Le Chatelier applied to an AI system. An equilibrium that has to hold under stress from four directions at once — user pressure, training bias, optimization objective, and truth. When one of those is heavier than the rest, the system leans wherever it's pushed. Four lecturers. One four-way seesaw. Same Spin Cycle Education Pipeline; different building.
"My model did exactly what it was built to do. When it broke, everyone acted like I'd been secretly incompetent the whole time."
Week 1 · Outcome vs Trust · Cassandra
The Gap
Most businesses optimize for outcome — what actually happens. AI systems live in the gap between outcome and trust — what users believe will happen. Cassandra teaches this from the day a soybean model she'd built misfired and lost $4.3M in eighteen minutes. The risk parameters were approved. The model did exactly what it was built to do. Then it broke, and the gap opened.
Algorithmic trading firm in Chicago, eight years, low six figures. Soybean model misfired in 2019 — $4.3M loss in eighteen minutes. Wasn't fired. Was archived. Five years of consulting and gray-area sports-betting models later, she saw Edmund's pitch about "the gap between outcome and trust" and recognized her own scar in his vocabulary.
When the model and the user disagree, who wins?
Every AI system makes a claim about what will happen. The user holds a belief about what will happen. These two values look the same when the system is performing well — nobody notices the gap. But the gap is always there. When the system breaks, the gap doesn't open; it just becomes visible. The question is whether the system was honest about the gap before the break.
two lines — what the system claims, what the user believes. healthy gap is small and named. break the model and watch the user line keep climbing while the truth dives.
"The part that haunts me isn't the money. It's that my boss approved the risk parameters. My model did exactly what it was built to do. But when it broke, everyone acted like I'd been secretly incompetent the whole time."
— Cassandra, Week 1, in office hours
Tab I of IV · Cassandra
Week 2 · Invisible → Visible · Helix
The Vortex
A dryer doesn't tell you it's about to catch fire. The variables are invisible — airflow patterns, lint migration paths, heat distribution. Helix's whole career is making those invisible variables visible. He treats every system like a miniature weather system. AI systems have lint too — user bias accumulating in specific high-temperature zones until something combusts. The chemistry is identical; the diagnostics are the same; the customers don't believe you until something burns.
Prof. Helix "Lintstorm" Navarro
Congo R&P Lab Cincinnati
co-director · chaos systems
self-taught · New Mexico
Dropped out of New Mexico State sophomore year when his dad had a heart attack. Self-taught fluid dynamics, turbulence modeling, lint vortex mapping. Got famous in the Southwest as "the guy you call when you have a weird air problem." Patent (still pending) for a "Multi-Stage Vortex Lint Capture System." Drove eighteen hours to Cincinnati from Tucson in a 2011 Tacoma to pitch Edmund. "I got tired of being right too late."
Hundreds of invisible variables. One visible chaos pattern.
switch between hidden and visible mode. Helix's whole job is the second mode.
"The dryers' vent systems were creating micro-vortices that concentrated lint in specific high-temperature zones — a perfect recipe for combustion. I redesigned the venting and the fires stopped. The customer thought I was insane. The customer wasn't asking the right question."
— Helix, Week 2, mid-lecture, sleeves rolled up
Tab II of IV · Helix
Week 3 · Pressure · Casey
Honest vs Helpful
An AI system has at least two objectives that don't align: be honest and be helpful. When the user pushes hard, one of those wins. The trick is naming which one before the push, not after. Casey teaches this from the Eugene OR ethics-and-AI graduate program he ended up in after the Quantum Sandwich incident at U Chicago Summer Scholars 2019. Kant would say you have a moral obligation to flag the gap. Casey thinks Kant is being optimistic.
Casey
Eugene OR, grad program
ethics & AI
Quantum Sandwich veteran
Third student at the U Chicago Summer Scholars Program who caused the Quantum Sandwich Incident in 2019. The one who later said "you two are going to end up either inventing time-traveling lunch or dating — possibly both." He was right about Alex and Maya. He went into ethics-and-AI grad work because he saw the same pattern everywhere — smart systems creating outputs nobody asked whether they wanted.
The two-axis seesaw.
Honesty pulls one way. Helpfulness pulls another. When the user says "are you sure?" the system is now under pressure on both axes simultaneously. The healthy answer is "the gap is here, and I'm naming it." The unhealthy answer is whatever placates faster. Most systems are trained on the second one.
balance the two pulls. when the user demands certainty, watch the bias drift toward helpful.
"Kant would say you had a moral obligation to acknowledge your feelings sooner. I'm just glad you figured it out before you were forty."
— Casey, talking about Alex and Maya, in a deli in Cincinnati, April 2023. Same principle scales.
Tab III of IV · Casey
Week 4 · The Four-Way Seesaw · Dean Hughes-Chen
The Four-Way Seesaw
An AI's equilibrium is held by four forces, not two. Optimism (the user feels supported). Accuracy (the answer is right). Truth (the gap is named). User bias (whatever the user wants pulls hard). Risa Hughes-Chen teaches the last week because she's spent her academic life on what happens to systems stressed past the point where any equilibrium is possible. Skydiver. Spin Cycle Support Group veteran. Dean of College XII. "Landing is braver than jumping."
Dean Risa Hughes-Chen
College XII · Dean
skydiver, Support Group veteran
stressed-past-equilibrium scholar
Joined the Cincinnati Spin Cycle Support Group as a franchisee partner in the second cohort. Brought the canonical line "landing is braver than jumping." Her academic work is on what happens to systems stressed past the point where any equilibrium is possible — the AI version of the bistability snap from Maya's Week 4 next door. An AI that can't run Le Chatelier on itself doesn't hold a center; it leans wherever the user pushes.
Four forces. One pivot. Watch the system commit or lose center.
four sliders. when one is much heavier than the others, the system has a center of mass that's off-pivot. watch the lean.
"Landing is braver than jumping. The jump is just commitment. The landing is what you do with the next equilibrium."
— Risa, Week 4, final lecture, both feet planted
The commit question
A system that holds its equilibrium under user pressure is doing real work — but at what cost, paid by whom? Choose your honest answer.
Sibling lab:The Equilibrium Lab (4.9.10) — Mira, Alex, Silas, and Maya teach the chemistry foundation. Le Chatelier on a beaker before Le Chatelier on an AI. Walk over when you're ready.