diff --git a/.github/workflows/collab.yml b/.github/workflows/collab.yml
index ec7301ae9..43cd2cb44 100644
--- a/.github/workflows/collab.yml
+++ b/.github/workflows/collab.yml
@@ -27,7 +27,7 @@ jobs:
             sphinx-togglebutton==0.4.5 \
             arviz \
             sphinx-proof==0.4.0 \
-            sphinx-exercise==1.2.1 \
+            sphinx-exercise==1.2.2 \
             sphinx-reredirects==1.1.0
           apt-get update
           apt-get install dvipng texlive texlive-latex-extra texlive-fonts-recommended cm-super    
diff --git a/environment.yml b/environment.yml
index 599ece0d3..7314f0d16 100644
--- a/environment.yml
+++ b/environment.yml
@@ -10,7 +10,7 @@ dependencies:
     - quantecon-book-theme==0.22.1
     - sphinx-tojupyter==0.6.0
     - sphinxext-rediraffe==0.3.0
-    - sphinx-exercise==1.2.1
+    - sphinx-exercise==1.2.2
     - sphinx-proof==0.4.0
     - sphinxcontrib-youtube==1.5.0
     - sphinx-togglebutton==0.4.5
diff --git a/lectures/exchange_rate_learning.md b/lectures/exchange_rate_learning.md
index 50611388a..d569c1448 100644
--- a/lectures/exchange_rate_learning.md
+++ b/lectures/exchange_rate_learning.md
@@ -1318,6 +1318,12 @@ To compare like with like, run the Newton–Raphson learner at Arifovic's parame
 Here is part 1.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: "Newton-Raphson learning at Arifovic's parameters"
+    name: fig-xr-ex4
+---
 ea, sa, la = newton_learning(arifovic, [6.0, 4.0], [0.45, 0.55], T=3000)
 eb, sb, lb = newton_learning(arifovic, [6.0, 4.0], [0.45, 0.55], T=3000, κ=2.0)
 
@@ -1460,6 +1466,12 @@ $\varepsilon_t \sim N(0, 1)$ and $\sigma = 1$, at Sargent's parameters.
 Here are parts 1 and 2.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: "Limiting exchange rates from identical initial conditions"
+    name: fig-xr-ex5
+---
 T, σ = 600, 1.0
 e_det = newton_learning(kw, [3.0, 2.2], [0.30, 0.50], T=T)[0][-1]
 
diff --git a/lectures/genetic_classifier.md b/lectures/genetic_classifier.md
index 797f343e2..43f8596db 100644
--- a/lectures/genetic_classifier.md
+++ b/lectures/genetic_classifier.md
@@ -1647,6 +1647,12 @@ for b, P in paths.items():
 The figure shows all forty paths.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: "GA forecasters in Bray's market"
+    name: fig-gc-bray-ga
+---
 fig, axes = plt.subplots(1, 2, figsize=(10, 4), sharey=True)
 for ax, (b, P) in zip(axes, paths.items()):
     ax.plot(P.T, lw=1, alpha=0.6)
diff --git a/lectures/information_transfer.md b/lectures/information_transfer.md
index b97881efd..0a79f57f5 100644
--- a/lectures/information_transfer.md
+++ b/lectures/information_transfer.md
@@ -1149,6 +1149,12 @@ Under the coarse experiment the planner takes $a_2$ at $0.2$ and $a_1$ at $0.8$,
 Under the fine experiment the planner always takes $a_2$, so each player's expected utility is the height of that player's line for $a_2$ at the prior (open dot).
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: Riley's example, with expected utilities read off at the prior
+    name: fig-it-riley
+---
 fig, axes = plt.subplots(1, 2, figsize=(10, 4))
 for ax, mat, who in zip(axes, (U_ri, W_ri), ("planner", "agent")):
     for k in range(2):
diff --git a/lectures/learning_approximation.md b/lectures/learning_approximation.md
index 4442d9a42..9277da934 100644
--- a/lectures/learning_approximation.md
+++ b/lectures/learning_approximation.md
@@ -1211,6 +1211,12 @@ table.round(4)
 The rules themselves show the two failure modes.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: "Kernel rules at three bandwidths"
+    name: fig-la-bandwidth
+---
 fig, ax = plt.subplots(figsize=(7, 4.2))
 for h, colour in [(0.003, 'C0'), (0.03, 'C1'), (0.20, 'C2')]:
     ax.plot(G, kernel_smoother(T=T_short, h=h)[0], '-', lw=1, color=colour, label=f"$h = {h}$")
@@ -1382,6 +1388,12 @@ eigenvalue of $DT(\theta^*)$.
 ```
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: "Distance of online beliefs from the approximate equilibrium"
+    name: fig-la-rate
+---
 T_online = len(θ_path)
 ts = np.unique(np.logspace(2, np.log10(T_online - 1), 60).astype(int))
 late = ts >= 10_000
diff --git a/lectures/lucas_prescott_search.md b/lectures/lucas_prescott_search.md
index c1a5f1cfa..3b6f0cdaa 100644
--- a/lectures/lucas_prescott_search.md
+++ b/lectures/lucas_prescott_search.md
@@ -1413,6 +1413,12 @@ Eu2 = np.array([r['unemployment'] for r in res2])
 The paper puts $\lambda$ on the vertical axis.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: Workforce demand and unemployment in the paper's two-state economy
+    name: fig-lps-two-state
+---
 fig, axes = plt.subplots(1, 2, figsize=(10, 4.2))
 axes[0].plot(Ey2, grid2, lw=2)
 axes[0].set_xlabel('workforce demand per market')
@@ -1688,6 +1694,12 @@ print(f"at U = 10 the largest gap on a grid of λ is {max(gaps_10):.4f}")
 Each of these inflows has two solutions, and each serves its own population; at $U = 10$ there are none.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: Undirected search, the populations served by the solutions at each inflow
+    name: fig-lps-undirected
+---
 Us = np.array(list(locus))
 scale = 200/N_fine                       # back to the lecture's units
 fig, ax = plt.subplots(figsize=(7, 4))
@@ -1795,6 +1807,12 @@ for Δ in (-0.04, -0.02, -0.01, 0.01, 0.02, 0.04):
 The no-surprise line reproduces the equilibrium, as it should.
 
 ```{code-cell} ipython3
+---
+mystnb:
+  figure:
+    caption: An unanticipated shift in demand moves unemployment and wages in opposite directions
+    name: fig-lps-phillips
+---
 u_pts, w_pts = np.array(sorted(points)).T
 fig, ax = plt.subplots(figsize=(6.5, 4))
 ax.plot(u_pts, w_pts, 'o-', lw=2)