Activities - EDA with Python, Pandas and Jupyter

Activities in EDA with Python, Pandas and Jupyter

All activities across workflows in this playbook

Activities 43

Name	Abbrev.	Workflow	Phase	Actions
Choose the analysis type	FPM-1	Frame the Problem	Frame the Problem
Formulate the driving questions	FPM-2	Frame the Problem	Frame the Problem	Read the 'FPM-2 Formulate the driving questions' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Frame the Problem'). Follow its guidance when assisting me. Write down the specific descriptive question(s) first ("what is happening / what is causing this"), then the predictive/prescriptive question(s) that follow from them ("what do we want to happen / what actions get us there"). Keep the questions in front of you for the rest of the playbook — every later chart and model exists to answer one of them.
Assess analyst readiness	FPM-3	Frame the Problem	Frame the Problem	Read the 'FPM-3 Assess analyst readiness' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Frame the Problem'). Follow its guidance when assisting me. Before opening any data, check you're bringing genuine interest/curiosity, skepticism, methodical practice, pattern-spotting, and enough domain familiarity to know what "normal" looks like for this data. Also plan to alternate between the statistician, programmer, and domain-expert "hats" as the work proceeds.
Select candidate datasets from the datalake catalog	FPM-4	Frame the Problem	Frame the Problem	Read the 'FPM-4 Select candidate datasets from the datalake catalog' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Frame the Problem'). Follow its guidance when assisting me. State what the "ideal" dataset for your questions would look like, then browse the provided datalake catalog (a Phainestai-style dataset catalog) for the closest available match(es), accepting a smaller/less feature-rich substitute if the ideal isn't present. Record which catalog entries you selected and why. This step replaces raw web/DB/file acquisition with selection from a pre-built catalog, since this playbook assumes the datalake already exists.
Assess dataset quality against meaningfulness traits	FPM-5	Frame the Problem	Frame the Problem	Read the 'FPM-5 Assess dataset quality against meaningfulness traits' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Frame the Problem'). Follow its guidance when assisting me. For each selected catalog entry, check it against the traits of meaningful data: sufficient volume, historical depth, consistency over time, multivariate coverage, atomic-level (not pre-aggregated) records, clean values, clear coding, dimensional structure, rich segmentation, and known pedigree/lineage. Drop or flag entries that fail badly on a trait that matters for your questions.
Write the Framing Brief	FPM-6	Frame the Problem	Frame the Problem	Read the 'FPM-6 Write the Framing Brief' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Frame the Problem'). Follow its guidance when assisting me. Consolidate the chosen analysis type, the driving questions, and the vetted dataset selection into a single short brief. This is the contract the rest of the playbook works against — Phase 2 starts from it, and Phase 2's final conclusions get checked back against it.
Load selected datasets into DataFrames	ODA-1	Organize and Clean Data	Explore Data and Find Signal	Read the 'ODA-1 Load selected datasets into DataFrames' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Organize and Clean Data'). Follow its guidance when assisting me. Read each selected catalog dataset into a pandas Series/DataFrame using the appropriate reader (read_csv with explicit sep/header/index_col/na_values, read_json, read_fwf, or chunked read_csv(chunksize=) for very large extracts). Repair obviously sparse/irregular files (supply explicit `names=`, then `ffill()` a sparse category column before `set_index()`).
Inspect structure	ODA-2	Organize and Clean Data	Explore Data and Find Signal	Read the 'ODA-2 Inspect structure' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Organize and Clean Data'). Follow its guidance when assisting me. Immediately after loading, check `.head()`/`.tail()`, `len()`, `.shape`, `.dtypes`, `.index`, `.columns` (and `.ndim` for any raw numpy arrays). This is the standard first look that catches wrong delimiters, misparsed headers, or unexpected types before you build anything on top of them.
Handle missing and invalid data	ODA-3	Organize and Clean Data	Explore Data and Find Signal	Read the 'ODA-3 Handle missing and invalid data' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Organize and Clean Data'). Follow its guidance when assisting me. Detect gaps with `isnull()`/`notnull()`. Then choose deletion (`dropna()`, whole rows/columns only), imputation (constant/mean via masked assignment, or `fillna(method="ffill"/"bfill")`), selective replacement (`replace()`), or backfill from a secondary source (`combine_first()`) — the right choice depends on domain knowledge about what a gap means for that variable, not a default rule.
