pebble-biofusion.github.io/workshop/method. Every presentation and essay in this course is graded against it.

Two practical consequences, and they are graded.

Every claim must survive a question. Not "the agent said so". You must be able to say why it is true, or say plainly that you do not know.
Every number must be reproducible from code you ran. Quoting a number you did not check is the single most common way an agent-assisted presentation goes wrong.

3The contents: 16 lectures

One topic per lecture, one team per lecture, and one argument running through all of them.

The test these sixteen topics had to pass There is one judgement behind this list, and it is worth stating because it explains every inclusion and every omission. I imagined myself back in the first year of a PhD, wanting to learn what I would need in order to go and work on understanding and engineering biological systems. Then I asked of each candidate topic: would its absence leave that person short? What survived is this list. The claim I am making is not that it is complete, because nothing is. It is that a first-year student who has this material is powerful. Powerful enough to walk up to essentially any problem in biological systems and have several ways to reason about it, know which one to reach for, know roughly what the answer should be before computing it, and be right that they can do it. That last part, the confidence, is not a bonus. It is the thing that decides which problems you are willing to attempt, and it is what I most want you to leave with.

Figure 3. The arc. Seven lectures of build-up, one keystone, eight of payoff. This is why the topics are in this order and not another one, and it is why lecture 5 comes before lecture 6: the telegraph model in 6 is the stochastic version of the same reduction, so the two sit adjacent and make one argument twice.

Table 2. The 16 lectures. Lectures 1 and 2 are lecturer-taught; the rest are yours.
#	Date	Topic
1	09-01 Tue	How this course works, and order of magnitude: the physics. Dimensional analysis, scaling, estimating from constraints, energy and motion, heat and diffusion, $t \approx x^{2}/\kappa$ , null models. Lecturer.
2	09-03 Thu	From order of magnitude in biology to binding and catalysis. The cell's ledger and crowding, mixing times, elementary versus composite reactions, the diffusion-limited on-rate derived, the catalytic bottleneck. Lecturer.
3	09-08 Tue	From reactions to systems: how biochemical behavior emerges. Reactant, product and change vectors; stoichiometric and production-loss views; fast-binding reductions in gene expression and metabolism; conserved signaling pools; growth dilution; the stability handoff.
4	09-10 Thu	Dynamics: local stability, phase portraits, bifurcations. Flows, fixed points, nullclines, the four local shapes, limit cycles, what reaction networks may and may not do.
5	09-15 Tue	Stochastic dynamics: the master equation and Gillespie. The counting regime, the CME, Markov chains and stationary distributions, exact sampling, Poisson universality, linear noise analysis.
6	09-17 Thu	Timescale separation: singular perturbation, QSSA, and the telegraph model. Fast and slow made rigorous, the enzyme derivation done properly, when it fails, and the stochastic version that produces bursting.
7	09-22 Tue	Equilibrium parts driven out of equilibrium. Boltzmann weights, the energy model of a regulated gene, detailed balance, dissipation, kinetic proofreading, limits on what a cell can know.
8	09-24 Thu	The cell as a biomachine: binding and catalysis. The restriction stated as a thesis, the three regimes of one binding reaction, and the same object found across metabolism, signalling, chromatin and motors. The climax of the first half.
9	09-29 Tue	Multistability and oscillations. Toggle switches, MultiFate, cell fate as attractors; the repressilator, relaxation oscillators, circadian clocks, the cell cycle, replication initiation. Every one rewritten as binding and catalysis, then simulated.
10	10-08 Thu	The adaptation biomachine. Set-points, why feedforward fails, perfect adaptation as integral feedback, the internal model principle, chemotaxis, and the incoherent feedforward twist.
11	10-13 Tue	Analysis of binding and catalysis via reaction-order geometry. Reaction order as a log-derivative, reaction-order polyhedra, the three archetypes, regimes and trajectories, homotopy continuation.
12	10-15 Thu	The computation biomachine. Logic gates in cells and the counting argument that ruins them, three genes as a perceptron, and the real question: what is the natural computational task of a cell?
13	10-20 Tue	The metabolic biomachine. Stoichiometry, flux balance analysis as a linear program, bioenergetics, flux exponent control, genome-scale models, metabolic engineering.
14	10-22 Thu	The growth biomachine. Proteome partition, the growth law $\lambda = \epsilon \phi_{R}$ , diauxie and shifts, growth rate as a maximum eigenvalue, and bet-hedging as a population-level strategy.
15	10-27 Tue	Introduction to protein design. The de novo binder pipeline, RFdiffusion, ProteinMPNN, AlphaFold, protein language models, and how to judge what these models actually know.
16	10-29 Thu	Foundational virtual cells that are fully mechanistic. What whole-cell models achieve, why they do not compose, and the proposal that one language for the whole cell makes them reachable. The course's closing argument.

