Analysis catalog

The layered set of questions this data is built to answer — and gradeable against ground truth

Every question below is answerable from a generated arm’s visible/ folder and — this is the point of the whole project — gradeable: the generating mechanisms are recorded in a hidden answer key, and scenario arms are exact counterfactual twins of the baseline (see Theory), so both estimates and methods can be scored against ground truth.

Questions are grouped by analytics layer, in the order a real engagement would tackle them. Applicability across arms: every question in Layers 0–4 and 6 applies to every generated arm, not just the baseline — each scenario folder is a complete instance of the same world, with its own recording-layer defects, its own tax accounting, and its own full answer key. Layer 5 is the one exception: its questions are about arm pairs by construction. The three-year arms (basic.year = 3) additionally unlock Layer 7 — questions a single year structurally cannot ask: trend, churn, structural breaks, regime-change forecasting, and capital decisions.

Layer 0 — Clean the records before trusting them

The recording layer deliberately plants ten defect families in the paperwork; every find is gradeable row-by-row against the hidden imperfections log.

#	Question	Approach
0.1	Which receipts were uploaded twice, and what was revenue really?	the all-even multiplicity rule; reconcile deduped receipts to the ledger
0.2	Which negative till lines are cancelled mis-rings, and which are real refunds?	partition by reference receipt; match void partners within receipts
0.3	Do supplier invoices reconcile to the ledger’s procurement line?	exact-key dedup; explain the residual gap
0.4	Does book stock reconcile with the paperwork, night by night?	perpetual-inventory identity per SKU-day
0.5	What is “shrinkage” here, and what causes it?	decompose write-offs by reason; trace stock-count corrections
0.6	Standardize the label mess	payment-label normalization, hour-0 placeholder handling, category typos, sensor gaps
0.7	After cleaning: which totals should tie exactly, and which gaps are supposed to remain?	run the full reconciliation contract

Layer 1 — Describe the business

#	Question	Approach
1.1	Where does the money come from and go?	P&L from the ledger; margin waterfall including tax
1.2	When do people shop?	day-of-week / hour-of-day profiles — the trap: this is schedule composition, not day preference
1.3	What sells when?	monthly seasonality indices per category and product type
1.4	What does a basket look like?	basket size/value distributions, category co-occurrence
1.5	Who are the customers?	RFM segmentation of the card panel; regulars vs. one-off guests vs. anonymous cash
1.6	How are prices architected?	charm-ending mix, repricing cadence, markdown depth
1.7	How often are shelves empty, and what rots?	out-of-stock incidence by SKU-month; write-off seasonality
1.8	How much tax does the shop handle?	effective rate by category group; remittance seasonality

Layer 2 — Diagnose causes

#	Question	Approach
2.1	Does weather move the business?	regression on rain/temperature anomaly with calendar controls, HAC errors, next-day rebound lags
2.2	What hit costs during a shock?	event-study on invoice cost indices by category
2.3	How much of a cost shock reaches the shelf, and how fast?	pass-through regression of tags on invoice costs; hysteresis/asymmetry tests
2.4	Does raising a price lose customers or just move them?	SKU-level IV elasticity vs. category-level two-way FE — substitution vs. walkaway
2.5	Is my instrument valid?	test a macro shock as an instrument (it also squeezes budgets → invalid) vs. an idiosyncratic cost event (valid)
2.6	Did the markdowns work?	naive lift vs. DiD vs. selection-aware estimates
2.7	Why does food rot faster some weeks?	write-off regression on temperature and the shock timeline
2.8	Can I trust card data to represent everyone?	test the missingness mechanism between card and cash baskets
2.9	What drives refunds and shrinkage?	rate modeling over time/category
2.10	Do pre-holiday days really spike?	small-N inference done honestly (a handful of days per year)