Combine and deduplicate datasets	ODA-4	Organize and Clean Data	Explore Data and Find Signal	Read the 'ODA-4 Combine and deduplicate datasets' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Organize and Clean Data'). Follow its guidance when assisting me. If two selected datasets are complementary (different variables, shared key), `pd.merge()`/`.join()` them; if they're similar-schema (same variables, e.g. across time periods or regions), `pd.concat()` them. After any vertical concat, check for and resolve duplicate index labels or rows (`duplicated()`, `drop_duplicates()`).
Reshape data for the framing questions	ODA-5	Organize and Clean Data	Explore Data and Find Signal	Read the 'ODA-5 Reshape data for the framing questions' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Organize and Clean Data'). Follow its guidance when assisting me. Convert between long ("tidy") and wide ("square") layouts with `stack()`/`unstack()`/`pivot()`, and build a hierarchical (multi-index) structure if your questions need more than one categorical grouping key (e.g. State + Year). Choose the shape each later step actually needs rather than leaving everything in whatever shape it was loaded in.
Compute descriptive statistics per candidate variable	EFS-1	Explore and Engineer Features	Explore Data and Find Signal	Read the 'EFS-1 Compute descriptive statistics per candidate variable' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Explore and Engineer Features'). Follow its guidance when assisting me. For every candidate variable, compute sum/mean/median/std/count/min/max (with `skipna`), sort/rank it, and look at cumulative/period-over-period measures (`cumsum`, `diff`) and uniqueness/value_counts for categoricals. This is the "smell and sense of the data" step before you commit to any Y or feature.
Choose candidate target variable(s) Y	EFS-2	Explore and Engineer Features	Explore Data and Find Signal	Read the 'EFS-2 Choose candidate target variable(s) Y' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Explore and Engineer Features'). Follow its guidance when assisting me. From the Framing Brief's questions, pick the concrete column(s) that would answer them if you understood what drove them — this is Y. Note whether Y is continuous (→ regression), categorical/binary (→ classification), or whether there is no single Y and the task is genuinely unsupervised (→ clustering, handled in ACE).
Engineer and derive features	EFS-3	Explore and Engineer Features	Explore Data and Find Signal	Read the 'EFS-3 Engineer and derive features' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Explore and Engineer Features'). Follow its guidance when assisting me. Build the feature table: arithmetic transforms across columns (index-aligned), `groupby()` aggregation to compare Y across categorical breakdowns, `cut()`/`qcut()` to discretize a continuous variable, `get_dummies()` to encode categoricals for models that need numeric input, and `series.map(func)` for one-off custom transforms.
Cross-tabulate categorical candidates against Y	EFS-4	Explore and Engineer Features	Explore Data and Find Signal	Read the 'EFS-4 Cross-tabulate categorical candidates against Y' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Explore and Engineer Features'). Follow its guidance when assisting me. Run `pd.crosstab(feature, Y, margins=True)` for each categorical feature candidate as a cheap first-pass independence check before committing to the heavier correlation/modeling work in the next workflow.
Compute correlation/covariance and significance between features and Y	EFS-5	Explore and Engineer Features	Explore Data and Find Signal	Read the 'EFS-5 Compute correlation/covariance and significance between features and Y' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Explore and Engineer Features'). Follow its guidance when assisting me. Compute pairwise correlation/covariance (`frame.corr()`, `.cov()`, `.corrwith(Y)`) across the feature table, then get significance via `scipy.stats.pearsonr()` since pandas doesn't supply p-values — align mismatched indexes manually first if needed. This produces the ranked candidate feature–Y relationships that Phase 2's ACE workflow will validate.