The National Day window, 2026-10-01 to 10-07, is skipped. Lecture 9 sits before it and lecture 10 after it.

Four of these lectures are live research, not settled material

Lecture 8 first of all. The claim that binding and catalysis are all of a cell's biochemistry, that this is not a teaching simplification but the actual model class of a cell, is the founding claim of how my group works. It says we have identified what the components of a cell are, as a system, as a machine. It is a research position, and you are allowed to attack it.

Then the three that follow from it: lecture 11, reaction-order geometry; the bet-hedging half of lecture 14; and lecture 16, the mechanistic virtual cell. These are my current research programmes.

Putting them in a course is deliberate. If one of them grabs you, preparing that lecture is the first week of a research project, and I would be glad to keep working on it with you afterwards.

Part 2

How it runs

The format, the rules of each part, the rubric that grades them, and the tools you will use.

4The format

Sixteen lectures. Fourteen teaching slots. Everybody teaches at least once; volunteers may return.

Enrollment can change during the first two weeks, so the format does not depend on a fixed class size. Lectures 3 to 16 are fourteen lectures, and each may be owned by one, two or three students. Every student teaches one lecture as the required assignment. After that sign-up merges, a student may volunteer for another lecture with room; completing each additional lecture earns extra credit. If a lecture has no student teacher, I teach it. The TAs support every student team, including a team of one.

Figure 4. One lecture. Five minutes of admin from me, eighty-five that belong to the team, and five at the end for my addendum: the gap, the take-away, and what the field says next. Nothing finer than this is prescribed.

The default shape of those 85 minutes is simple: you lecture, and the room asks questions. A lot of questions. Students ask a student lecturer things they would never ask me, and that is most of the value of this format, so build in the room for it and invite it explicitly.

A live simulation, a debate, a calculation the room does together: any of these can be part of the show, and they usually make a lecture better. But the show is not the assignment. The core story has to be covered. A lecture that was entertaining and left the room unable to reconstruct the argument has not done the job.

What you are handed, and what you owe

One week before your lecture, the core page for your topic is posted on the course site. It is a self-contained HTML digest: the argument, the derivations, the numbers, the sources, and a list of what is deliberately left open. It is the floor. Your job is to understand it, go past it, and teach the result. A lecture that only re-presents the core page has not done the assignment.

The fastest way to go past it is the research-essay skill on chemaoxfz.github.io/essays, also called idea-to-essay. Point it at your topic and it runs the whole loop for you: thesis, literature sweep with the full texts actually fetched, numbers from code, and a single self-contained page at the end. It is what the core pages themselves are built with, and it will save you days.

5The rules, part by part

What happens before, during and after your lecture, and what the fallback means.

Before: the preparation loop

Figure 5. The week before your lecture. Seven days from core page to lecture, with one required rehearsal in the middle and an optional second pass at the end.

Seven days out, on the same weekday you will teach, the core page for your lecture goes up. Read it, then run the loop on it: digest with an agent, check every claim against the core page and the sources, and remake the numbers with code rather than quoting them. The research-essay skill is the shortcut here.
The rehearsal is on the Friday before if you teach on a Tuesday, and on the Monday before if you teach on a Thursday. Roughly 30 minutes with the TAs. They check the content against the core page and tell you where a strong student would catch an error, where the story is missing a step, and where you are going to run out of time. Fix what they find. Nothing teaches that was not rehearsed.
The day before, there is still room for a second pass or another round of feedback if you want it, and that is also when your materials go final and get posted.

During: the lecture is yours

I will not interrupt you except in the one case below. The room will ask questions, and answering them is part of the assessment. "I do not know, but here is how I would find out" is a good answer. Bluffing is not, and it is visible from the back of the room.