Layer 3 — Predict

#	Question	Approach
3.1	What will next week sell?	category demand forecasting vs. a seasonal-naive benchmark
3.2	What would demand be if shelves never emptied?	censored-demand modeling: impute stockout-lost sales
3.3	How wrong is the owner’s own forecast, and why?	autopsy the trailing-average rule; quantify the censoring spiral
3.4	Which SKUs will stock out next week?	out-of-stock risk classification from cover, seasonality, delivery cadence
3.5	Which regulars are sliding into trouble?	early-warning detection of sustained down-trading spells
3.6	How far can one year of history be trusted?	quantify the single-seasonal-cycle limit — priced empirically against the three-year arm

Layer 4 — Prescribe

#	Question	Approach
4.1	What is better analytics worth, in euros?	rebuild the ordering policy on de-censored demand; close the believed/realized/oracle gap
4.2	How should perishables be ordered?	spoilage-aware cover per category (waste vs. stockout trade-off)
4.3	Where is margin safely adjustable?	category-level repricing using elasticity structure
4.4	What should be delisted or added?	dead-SKU economics: appeal, storage, listing fees
4.5	When should promotions run, on what, how deep?	redesign the markdown trigger with selection bias removed
4.6	Should I hire and extend hours?	cost-benefit of recovered demand vs. employer-priced labor
4.7	How much cash must the till hold?	working-capital policy: credit line, tax remittance timing

Layer 5 — The policy laboratory (counterfactuals)

Each scenario arm is a CRN twin: the arm-vs-baseline difference is the causal effect with zero sampling error between arms. The deepest exercise here is method validation: estimate the effect observationally inside one arm, then check yourself against the twin difference.

#	Question	What the twin diff reveals
5.1	Who actually bore a tax cut — customers or the owner?	tax incidence split; pass-through speed and completeness
5.2	What did households do with a rebate?	an emergent marginal propensity to consume
5.3	What does a broad supply shock do to a grocer?	revenue up, units down — demand destruction decomposed by category resilience
5.4	What does a storm cost, net of the catch-up?	intertemporal substitution via pantries
5.5	Does my observational elasticity generalize?	external validity: predict a twin’s outcome from baseline-estimated structure
5.6	Was hiring a clerk worth it?	full-cost staffing counterfactual, payroll tax included

Layer 6 — Advanced and structural

#	Question	Approach
6.1	What do customers want?	discrete-choice (conditional logit) estimation on the card panel
6.2	Can partial pooling beat per-SKU noise?	hierarchical Bayesian demand models
6.3	What drives sales, decomposed?	marketing-mix modeling: base, seasonality, weather, price, promotions
6.4	Where did the missing demand go?	structural four-cause decomposition (closed / stockout / budget / outside)
6.5	How much business is passing trade?	mixture modeling of the token panel: regulars vs. single-use guests
6.6	Is the documented causal graph consistent with the data?	test the DAG’s implied conditional independencies

Layer 7 — The three-year arc (time, churn, and capital)

Asked of the three-year baseline, graded against its hidden answer key and its own CRN twins. These are the questions one year of data structurally cannot ask.

#	Question	Approach
7.1	Is the business growing, or is it just summer?	trend–seasonality decomposition, now identified with three annual cycles
7.2	Which customers left, which arrived, and who was never going to stay?	churn inference from token silence, cohort retention, survival modeling
7.3	What happened when a competitor entered?	structural-break detection, then who defected — and the twin that prices it
7.4	Did customers trade up after the discounter opened?	composition vs. behavior decomposition
7.5	Does a model trained on years one–two survive year three?	true holdout-year evaluation; forecast breakdown at a regime change
7.6	Did the promotions work?	naive lift vs. entry-aware estimates — the endogenous-response confound
7.7	What do the repeated cost-shock episodes have in common?	compare pass-through across episodes with different competitive regimes
7.8	What did a one-off event (equipment failure, a local surge) cost or earn?	event studies on narrated one-offs
7.9	Was the expansion a good investment?	capital budgeting from the books — NPV the bet, graded against its own twin
7.10	How does capital actually flow through a small shop?	financial-statement analysis of the widened cost sheet
7.11	Renew the lease or close?	the capstone prescriptive: decompose the result into its causes and project forward