Interpret correlation with p-value and rule out spurious relationships	ACE-1	Assess Information Gain and Conclude	Explore Data and Find Signal	Read the 'ACE-1 Interpret correlation with p-value and rule out spurious relationships' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Assess Information Gain and Conclude'). Follow its guidance when assisting me. Cross each correlation's sign/magnitude against its p-value (p≤.01 vs not) before treating it as evidence. Remember correlation is not causation — a real correlation may share a hidden confounder — and a non-significant linear correlation doesn't rule out a non-linear relationship.
Design and run a predictive experiment	ACE-2	Assess Information Gain and Conclude	Explore Data and Find Signal	Read the 'ACE-2 Design and run a predictive experiment' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Assess Information Gain and Conclude'). Follow its guidance when assisting me. Split the feature table 70:30 train/test (seed the RNG first). Choose the model by Y's type (per the Target Variable (Y) Definition from EFS): continuous Y → OLS `LinearRegression`, or `Ridge` if predictors are collinear; categorical/binary Y → `LogisticRegression` (use `cut()` first to quantize a continuous proxy into labels if needed, and set `C` well above the default 1.0, e.g. ~20). Fit on training data only, engineering nonlinear predictors (squares/logs/interaction terms) first if a plain linear fit under-captures curvature, then apply the fitted model to the held-out test set. Never touch the test data while fitting.
Evaluate model quality	ACE-3	Assess Information Gain and Conclude	Explore Data and Find Signal	Read the 'ACE-3 Evaluate model quality' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Assess Information Gain and Conclude'). Follow its guidance when assisting me. Build a confusion matrix (classification) or compare predicted-vs-actual on train and test sets (regression); derive accuracy/precision/sensitivity/specificity and choose which one matters for the question at hand. Compare train-set fit against test-set fit to judge overfitting before trusting the model.
Use clustering as an alternative when correlation is weak	ACE-4	Assess Information Gain and Conclude	Explore Data and Find Signal	Read the 'ACE-4 Use clustering as an alternative when correlation is weak' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Assess Information Gain and Conclude'). Follow its guidance when assisting me. When linear correlation is weak or absent but you suspect structure, scale the features (`sklearn.preprocessing.scale()`), run k-means (`KMeans(n_clusters=k)`, k chosen by judgment), and assign human-meaningful labels to the resulting clusters — cluster IDs alone mean nothing.
Rank features by information gain and draw conclusions	ACE-5	Assess Information Gain and Conclude	Explore Data and Find Signal	Read the 'ACE-5 Rank features by information gain and draw conclusions' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Assess Information Gain and Conclude'). Follow its guidance when assisting me. Fit a Random Forest Regressor on the feature table against Y to get a feature-importance ranking (an ensemble-of-trees estimate of how much each feature actually informs Y), cross-check it against the regression coefficients and correlations from earlier steps, and write the surviving, validated findings as the phase's Conclusions.
Identify the analytical pattern type per conclusion	ECR-1	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-1 Identify the analytical pattern type per conclusion' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. For each conclusion from Phase 2, classify it as time-series, part-to-whole/ranking, deviation, distribution, correlation, or multivariate before choosing anything about how to draw it — this single classification drives every remaining choice in this workflow and the next.
Select the optimal quantitative scale	ECR-2	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-2 Select the optimal quantitative scale' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. Bar graphs must start their scale at zero (never narrow it — that's a visual lie); every other chart type (line, scatter, dot plot, box plot) should start a little below the lowest value and end a little above the highest. Begin/end/interval values should all be round numbers.
Add reference lines and regions	ECR-3	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-3 Add reference lines and regions' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. Whenever a value must be judged against a threshold or norm (budget, target, mean±SD, percentile band), add a labeled reference line or shaded region instead of leaving the comparison to be computed mentally. Set the reference to 0 / 0% / 100% to match whether the deviation is absolute, a percentage difference, or a percentage of the reference.
Choose pre-attentive encodings by precision hierarchy	ECR-4	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-4 Choose pre-attentive encodings by precision hierarchy' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. Reserve length and 2-D position — the only two pre-attentive attributes perceived with real quantitative precision — for the values that must be compared precisely. Use hue or shape (the best categorical discriminators) to separate groups, and treat size/color-intensity/blur as secondary/tertiary encodings only.