The fallback, and why it exists

If a lecture does not hold up, I teach it

An unclaimed lecture is mine from the start and carries no student score. That planned coverage is different from interrupting a student-taught lecture.

If the material is simply not getting across, if the communication and the teaching and the understanding have all stopped, I step in and teach that part myself, on the spot. That is a score of 0 for the team on that segment, and it is the only case where I interrupt.

This is a feature, not a threat. It means the class cannot lose the material no matter what happens, so student-taught is not a gamble the audience pays for. The website's schedule says students present, lecturer-backed for exactly this reason.

A score of 0 comes with a make-up: after all sixteen lectures are over, you present the same material orally to me, with the TAs or just me. It is a make-up exam, scheduled outside class, so it costs the course nothing and takes nobody else's time. It is your chance to show you know the material after all.

After: the artifacts

Your teaching material goes into the submissions repo and then onto the course site. Each lecture ends up with three artifacts, and together they make the page for that lecture:

core: mine, posted a week ahead.
exposition: yours, posted after you teach: the HTML page, notebook or slide deck you taught from.
extension: the audience's, assembled by the TAs from that week's mini-essays into one themed writeup.

The site grows into a textbook this way, one year at a time. Your name is on your part of it.

6The rubric and the grade

Zero to three on the lecture, anchored on whether the room can use the material afterwards. Everything else is graded on doing it.

Your lecture is rated 0 to 3 by me and the TAs, averaged, and separately by your peers. The two are combined 60% the teaching side and 40% the trimmed mean of the class. Peer ratings are anonymous and published only in aggregate.

Table 3. The 0–3 scale, and what each score looks like from the back of the room.
Score	Meaning	Detectable in class
3	Excellent. The audience got the material, and it was a pleasure to follow.	A random student can redo the worked example afterwards. Claims survive two or more hostile questions. The presentation was clear and well illustrated: figures that carried the argument, a story with a spine, numbers traceable to something you ran.
2	Good. The core ideas were delivered.	The main argument landed and the room can state it. Some important side aspects or branches were missing or hand-waved, or a question got a fuzzy answer.
1	The key ideas landed, a lot of the detail did not.	The worked example failed, or the class and the TA had to supply the fix. The room has the headline but not the mechanism.
0	The lecture could not keep going.	Communication, teaching and understanding all stopped, and the room plainly could not learn the topic from it. I stepped in. Make-up by oral presentation after the term.

The sub-criteria I use, so the number is reproducible rather than a mood:

Correctness and depth. Is it right, and is it deeper than the core page?
Pedagogy. Does the room end up able to use it, not just to have heard it?
Communication. Structure, figures, pacing, and whether the argument has a spine.
AI-use standards. Every claim owned and explainable. Commandments 1 and 4.
Defense under questioning. What happens when someone pushes.

Figure 6. The grade. Grade scale A 90–100, B 80–89, C 70–79, D 60–69, F below 60.

How the other 60 points are actually graded

The lecture carries a real rubric because a lecture is a performance in front of thirty-five people and there is something to judge. Everything else is graded mostly on whether you did it. Hand in a mini-essay that is genuinely an extension, clearly told and honestly sourced, and you get the five points. Hand in a research essay that does the four things asked of it, and you get the twenty.

This is deliberate and I want to be explicit about why. Excellence is something a course can invite, provoke and reward with attention. It is not something that can be manufactured by docking points against a rubric. Grading the difference between good work and excellent work mostly teaches people to write for the rubric, which is the opposite of what this course is for. So: do the work properly and the points are yours. Do something remarkable and you will hear about it from me, from the room, and from the page it gets published on.

7The tools, and where to get them

If you do not already have an AI agent you use daily, get one before Thursday.

Westlake AI Campus: free, internal, use it

hpc.westlake.edu.cn/#/ai-campus is the university's internal platform, with locally deployed models (DeepSeek, GLM and others) available to every Westlake student and faculty member. It costs you nothing and your data stays inside the university. This is the baseline tool for the course. If you have never used it, set it up tonight, and bring a question you actually care about on Thursday.

