Summary

Style audit of the lecture-python.myst series.

• Audit date: 2026-08-24

• Corpus snapshot: e25fdf2345

• Lectures audited: 145

• Average overall score: 7.7 / 10

• Average per-category scores: writing 4.5, math 7.0, code 7.6, figures 6.5, references 9.5, links 9.8, admon 10.0

• JAX: out of scope — the qe-jax-* rules target lecture-jax.

• Judgment-review coverage: all lectures reviewed.

Writing is the binding constraint. At 4.5 it is the weakest of the seven categories — as it was in 2026-05, when it scored 5.9 — and it is now at or below the 4-point floor in 72 of the 81 HIGH lectures. Math floors 26 and Figures 8; Code, References, Links and Admonitions floor none. In 49 of the 81, Writing is the only floored category, so a Writing pass is the one action that can clear more than half this series’ HIGH list on its own. Two very different jobs sit under it. qe-writing-006 (sentence case below H1, the highest-weighted Writing rule) appears in 59 of the 72 and qe-writing-008 (runs of extra whitespace, the lowest-weighted) in 60, and both are mechanical; but 57 of the 72 also carry qe-writing-001 or qe-writing-004, which want an editor rather than a script.

At 145 lectures this series is 42 % of the corpus and holds 81 of its 197 HIGH lectures, and it scores like the corpus largely because it is the corpus: overall 7.7 against 7.7, math 7.0 against 7.0, figures 6.5 against 6.5. The floor proportions match too — Writing 89 %, Math 32 %, Figures 10 % of the HIGH list, the same three shares as corpus-wide. The extremes are what distinguish it. cross_product_trick at 5.5 is the lowest-scoring lecture in all 348, and qe-writing-008 at 2,569 occurrences across 89 lectures is the largest count of any rule in any series — 36 % of the corpus total for that rule, with ak2 alone at 197. The published Writing figure fell 1.4 points since 2026-05, but that is the judgment layer — the 2026-05 row folds in no review overlay and this one folds one into all 145 lectures. Like for like, on the evidence layer alone (history_mechanical.csv), Writing here rose 5.9 → 7.0, Math 6.6 → 7.4 and Figures 6.3 → 6.5, while the series grew from 110 lectures to 145 — so the comparable movement is an improvement, part of it composition.

A caveat this series shares with lecture-dp. 31 of these 145 filenames also exist in lecture-dp, which syncs them from here; 16 of the 31 are HIGH here and 15 are HIGH in both. Only 6 are byte-identical at this snapshot (cross_product_trick, ifp_discrete, ifp_opi, lq_inventories, mccall_model_with_separation, os_numerical) — for those, this repo is where a fix belongs and both copies clear. The other 25 share an origin and have since diverged, so the same defect usually appears in both but each copy needs its own edit. Whether the corpus totals should de-duplicate the 6 is open as compliance-lecture-style#3.

Priority distribution

Priority	Count	%
HIGH	81	55.9%
MEDIUM	1	0.7%
LOW	46	31.7%
NONE	17	11.7%

Top systemic issues across the series

Ranked by how many of the series’ lectures each rule reaches.

1. qe-fig-005 — Descriptive figure names for cross-referencing — 110 / 145 lectures, 446 occurrences.

2. qe-fig-001 — Do not set figure size unless necessary — 107 / 145 lectures, 462 occurrences.

3. qe-writing-008 — Remove excessive whitespace between words — 89 / 145 lectures, 2569 occurrences.

4. qe-fig-003 — No matplotlib embedded titles — 79 / 145 lectures, 329 occurrences.

5. qe-writing-006 — Capitalize lecture titles properly — 71 / 145 lectures, 410 occurrences.

6. qe-fig-008 — Use lw=2 for line charts — 69 / 145 lectures, 393 occurrences.

7. qe-writing-001 — Use one sentence per paragraph — 65 / 145 lectures, 165 occurrences.

8. qe-math-010 (proposed) — Blackboard \mathbb{P}, \mathbb{E}, \mathbb{V} with braces — 60 / 145 lectures, 689 occurrences.

9. qe-writing-004 — Avoid unnecessary capitalization in narrative text — 43 / 145 lectures, 148 occurrences.

10. qe-ref-001 — Use correct citation style — 36 / 145 lectures, 105 occurrences.

Clean across the series

Checked rules with no violation anywhere in the series — the conventions this series already holds to.