Plan trellis or multi-view layout	ECR-5	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-5 Plan trellis or multi-view layout' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. If one chart per conclusion would be cluttered, or you need to compare shapes across many groups, plan a trellis/small-multiples grid (one categorical variable) or a visual crosstab (two categorical variables): every panel must share the same chart type, size and scale, sequenced by magnitude rather than alphabetically. For open-ended exploration, plan tightly-coupled multiple views with shared filtering/brushing instead of one fixed chart.
Reduce over-plotting	ECR-6	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-6 Reduce over-plotting' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. If points visibly collide in a scatter or dense line chart, apply remedies in order of increasing intervention: shrink point size → remove fill (outline only) → change shape → jitter → add transparency → encode density (contours/heatmap cells) → reduce the data itself (aggregate, filter, trellis, or sample). Stop at the mildest remedy that solves the problem.
Check for spurious or missed patterns	ECR-7	Encode and Reduce Clutter	Design the Visual Story	Read the 'ECR-7 Check for spurious or missed patterns' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Encode and Reduce Clutter'). Follow its guidance when assisting me. Before locking in the encoding plan, verify every apparent pattern against the underlying detail data, and deliberately look again with fresh eyes for a real pattern you may have missed. This is the QA gate before any chart gets built.
Design a time-series display	CTE-1	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-1 Design a time-series display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. Default to a line graph for continuous change through time; use bars for period-to-period comparison, a dot plot (never connected by a line) for irregularly-spaced samples, box plots to show a distribution shifting over time, and radar/cycle plots or heatmaps for cyclical/high-volume data. Smooth volatility with a running average rather than a single trend line, use a log scale or % change when comparing growth rates of very different magnitude, and never render a missing value as zero — break the line instead.
Design a part-to-whole/ranking display	CTE-2	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-2 Design a part-to-whole/ranking display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. Sort items by value and use a bar graph (never a pie chart) for ranking/part-to-whole comparison; switch to a dot plot when the scale must be narrowed off zero; use a Pareto chart (bars + cumulative-% line) for 80/20 analysis; re-express with √ (mildest), then log, then inverse when values span a huge range; use a line graph only to show rank changes over time (a bumps chart), never a single ranking snapshot.
Design a deviation display	CTE-3	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-3 Design a deviation display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. Pair a bar or line graph with a 0 / 0% / 100% reference line depending on whether the deviation (actual vs. budget/target/forecast/prior period/peer) is shown as an absolute difference, a percentage difference, or a percentage of the reference value. Use bars to emphasize individual-period deviations and lines when the overall shape of deviation through time matters more.
Design a distribution display	CTE-4	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-4 Design a distribution display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. For a single distribution use a histogram (shape+magnitude, consistent bin widths, bin count tuned by trial and error), a frequency polygon (shape only), a strip plot (every raw value, small N), or a stem-and-leaf plot (≤100 values, quick by hand). For comparing several distributions, box plots beat everything else — add outlier dots beyond the whiskers rather than stretching them. Prefer median/percentiles over mean/SD when outliers are present.
Design a correlation display	CTE-5	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-5 Design a correlation display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. Use a scatterplot with a square aspect ratio, both axes trimmed near min/max, for two variables; add a trend line matching the actual shape (linear/logarithmic/exponential/low-order polynomial — never an overfit high-order polynomial), with a separate trend line per subgroup if slopes visibly differ. For 3+ variables use a scatterplot matrix, table lens, or parallel-coordinates plot; de-clutter dense scatterplots (transparency/jitter) and add grid lines when comparing several side by side.
Design a multivariate display	CTE-6	Choose Chart per Analysis Type	Design the Visual Story	Read the 'CTE-6 Design a multivariate display' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Choose Chart per Analysis Type'). Follow its guidance when assisting me. Avoid glyph-based displays (Chernoff faces, stars, whiskers) — no demonstrated real-world effectiveness. Prefer a parallel-coordinates plot with brushing/filtering, or a colorblind-safe diverging-color heatmap (never red/green, never black as the "near zero" color), to compare multivariate profiles. Use a clustering/similarity-ranking algorithm, not eyes alone, to group items, then visualize the groups (e.g. in a trellis).
