causaldemand 1.0.0__tar.gz

causaldemand-1.0.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

causaldemand-1.0.0/MANIFEST.in

@@ -0,0 +1,2 @@
+# Files of the source distribution beyond the packages (setuptools adds the README, the license and tests/test*.py).
+include tests/conftest.py

causaldemand-1.0.0/PKG-INFO

@@ -0,0 +1,297 @@
+Metadata-Version: 2.4
+Name: causaldemand
+Version: 1.0.0
+Summary: CausalDemand: synthetic retail sales datasets for testing whether a method predicts how sales respond to price changes. Download, scoring and regeneration tools.
+Author: Juwon Hong, Minha Hwang, Venkatesh Shankar
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/jean-jsj/CausalDemand
+Project-URL: Repository, https://github.com/jean-jsj/CausalDemand
+Project-URL: Dataset, https://huggingface.co/datasets/jean-jsj/CausalDemand
+Keywords: benchmark,causal inference,counterfactual prediction,demand estimation,price elasticity,retail,synthetic data
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering
+Classifier: Topic :: Scientific/Engineering :: Information Analysis
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: pandas>=2.0
+Requires-Dist: scipy>=1.13
+Requires-Dist: pyarrow>=14
+Requires-Dist: PyYAML>=6.0
+Requires-Dist: huggingface_hub>=0.23
+Provides-Extra: generate
+Requires-Dist: numpy==1.24.3; python_version < "3.11" and extra == "generate"
+Requires-Dist: pandas==2.3.3; python_version < "3.11" and extra == "generate"
+Requires-Dist: scipy==1.13.1; python_version < "3.11" and extra == "generate"
+Requires-Dist: statsmodels==0.14.6; python_version < "3.11" and extra == "generate"
+Requires-Dist: pyarrow==21.0.0; python_version < "3.11" and extra == "generate"
+Requires-Dist: PyYAML==6.0.3; python_version < "3.11" and extra == "generate"
+Requires-Dist: scikit-learn==1.6.1; python_version < "3.11" and extra == "generate"
+Requires-Dist: matplotlib==3.9.4; python_version < "3.11" and extra == "generate"
+Requires-Dist: xlrd==2.0.2; python_version < "3.11" and extra == "generate"
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Dynamic: license-file
+
+# CausalDemand
+
+CausalDemand is a benchmark for methods that predict how retail sales respond to a change in price. It tests whether
+a method that forecasts sales well also gets the effect of price right when prices are confounded: in half of the
+datasets, stores deepen discounts when an unobserved demand shock is high, so deep discounts and high sales coincide
+for a reason other than the discount, and a model that learns the price effect from this co-movement overstates it. A
+held-out forecast cannot reveal the error, because the held-out weeks follow the same pricing. The generator records
+every random quantity it draws, so it computes what sales would have been at prices that were never charged: in closed
+form for the log-linear model, and for the discrete-choice model by averaging over 10,000 simulated shoppers per store
+and week, the same shoppers at the actual and at the changed price.
+
+The datasets are synthetic weekly sales of 40 facial-tissue products in 731 stores over 156 weeks, from a generator
+calibrated on licensed store sales records. 20 further datasets repeat the design for a second product category,
+yogurt.
+
+| | |
+|---|---|
+| Datasets | Hugging Face: <https://huggingface.co/datasets/jean-jsj/CausalDemand> (DOI [10.57967/hf/10744](https://doi.org/10.57967/hf/10744)) |
+| Code | this repository, release [v1.0](https://github.com/jean-jsj/CausalDemand/releases/tag/v1.0) |
+| Paper | "CausalDemand: Benchmarking Causal Recovery of Demand from Retail Transactions and Product Text" (KDD 2027 Datasets and Benchmarks Track, submitted) |
+
+## Datasets
+
+Each dataset is one **configuration**, a demand model with confounding off or on, generated with one of five seeds:
+2 demand models x 2 confounding settings x 5 seeds = 20 datasets, named `<configuration>_seed<NNN>`.