• qe-admon-002 — Use dropdown class for solutions

• qe-admon-003 — Use tick count management for nested directives

Series-level recommendations

1. qe-writing-006 — sentence-case the headings below H1 (71 / 145, 410 headings). The highest-weighted Writing rule in the spec, and present in 59 of the 72 Writing-floored lectures — the best return per heading edited. Scriptable against the curated proper-noun list in tools/qestyle_rules.py, but that list is heuristic, so read the diff before it lands. linear_models (25), markov_asset and linear_algebra (18 each) are the densest.

2. qe-writing-008 — collapse repeated spaces (89 / 145, 2,569 occurrences). The largest count of any rule in any series in the corpus, and present in 60 of the 72. It is the lowest-weighted Writing rule, so it moves a score less than its size suggests — but it is entirely safe to automate, which is why it ranks here rather than lower. ak2 holds 197 of them and var_dmd 180.

3. qe-writing-001 and qe-writing-004 (65 / 145, 165 blocks; 43 / 145, 148 words). Reading passes rather than sweeps: splitting a paragraph and deciding what is a proper noun both need an editor. Unavoidable, though — 57 of the 72 Writing-floored lectures carry one or the other, so items 1–2 alone will not lift them off the floor. Items 1–3 together are the only route to the 49 lectures floored on Writing alone, and a precondition for the 23 floored on Writing plus something else.

4. qe-math-010 (proposed) — expectation and probability operators, with braces (60 / 145, 689 occurrences). The largest Math finding, and present in 19 of the 26 Math-floored lectures. A mechanical substitution, but it edits equations, so review it rather than running sed blind. util_rand_resp (113) and prob_matrix (81) dominate.

5. qe-math-004 — un-bold the vectors and matrices (18 / 145, 509 occurrences) and qe-math-002 — ^\top for transpose (24 / 145, 494). Narrow and dense rather than broad: three lectures hold 370 of the 509 (hansen_singleton_1983 143, lln_clt 121, mle 106), and linear_algebra alone holds 114 of the 494. Good single sittings. With item 4 they cover the Math floor — of which only six lectures (affine_risk_prices, hansen_singleton_1983, imp_sample, likelihood_bayes, likelihood_var, mle) are floored on Math alone.

6. The figure rules, for the average rather than the triage. qe-fig-005 — name the figures (110 / 145, 446 figures) and qe-fig-001 — drop figsize= (107 / 145, 462 overrides) are the two largest reaches in the series and pure sweeps, but Figures is the floor in only 8 lectures and just two of those (ak_aiyagari, sargent_surico) are floored on Figures alone. qe-fig-003 — titles into captions (79 / 145, 329 calls) is a reading pass instead: each ax.set_title(...) becomes a caption someone has to write. two_computation (28) and navy_captain (20) are the heaviest.

7. Two structural one-line fixes, independent of the HIGH list. cross_product_trick.md:133 has {eq}`eq:Kalman102} — mismatched braces, pointing at a bare align* block that carries no label — and ifp_advanced.md:158 has a raw \label{a:y0} inside $$, which MyST does not resolve. Both change what the build produces, and both exist in the lecture-dp copies too.

8. Sequence the sweeps here before lecture-dp runs its own, and get the two proposed rules encoded upstream. This series is the sync source for the 31 shared filenames, so a fix landed in lecture-dp on one of the 6 byte-identical lectures is overwritten by the next sync; for the 25 diverged copies the sweeps have to run in both repos. And qe-math-010 (proposed) (item 4) and qe-writing-009 (proposed) (15 / 145, 27 occurrences) are scored here from the manual style guide but are not encoded in action-style-guide — they are proposed in issue #18. Sweeping 689 occurrences with nothing in CI to hold the line means measuring the same debt again next period.

9. Start with the four lowest: cross_product_trick (5.5, the corpus minimum), qr_decomp (5.9), navy_captain (6.0) and two_auctions (6.0). The first is byte-identical to its lecture-dp copy, so fixing it here clears both.

Lectures ranked by priority (lowest score first)

Scores are 0–10 per category; Overall is the mean of the in-scope categories, and Priority follows spec §4. A dash means the category is not applicable to that lecture. Click a lecture for its full report.