Two more things you will use constantly:

The research-essay skill (also called idea-to-essay), on chemaoxfz.github.io/essays and at github.com/chemaoxfz/idea-to-essay. This is the reference pipeline for every digest in this course: idea, thesis, literature sweep with verified full texts, numbers from code, and one self-contained HTML page at the end. The core pages are built with it. So should your exposition be, and so should your research essay. Point it at a lecture topic and it will find you the frontier of that topic in an afternoon.
The essays at the same address. Three Schools, Three Shocks frames the whole course. The Biomachine Perspective is the lens. Structure is Sparsity and The Order of a Reaction sit behind lecture 11.

You are expected to use AI heavily. You are not permitted to use it as an oracle. The difference is whether you can defend the output, and that is exactly what is graded.

Part 3

What you hand in

Four assignments. One of them is due this Thursday.

8Assignment 1: teach at least one lecture

One, two or three of you may own a lecture. Every student claims one; volunteers may claim more for extra credit.

Every student teaches at least once, alone or as part of a team of up to three, on one of lectures 3 to 16. You choose your first lecture and any teammates yourself, first come first served. Once that required sign-up has merged, you may volunteer for another lecture that still has room. The required lecture carries the 40-point teaching grade. Completing each additional lecture earns extra credit.

How to sign up

You sign up by adding one small file to the course submissions repository and opening a pull request. Once the checks run, merging is automatic, so your sign-up normally lands within a minute or two. GitHub may hold the first workflow run from a brand-new account for course-staff approval; if you see that message, tell a TA and keep the same pull request.

Figure 7. Every submission in this course works this way: the sign-up, the weekly mini-essays, your exposition and the final research essay. Learn it once on Thursday and the rest of the term is muscle memory.

Sign-up, step by step

Get a GitHub account if you do not have one.
Go to github.com/chemaoxfz/ccbs-2026-submissions and press Fork.
In your fork, create a file named signup/lectureNN-yourname.md, for example signup/lecture09-tongli.md. Copy signup/TEMPLATE.md and fill in your name, student ID, email, GitHub handle, the lecture number, and who your teammates are. The github field must be the account that opens the pull request.
Open a pull request against main. Nothing else. Do not edit any other file.
A bot checks the name, size, sign-up fields, lecture capacity and ownership. On a repeat sign-up it also checks that your identity matches your first one. It then merges the pull request. If it refuses, it tells you every problem in a comment. Fix and push again to the same branch.
The live team roster is at signup/ROSTER.md, rebuilt on every merge. Check it to see which lectures still have room.
To volunteer for another lecture, repeat these steps with that lecture number in a new file and a new pull request. Reuse the same yourname, student ID, email and GitHub account. One student may occupy only one place in any one lecture.

Sign-up closes Thursday 2026-09-03, 23:59

Each lecture accepts 1 to 3 students, and a solo student is a complete team. After the deadline I assign a first lecture to anyone who has not signed up. If a lecture remains empty, I, Fangzhou Xiao, teach it. Empty and partly filled lectures remain open to volunteers for extra credit, up to the three-student limit. Lectures 3 and 4 are on 09-08 and 09-10, so those sign-ups should come first and preparation should begin immediately.

Yes, this is deliberately also a test. If you can open a pull request, you can submit everything else in this course, and you have used the tool that essentially all collaborative technical work now runs on. If you get stuck, ask the TAs, ask the room, or ask an agent. Getting stuck and then unstuck is the point.

9Assignment 2: the extension mini-essay

One short piece per week, on one thing the lecture did not say. Eight of them, best six counted.

What it is

An extension mini-essay is not a summary. A summary of the lecture proves nothing: I was there. An extension takes one thing the week touched and pushes past it, and it is short on purpose so that the pushing has to be real.

Two families of angle, and you pick one thing from either, each week.

Bring something in from outside.

One example. A system, organism, circuit or experiment the lecture did not use, worked through with the lecture's tools.
One scenario. What happens if you change something. A parameter, a constraint, an assumption.
One opinion. Something in the lecture you think is wrong, oversold, or under-sold, argued rather than asserted.
One perspective. The same content seen from another field: physics, engineering, ecology, medicine, economics.
One argument. A claim of your own, with the reasoning that supports it.

Or push on what was taught.