+
+| Configuration | Demand model | Confounding | Seeds |
+|---|---|---|---|
+| `loglinear_confounding_off` | log-linear | off | 1, 10, 20, 30, 40 |
+| `loglinear_confounding_on` | log-linear | on | 1, 10, 20, 30, 40 |
+| `discretechoice_confounding_off` | discrete-choice | off | 1, 10, 20, 30, 40 |
+| `discretechoice_confounding_on` | discrete-choice | on | 1, 10, 20, 30, 40 |
+
+- **Log-linear model:** expected weekly units are a log-linear function of the product's regular price and discount,
+  the prices and discounts of the other products (more strongly for products that are closer in a product distance
+  computed from the product descriptions), and demand terms for product, brand, store, region and week.
+- **Discrete-choice model:** in each store and week, simulated households decide whether to buy and which stocked
+  product to buy; their tastes are correlated more strongly across closer products.
+- **Confounding on:** the discount of a promoted week responds to the unobserved demand shock (loading 0.30).
+  **Confounding off:** it does not. The two datasets of a demand model and seed are generated from the same random
+  draws and differ in the discounts of promoted weeks and in what follows from them.
+
+20 further **dose-response datasets** repeat the confounding-on configurations at confounding strengths 0.15 and 0.45.
+
+20 **yogurt datasets** repeat the four configurations and five seeds for a second product category: 100 yogurt
+products in 721 stores over the same 156 weeks, generated by the same generator and calibrated on yogurt sales
+records. Beyond that, they differ from the facial-tissue datasets in four main ways: the confounding strength with
+confounding on is 0.42 (log-linear) and 0.36 (discrete-choice); the product distance is estimated from the products
+between which households switch, not from product descriptions; the product table holds the pack size and catalogue
+attributes instead of a description; and each market's seasonal pattern follows its own yogurt sales instead of
+influenza activity. They are named `yogurt_<configuration>_seed<NNN>` and lie in `yogurt/` on Hugging Face, outside
+the dataset subsets, with the predictions and run records of their reference-model runs in `runs/yogurt/`.
+[`docs/DATA.md`](docs/DATA.md) describes every file and column (yogurt: Section 11).
+
+## Install
+
+In a copy of this repository (Python 3.9 to 3.12 and 3.14 are tested):
+
+```
+pip install .
+```
+
+From PyPI: `pip install causaldemand`; from GitHub: `pip install git+https://github.com/jean-jsj/CausalDemand`. Either installs the `causaldemand` command and the scorer.
+Regenerating datasets and running the reference models need their own environments ([`locks/README.md`](locks/README.md)).
+
+## Quickstart
+
+Run the commands from the repository root. Every command that reads released files downloads them from Hugging
+Face, or reads a local copy of the release with `--source <folder>`. An output folder that a command creates
+(`causaldemand_data/`, `scores/` and so on) holds a `.gitignore` file, so git ignores it.
+
+1. **Download a dataset** with its ground truth (776 MB of CSV):
+
+   ```
+   causaldemand download loglinear_confounding_on_seed001 --ground-truth
+   ```
+
+   This writes `causaldemand_data/loglinear_confounding_on_seed001/`: `public/` holds the files a method may use
+   (training sales, the prices of the 16 evaluation weeks, the 32 counterfactual scenarios, products and stores),
+   `ground_truth/` the expected units the scores compare with.
+
+2. **Predict.** Fit a method on the 140 training weeks and write two files from the same fitted model:
+   `forecast.csv` (`product_id, store_id, week, predicted_units`, for every row of
+   `public/transactions_holdout_context_public.csv`) and `scenarios.csv` (`intervention_id, product_id, store_id,
+   week, predicted_delta_units`, the predicted change in units under each scenario, for every row of
+   `public/counterfactual_sweep_context_panel.csv` whose `intervention_price` differs from its `baseline_price`).
+   [`notebooks/quickstart.ipynb`](notebooks/quickstart.ipynb) writes both files for a trivial method in a few lines,
+   into `notebooks/quickstart/` (see [Notebooks](#notebooks) for how to run it).
+
+3. **Score** the two files, here those of the notebook:
+
+   ```
+   causaldemand score --dataset causaldemand_data/loglinear_confounding_on_seed001 --forecast notebooks/quickstart/forecast.csv --scenarios notebooks/quickstart/scenarios.csv --out scores
+   ```
+
+   For the trivial method of the notebook (each pair's mean weekly training units, no response to any price change):
+   forecast error 0.4625, counterfactual error 0.1384, mean absolute counterfactual bias 0.9934, share of scenarios
+   with an overstated response 0, status valid.