#	Lecture	Writing	Math	Code	Figures	References	Links	Admon	Overall	Priority
1	cross_product_trick	3.5	3	—	—	—	10	—	5.5	HIGH
2	qr_decomp	3	3	7.5	—	—	10	—	5.9	HIGH
3	navy_captain	3	6.5	7.5	3	—	10	—	6.0	HIGH
4	two_auctions	3	4.5	6.5	3	10	9	—	6.0	HIGH
5	likelihood_var	4.5	3.5	7.5	5	—	10	—	6.1	HIGH
6	prob_matrix	3	3	5.5	5	—	10	10	6.1	HIGH
7	two_computation	5.5	3	6	3	10	10	—	6.2	HIGH
8	rs_inventory_q	3	6.5	7.5	5	—	10	—	6.4	HIGH
9	linear_models	3	3	6	8	—	9	10	6.5	HIGH
10	multivariate_normal	3	3	7.5	5.5	—	10	10	6.5	HIGH
11	perm_income_cons	3	4	7.5	5.5	10	9	—	6.5	HIGH
12	likelihood_ratio_process	3	3	7	3.5	10	10	10	6.6	HIGH
13	lagrangian_lqdp	3	3	7	—	10	7.5	10	6.8	HIGH
14	prob_meaning	3	7.5	6	4.5	—	10	10	6.8	HIGH
15	ifp_advanced	3	3	7	7	8.5	10	10	6.9	HIGH
16	kalman_2	5	7.5	7.5	4.5	—	10	—	6.9	MEDIUM
17	linear_algebra	3	4.5	7.5	5.5	10	7.5	10	6.9	HIGH
18	markov_asset	3	4.5	7.5	6.5	8.5	8	10	6.9	HIGH
19	misspecified_recovery	3	3	5.5	6.5	10	10	10	6.9	HIGH
20	multi_hyper	3.5	6.5	7.5	7	—	10	—	6.9	HIGH
21	pandas_panel	3.5	—	7.5	4.5	—	9	10	6.9	HIGH
22	stats_examples	3	4.5	7	7	—	10	10	6.9	HIGH
23	ols	3	7	7.5	5	7.5	9	10	7.0	HIGH
24	perm_income	3	4	7.5	6	8.5	10	10	7.0	HIGH
25	phillips_lost_conquest	3.5	6	7.5	4.5	7.5	10	10	7.0	HIGH
26	wald_friedman_2	3	6	7.5	5	9	8.5	10	7.0	HIGH
27	finite_markov	3	3.5	8	6.5	10	9	10	7.1	HIGH
28	lln_clt	3.5	3	6.5	7	10	10	10	7.1	HIGH
29	lq_inventories	3	3	7.5	7	10	10	10	7.2	HIGH
30	mccall_model	3	7.5	6	6	10	8	10	7.2	HIGH
31	mle	4.5	3	7.5	5.5	10	10	10	7.2	HIGH
32	sargent_surico	6.5	5.5	4.5	4	10	10	10	7.2	HIGH
33	var_subsets	6.5	7	5	5	—	10	10	7.2	LOW
34	ge_arrow	3	3	7.5	7.5	10	10	10	7.3	HIGH
35	opt_transport	3	3	7	8	10	10	10	7.3	HIGH
36	pricing_information	5.5	4.5	5	6	10	10	10	7.3	LOW
37	util_rand_resp	4	4	7.5	9.5	9	10	—	7.3	HIGH
38	von_neumann_model	3	5	7.5	7	8.5	10	10	7.3	HIGH
39	affine_risk_prices	4.5	4	8.5	5.5	9	10	10	7.4	HIGH
40	blackwell_kihlstrom	3.5	3	8.5	7.5	9	10	10	7.4	HIGH
41	imp_sample	4.5	4	10	8.5	—	10	—	7.4	HIGH
42	likelihood_ratio_process_2	3	9.5	7.5	4	7.5	10	10	7.4	HIGH
43	markov_perf	4	5	7	6	10	10	10	7.4	HIGH
44	odu	3	9	7.5	5	9	8.5	10	7.4	HIGH
45	ross_recovery	3.5	5.5	6.5	6	10	10	10	7.4	HIGH
46	uncertainty_traps	3	5.5	7.5	6.5	9	10	10	7.4	HIGH
47	inventory_q	3	6	10	6	10	10	—	7.5	HIGH
48	lqcontrol	4.5	5	7.5	5.5	10	10	10	7.5	LOW
49	mccall_risk	6.5	6	5.5	7	—	10	10	7.5	LOW
50	measurement_models	3	4	6	9.5	10	10	10	7.5	HIGH