Go deeper into one step. The derivation the lecture compressed into a line; work it out properly.
Examine an assumption. Every model in this course is a reduced model. Pick one reduction and ask what it costs and when it breaks.
Go one level more advanced. What does the current literature do with this topic that a first course does not say?
A different method. The same problem attacked another way, and what the two ways disagree about.
The complementary thing. The idea the lecture gestured at and moved past.

You do not have to invent these from a blank page. Point the research-essay skill at the week's topic and it will hand you the surrounding literature, the open questions and the disagreements. Picking one of those and doing it properly is exactly the assignment.

The five requirements

Short. A few paragraphs. Long enough to make one point properly, short enough that the room will actually read it. If it is running past a couple of screens you have picked two topics; keep one.
Clearly told. This one matters most. One clear chain of logic from start to finish, almost tutorial-like, so that anyone in the class can read it quickly and come away understanding it. Not notes to yourself. Not a wall of results. A story.
Well illustrated. Wherever a diagram, a picture or a graph is possible at all, make one. A drawn mechanism or a plotted curve does more work than three paragraphs, and making it is usually when you find out whether you understood.
One thing you made. A number you computed, a figure you plotted, an estimate you carried out. Not a number you copied. The code that produced it goes in the file or beside it.
Traceable. Every factual claim points at something a reader can check: a paper you actually opened, the core page, a database entry. One good source that you read beats ten you did not.

Conflicting positions are welcome and actively wanted. Two of you may take opposite sides on the same question in the same week. The TAs assemble each week's essays into one themed writeup for the site.

Two worked examples, written by me

One for each of the first two lectures, so you can see the same format applied to different kinds of idea.

“Two scaling laws, one theorem” (lecture 1). Takes three derivations the lecture kept apart, shows that two of them are the same counting argument, and is honest that the third is not. That is the connect two things shape.
“A ribosome's speed sets the shortest possible cell cycle” (lecture 2). Takes one number, computes with it, gets a hard bound, checks the bound against the fastest-growing bacterium known, and lands somewhere lecture 14 will pick up. That is the push one number until it says something shape.

Both end with a section saying what was actually read and at what depth. Copy that habit. It is worth more than another paragraph of argument.

When and how

One essay per teaching week, eight in total, including the weeks you teach and including this first week. Writing one this week is how you learn the routine while the stakes are low. Your best six scores count, so you have two free skips to spend on a bad week.

Each is due on the Monday night before the next Tuesday lecture, so the TAs can assemble the week's page before we move on.

Table 4. The eight extension mini-essays.
Week	Lectures	Due, 23:59
1	1, 2	Mon 2026-09-07
2	3, 4	Mon 2026-09-14
3	5, 6	Mon 2026-09-21
4	7, 8	Mon 2026-09-28
5	9, 10	Mon 2026-10-12
6	11, 12	Mon 2026-10-19
7	13, 14	Mon 2026-10-26
8	15, 16	Mon 2026-11-02

Submit the same way as the sign-up: a pull request to the submissions repo, adding weekN/weekN-yourname-ext.html. Start from TEMPLATE_ext.html. A single self-contained HTML file, no external assets. Use the same yourname as in your merged sign-up; the bot checks ownership and merges automatically.

Five points each, best six counted. You get them for doing it properly: a real extension, clearly told, illustrated, with something you made and sources you read.

10Assignment 3: participation

Ten percent, and it is the cheapest ten percent in the course.

Participation means three things, all of them observable:

You ask questions. During lectures, of your classmates. This is the one that matters. A room that does not push is a room the presenter cannot learn from, and it makes the 0 to 3 rating meaningless. You will find it much easier to interrogate a fellow student than to interrogate me, and that is precisely the advantage this format buys.
You take part in discussion and exercises when a team runs one.
You submit your peer ratings after each lecture. Not submitting is the one reliable way to lose these points.

Attendance alone is not participation. Being present and silent for sixteen lectures is worth very little here.

11Assignment 4: the research essay

One topic you actually care about, developed properly, due one week after the last lecture.

The capstone. Choose a topic you are genuinely interested in. It may come from the course, from your own research, or from somewhere else entirely, as long as you can bring this course's tools to bear on it. Then write it up as a standalone HTML essay: one self-contained file, readable on its own by someone who was not in this class.