+
+`causaldemand --help` lists every command (download, score, elasticity, rescore, generate).
+
+## Notebooks
+
+- [`notebooks/quickstart.ipynb`](notebooks/quickstart.ipynb) runs steps 1 to 3 of the quickstart for a trivial
+  method.
+- [`notebooks/full_run.ipynb`](notebooks/full_run.ipynb) runs a reference model of the paper (LightGBM) on the same
+  dataset, compares its scores with the released run, and rebuilds the default-settings table from the released
+  predictions. It runs the model with the CPU-model environment: create it as `.venv-models` in the repository root
+  ([`locks/README.md`](locks/README.md); on macOS, LightGBM also needs the OpenMP runtime `libomp`), or set the
+  environment variable `CAUSALDEMAND_MODEL_PYTHON` to the Python of that environment.
+
+Both read the dataset from `causaldemand_data/` (step 1), downloading it there if it is missing, and write their
+output into `notebooks/quickstart/` and `notebooks/full_run/`. To read a local copy of the release instead of Hugging
+Face, set the environment variable `CAUSALDEMAND_SOURCE` to its folder. The package does not install Jupyter; to run
+a notebook from the command line, in the package environment:
+
+```
+pip install nbconvert ipykernel
+jupyter nbconvert --to notebook --execute --output-dir notebooks/executed notebooks/quickstart.ipynb
+```
+
+This writes the executed copy to `notebooks/executed/`. The quickstart takes about 15 seconds plus the download,
+`full_run.ipynb` about 2 minutes.
+
+## Reproduce the paper
+
+From the repository root:
+
+```
+causaldemand rescore --out rescored --compare-with results
+```
+
+This re-scores the released predictions of 560 reference-model runs on the facial-tissue datasets (280 CPU runs and
+280 foundation-model runs) and compares every score with the archived one in `results/` (about 3 minutes, plus a
+download of about 1.8 GB). The 320 runs on the yogurt datasets (CPU and foundation models) are re-scored with
+`causaldemand rescore --runs yogurt --out rescored --compare-with results` (about 3.5 minutes, plus a download of about
+2.3 GB). [`docs/REPRODUCE.md`](docs/REPRODUCE.md) gives four tiers:
+rescoring the released predictions, rerunning the CPU models, regenerating a dataset from the calibration bundle, and
+rerunning the foundation models on a GPU.
+
+## Tasks and scores
+
+A method is fitted on the 140-week training window of a dataset, using only that dataset's input files, and evaluated
+on the following 16 weeks in two tasks.
+
+- **Sales forecasting:** predict the units `q̂` of every product-store-week of the evaluation file at the actual
+  prices. Score: the **forecast error** `Σ |q̂ - q| / Σ q`, where `q` are the true expected units.
+- **Counterfactual prediction:** predict how these units change if the price of one product or one brand is raised or
+  lowered by 10% through its promotion discount (32 scenarios: the top 12 products and the top 4 brands, each +10% and
+  -10%). On the rows `R` where a scenario changes a target's price, the predicted percentage change is applied to the
+  true units, `q̃_cf = q (1 + Δq̂ / q̂)`, so that an error in the sales level cannot offset an error in the response:
+  - **counterfactual error** `= Σ_R |q̃_cf - q_cf| / Σ_R q_cf`;
+  - **counterfactual bias** `= Σ_R (q̃_cf - q_cf) / Σ_R |q_cf - q|` (a method that overstates the effect of price has a
+    negative bias after a price increase and a positive bias after a decrease).
+
+The headline score is the mean counterfactual error over the 32 scenarios; the mean absolute bias and the share of
+scenarios with an overstated response describe the direction of the errors. Methods are ranked by the mean
+counterfactual error on the confounding-on configurations, separately for each demand model, and scores are reported as
+means over the five seeds with 95% t-intervals. A score is valid only if every ground-truth row has a prediction, no
+row is excluded, all 32 scenarios are scored, no key repeats in the forecast file and the scores are finite. A method
+may also submit its 40 x 40 table of elasticities, which is compared with the true table and reported, not ranked
+(`causaldemand elasticity`). [`docs/SUBMISSION.md`](docs/SUBMISSION.md) gives the exact rules.