51	phillips_priors	4.5	7	8.5	5.5	7	10	10	7.5	LOW
52	samuelson	6.5	9.5	7.5	5	8.5	8	—	7.5	LOW
53	wald_friedman	3	8.5	7.5	4.5	10	9	10	7.5	HIGH
54	ak_aiyagari	5	10	8	4	8.5	10	—	7.6	HIGH
55	cass_fiscal	3	9	7.5	4	10	10	10	7.6	HIGH
56	hansen_singleton_1983	6	3	7	9.5	10	10	—	7.6	HIGH
57	olg_adaptive_money	3.5	7.5	6.5	6	10	10	10	7.6	HIGH
58	os_stochastic	3	7	7.5	7.5	10	8	10	7.6	HIGH
59	ar1_turningpts	3	7.5	8.5	8	9	10	—	7.7	HIGH
60	cass_koopmans_2	3	9.5	8.5	6	10	7	10	7.7	HIGH
61	kalman	5	4.5	8.5	7	9	10	10	7.7	LOW
62	long_run_risk_operator	3	7	7.5	6.5	10	10	10	7.7	HIGH
63	phillips_learning	5	7.5	8.5	6	7	10	10	7.7	LOW
64	phillips_two_stories	3	10	8.5	5	7.5	10	10	7.7	HIGH
65	back_prop	3	7.5	7.5	9	—	10	10	7.8	HIGH
66	ifp_opi	5	10	7	6	—	9	10	7.8	LOW
67	likelihood_bayes	5.5	4	7.5	7.5	10	10	10	7.8	HIGH
68	ls_learning	4.5	5	8.5	6.5	10	10	10	7.8	LOW
69	mccall_q	4	9.5	7	7	9	10	—	7.8	HIGH
70	newton_method	4	9.5	7	6	—	10	10	7.8	HIGH
71	os_numerical	4.5	9.5	7.5	5.5	—	10	10	7.8	LOW
72	re_with_feedback	3	8.5	7	6	10	10	10	7.8	HIGH
73	sir_model	4.5	7.5	8.5	8.5	—	10	—	7.8	LOW
74	svd_intro	3	9.5	7.5	6.5	—	10	10	7.8	HIGH
75	var_dmd	3	9.5	—	—	7.5	9	10	7.8	HIGH
76	eig_circulant	3	7.5	10	7	—	10	10	7.9	HIGH
77	exchangeable	3.5	7	7.5	7.5	10	10	10	7.9	HIGH
78	information_market_equilibrium	4	5.5	8.5	7.5	10	10	10	7.9	HIGH
79	kalman_filter_var	5.5	7.5	5.5	6.5	10	10	10	7.9	LOW
80	lake_model	5.5	6.5	7.5	5.5	10	10	10	7.9	LOW
81	learning_approximation	5.5	8	5.5	6.5	10	10	10	7.9	LOW
82	mccall_fitted_vfi	6	8	7	5.5	9	10	10	7.9	LOW
83	phillips_drifts_volatilities	3.5	9.5	8	4.5	10	10	10	7.9	HIGH
84	phillips_escaping_nash	5	9.5	8.5	4.5	7.5	10	10	7.9	LOW
85	rational_expectations	3.5	5	7.5	10	9	10	10	7.9	HIGH
86	robust_permanent_income	3.5	9	6.5	6	10	10	10	7.9	HIGH
87	aiyagari_egm	3	8.5	10	5.5	9	10	10	8.0	HIGH
88	cass_koopmans_1	3	9.5	8.5	6	10	9	10	8.0	HIGH
89	divergence_measures	4.5	6.5	8.5	6.5	10	10	10	8.0	LOW
90	hansen_singleton_1982	6	5	5.5	9.5	10	10	10	8.0	LOW
91	ifp_egm_transient_shocks	4	10	7.5	5.5	9	10	10	8.0	HIGH
92	jv	3	10	7.5	6.5	9	10	10	8.0	HIGH
93	marimon_mcgrattan_sargent	4	9.5	6.5	7.5	8.5	10	10	8.0	HIGH
94	market_diffusion	6.5	7	4.5	8	10	10	10	8.0	LOW
95	mccall_model_with_sep_markov	5.5	10	7	5.5	—	10	10	8.0	LOW
96	mix_model	5	9	7	7	—	10	10	8.0	LOW
97	theil_1	4.5	6.5	8.5	6.5	10	10	10	8.0	LOW
98	troubleshooting	6	—	—	9	—	9	—	8.0	LOW