Sequence the narrative	SNK-1	Storyboard the Notebook	Design the Visual Story
Draft one storyboard entry per chart	SNK-2	Storyboard the Notebook	Design the Visual Story	Read the 'SNK-2 Draft one storyboard entry per chart' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Storyboard the Notebook'). Follow its guidance when assisting me. For every chart in the sequence, write down: chart type (from CTE), axis/scale choice and reference lines (from ECR), any trellis/layout decision, and the one-sentence insight/conclusion it must support. This entry-per-chart table is the literal blueprint Phase 4 codes against.
Review the storyboard against conclusions	SNK-3	Storyboard the Notebook	Design the Visual Story	Read the 'SNK-3 Review the storyboard against conclusions' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Storyboard the Notebook'). Follow its guidance when assisting me. Check the storyboard against Phase 2's Conclusions Summary: does every conclusion that matters have a designated chart, and does every planned chart trace back to a real conclusion (not just an interesting-looking variable)? Fix gaps before writing any Plotly code.
Set up the Plotly/Jupyter environment	IVS-1	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-1 Set up the Plotly/Jupyter environment' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Import `plotly.express as px` and `plotly.graph_objects as go`, set the notebook renderer (`pio.renderers.default = "notebook"` or "notebook_connected"), and decide once whether figures need to be interactive (keep Plotly) or also exported statically (`kaleido` for PNG/PDF export).
Implement each storyboard entry as a Plotly figure	IVS-2	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-2 Implement each storyboard entry as a Plotly figure' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Map each storyboard chart type to Plotly: time-series → `px.line`/`go.Scatter(mode="lines")`; part-to-whole/ranking → `px.bar` (sorted, never `px.pie`) or `px.scatter` for a dot plot; deviation → `px.bar`/`px.line` plus a reference shape; distribution → `px.histogram`, `px.box`, or `px.strip`; correlation → `px.scatter` with `trendline="ols"/"lowess"`; multivariate → `px.imshow` (diverging colorscale) or `px.parallel_coordinates`. Use the exact chart type the storyboard specifies — do not substitute a "nicer-looking" one.
Apply scale/reference-line/annotation rules via layout methods	IVS-3	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-3 Apply scale/reference-line/annotation rules via layout methods' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Enforce ECR's decisions in code: `fig.update_yaxes(rangemode="tozero")` for bar charts, explicit `range=[min,max]` with round numbers for line/scatter/box; `fig.add_hline()`/`add_vline()`/`add_shape()` for reference lines/regions and target bands; `fig.update_traces(marker=dict(opacity=...))` or `size=` for over-plotting fixes (jitter via manual x/y offset, density via `px.density_heatmap`).
Build trellis/small-multiples and multi-view linking	IVS-4	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-4 Build trellis/small-multiples and multi-view linking' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Use `facet_col=`/`facet_row=` (with `facet_col_wrap=`) in plotly.express for trellis/visual-crosstab layouts, keeping `category_orders=` set to sort panels by magnitude rather than alphabetically; use `plotly.subplots.make_subplots` for hand-built multi-panel layouts; for linked brushing/filtering across views, use `go.FigureWidget` in the notebook or export to Dash if true cross-filtering is required.
Assemble the Jupyter storybook notebook	IVS-5	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-5 Assemble the Jupyter storybook notebook' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Interleave a markdown cell (the question/insight from the storyboard) before each code cell that renders its Plotly figure, in the sequence fixed by SNK. Keep one figure's setup self-contained per cell group so the notebook can be read top-to-bottom as the story, not just executed.
Review and export the final notebook/report	IVS-6	Implement Plotly Visuals	Build the Story in Plotly	Read the 'IVS-6 Review and export the final notebook/report' activity from playbook 'EDA with Python, Pandas and Jupyter' (workflow 'Implement Plotly Visuals'). Follow its guidance when assisting me. Re-run the notebook top-to-bottom on a clean kernel to confirm every cell executes and every chart still matches its storyboard entry and conclusion. Export to HTML (`jupyter nbconvert --to html`) or PDF for the data sponsor, alongside the notebook itself, per the Framing Brief's audience.