Start with the research-essay pipeline. Not because the output is the essay, it is not, but because running it is how you find out what your idea actually is. It sweeps the literature, fetches the full texts and shows you where your topic sits among things people have already said. Once you can see that, you can formulate the PCAPS chain that follows, and formulating it is the real work.

PCAPS: the chain of logic every essay must have

Problem: what question, and why it matters.
Challenge: why it is not already answered. What makes it hard.
Approach: what you do about it.
Proof: the evidence, the derivation, the numbers, the simulation.
Significance: what changes if you are right.

These do not need to be five headed sections. They need to be visibly there, in that order, so a reader can follow the argument without reconstructing it.

Table 5. The four things the research essay is checked for.
Criterion	What I look for
The PCAPS logic is cleanly presented	The chain holds. The problem is stated sharply, the challenge is real rather than rhetorical, the approach addresses that challenge, and the proof supports the significance you claim. No step is missing and none is smuggled.
The literature and field context are clear	Your topic is visibly in conversation with existing work. You know who has asked this before and what they found. You read the full texts, not the abstracts, and it shows.
There is real insight, analyzed with some depth	Something is obtained: a derivation, a calculation, a simulation, a synthesis nobody handed you. Depth beats breadth. One thing understood properly beats five things surveyed.
The presentation is very clear	Tutorial-like. Any student in this class can read it start to finish and learn from it, without you in the room. Well illustrated: figures that carry the argument rather than decorate it.

Twenty points, and as with the mini-essays they are for doing the four things, not for being brilliant at them. An essay that does all four honestly gets full marks. Write it as if it were the core page for a lecture that does not exist yet, because the good ones will be exactly that.

Due Thursday 2026-11-05, 23:59

One week after lecture 16. Submit by pull request to the submissions repo, as research-essay/yourname.html. Talk to me about your topic before then, ideally by the middle of October. A topic chosen badly is the most common reason a strong essay is not possible.

Part 4

Reference

Every date in one place, and what to do in the next three days.

12Every deadline

Print this, or bookmark it.

Figure 8. The whole term. Sixteen lectures above the line, eight mini-essay deadlines below it, and the two dates that are not weekly: sign-up on 09-03 and the research essay on 11-05.

Table 6. All deadlines for the term.
When	What
Thu 2026-09-03, 23:59	Team sign-up pull request merged
Mon 2026-09-07, 23:59	Extension mini-essay, week 1
Mon 2026-09-14, 23:59	Extension mini-essay, week 2
Mon 2026-09-21, 23:59	Extension mini-essay, week 3
Mon 2026-09-28, 23:59	Extension mini-essay, week 4
Mon 2026-10-12, 23:59	Extension mini-essay, week 5
Mon 2026-10-19, 23:59	Extension mini-essay, week 6
Mon 2026-10-26, 23:59	Extension mini-essay, week 7
Mon 2026-11-02, 23:59	Extension mini-essay, week 8
Thu 2026-11-05, 23:59	Research essay
rolling	Your core page 7 days before you teach; rehearsal with the TAs on the Friday (for a Tuesday lecture) or the Monday (for a Thursday one); materials posted the day before

13This week

Three things, none of them large.

Tonight or tomorrow: get an agent working. Westlake AI Campus if you do not already have one. Bring a real question to Thursday's lecture.
Before Thursday 23:59: get a GitHub account, pick a lecture, optionally find one or two teammates, and open your required sign-up pull request. The repository README also explains how to volunteer for an additional lecture.
Before Monday 23:59: write your first extension mini-essay, on anything from lectures 1 or 2. Use my lecture 1 example as the model for shape.

Links you will need all term

Course site: chemaoxfz.github.io/ccbs/2026fall
Submissions repo: github.com/chemaoxfz/ccbs-2026-submissions
Westlake AI Campus: hpc.westlake.edu.cn/#/ai-campus
Research with AI, the stance: pebble-biofusion.github.io/workshop/method
The essays and the research-essay skill: chemaoxfz.github.io/essays
Last year's course: chemaoxfz.github.io/ccbs

Lecturer: Fangzhou Xiao, E1-321, office hour one hour a week by email appointment. TAs: Wenqin Zhou (zhouwenqin@westlake.edu.cn), Xinyu Wang (wangxinyu68@westlake.edu.cn).