99	ak2	3.5	10	8.5	5	10	10	10	8.1	HIGH
100	cass_fiscal_2	4	10	7.5	5.5	10	10	10	8.1	HIGH
101	chow_business_cycles	5.5	6.5	8.5	6.5	10	10	10	8.1	LOW
102	ifp_egm	3	9.5	7.5	6.5	10	10	10	8.1	HIGH
103	lq_robust_bewley	6.5	5.5	7.5	7.5	10	10	10	8.1	LOW
104	merging_of_opinions	4.5	7	8.5	7.5	9	10	10	8.1	LOW
105	os_time_iter	4.5	7.5	7.5	8	9	10	10	8.1	LOW
106	phillips_self_confirming	5.5	7	8.5	6	10	10	10	8.1	LOW
107	rand_resp	3.5	9.5	7.5	—	10	10	—	8.1	HIGH
108	rational_learning_re	5.5	7.5	7	7	10	10	10	8.1	LOW
109	theil_2	4.5	7.5	6.5	8.5	10	10	10	8.1	LOW
110	wealth_dynamics	3	9.5	8	6.5	10	10	10	8.1	HIGH
111	aiyagari	4.5	8.5	8.5	7	10	9	10	8.2	LOW
112	bayes_nonconj	5	10	8.5	6.5	—	9	10	8.2	LOW
113	endogenous_lake	5.5	7.5	8.5	6	10	10	10	8.2	LOW
114	mccall_persist_trans	6.5	6	7	8	10	10	10	8.2	LOW
115	survival_recursive_preferences	5	9.5	8.5	4.5	10	10	10	8.2	LOW
116	house_auction	3	10	7	—	10	10	10	8.3	HIGH
117	lq_permanent_income	5.5	7.5	8.5	6.5	10	10	10	8.3	LOW
118	hoist_failure	5.5	6.5	8.5	8	10	10	10	8.4	LOW
119	ifp_discrete	5.5	7	7.5	8.5	10	10	10	8.4	LOW
120	kesten_processes	5.5	5	10	8	10	10	10	8.4	LOW
121	lq_robust_smoothing	5.5	7.5	8.5	7.5	10	10	10	8.4	LOW
122	organization_capital	4.5	7	8.5	9	10	10	10	8.4	LOW
123	phillips_adaptive	5.5	9.5	8.5	6.5	8.5	10	10	8.4	LOW
124	phillips_credibility	5	9.5	9	7	8.5	10	10	8.4	LOW
125	phillips_misspecified	5.5	9.5	8.5	6.5	8.5	10	10	8.4	LOW
126	morris_learn	3	9.5	8.5	10	8.5	10	10	8.5	HIGH
127	os_egm	4	9.5	7	9	10	10	10	8.5	HIGH
128	phillips_credible_policies	6.5	9.5	8.5	6	10	9	10	8.5	LOW
129	ar1_bayes	7	10	7.5	8	10	8	10	8.6	NONE
130	bounded_rationality	5.5	10	8.5	6.5	10	10	10	8.6	NONE
131	career	6	9.5	8.5	6.5	10	10	10	8.6	NONE
132	exchange_rate_learning	5	9.5	10	5.5	10	10	10	8.6	NONE
133	lq_bewley_complete_markets	6	10	8.5	6	10	10	10	8.6	NONE
134	mccall_model_with_separation	7	8.5	6.5	8.5	10	10	10	8.6	NONE
135	os_egm_jax	8	10	6.5	7	—	10	10	8.6	NONE
136	unemployment_linear	6	8	8.5	8.5	9	10	10	8.6	NONE
137	genetic_classifier	6	10	8.5	7	10	10	10	8.8	NONE
138	harrison_kreps	6.5	8	8.5	—	10	10	10	8.8	NONE
139	os	5.5	9.5	8.5	8	10	10	10	8.8	NONE
140	prospects_bounded_rationality	6.5	10	7.5	8.5	10	10	—	8.8	NONE
141	inventory_dynamics	5.5	10	9.5	7	10	10	10	8.9	NONE
142	unemployment_shocks	6.5	9.5	7.5	10	—	10	10	8.9	NONE
143	status	10	—	9	—	—	10	—	9.7	NONE
144	intro	10	—	—	—	—	10	—	10.0	NONE
145	zreferences	10	—	—	—	—	10	—	10.0	NONE