+
+## Before you use the results
+
+- **Rounding floor.** Prices paid are rounded to end in 9 cents, while demand responds to the unrounded price. A
+  reference that knows the true demand but reads the scenario file's price change at face value scores a
+  counterfactual error of 0.038 to 0.042 (mean over seeds per configuration). These reference rows are in
+  [`results/main/reference_rows/`](results/main/reference_rows/README.md).
+- **Pseudonyms can be decoded by inference.** Brands, markets, chains and manufacturers are replaced by codes, and
+  product and store identifiers by numbers, but the codes keep the order of the real labels (so that regeneration
+  reproduces the data exactly) and the released sales and public influenza data reveal more. Many codes can be
+  matched to a real market, chain, brand, manufacturer, product or store ([`docs/DATASHEET.md`](docs/DATASHEET.md)).
+- **Reference models fill the seasonal index of the evaluation weeks with 1.0**, a high-season value on the
+  discrete-choice index. The archived results keep it. Filling the training-window mean instead changes one examined
+  run's counterfactual error by 0.002 and its forecast error by 0.013; the five-seed counterfactual error of that
+  configuration does not change at three decimals ([`docs/EXPERIMENT_RULES.md`](docs/EXPERIMENT_RULES.md), Section 4.4).
+- **The elasticity diagnostic** (Appendix F.4 of the paper) is supplementary. The true table describes a change of the
+  regular price; the log-linear cross elasticities count a competitor only where it is stocked
+  ([`docs/DATA.md`](docs/DATA.md), Section 4.4).
+- **Foundation-model runs made again.** The predictions of the original foundation-model runs on the facial-tissue
+  datasets were not kept, so the runs of the main experiment and of the confounding strengths 0.15 and 0.45 were made
+  again on 2026-10-03 with the original settings. Their predictions and run records are released, and the scores in
+  `results/` are theirs. The foundation-model entries of the main results table and of the confounding-strength table
+  at each strength are unchanged at three decimals. A run repeated on the GPU model of the original run gave the same
+  scores; on another GPU model the counterfactual and forecast errors of single runs differ by at most 0.0017 (GPU
+  rounding; [`results/README.md`](results/README.md)).
+- **Not released:** the licensed store sales records and store profiles the generator was calibrated on; the datasets
+  and predictions of the household-switching robustness check (its scores are in `results/`); the pipeline that wrote
+  the product descriptions and computed the product distances (both are released as files); the code and runs of the
+  paper's real-data check.
+- **Yogurt.** The 20 yogurt datasets, their ground truth and the predictions and run records of the 320 yogurt runs
+  (100 CPU runs with default settings, 80 forecast-tuned CPU runs and 140 foundation-model runs) are released as files.
+  The licensed yogurt sales records, product dictionary and household purchase panel they rest on, and the code that
+  computed their product distance, are not released; no calibration bundle is given for them, so they cannot be
+  regenerated. Their product, store, brand, chain and manufacturer codes are numbered separately from those of the
+  facial-tissue datasets; only the market codes are shared. All 20 fail the generator's seasonal-swing check (a swing
+  of 0.016 to 0.044 against the band 0.057 ± 0.01) and are released as generated ([`docs/DATA.md`](docs/DATA.md),
+  Section 11).
+
+## Documentation
+
+| Document | Contents |
+|---|---|
+| [`docs/DATA.md`](docs/DATA.md) | Every file, column and key of the datasets, the runs and the generator inputs |
+| [`docs/SUBMISSION.md`](docs/SUBMISSION.md) | Prediction formats, what a method may use, validity, scores, reporting |
+| [`docs/REPRODUCE.md`](docs/REPRODUCE.md) | Four tiers: rescore the released predictions, rerun the CPU models, regenerate a dataset, rerun the foundation models |
+| [`docs/EXPERIMENT_RULES.md`](docs/EXPERIMENT_RULES.md) | The protocol behind the reference results: scenarios, scores, models and settings |
+| [`docs/DATASHEET.md`](docs/DATASHEET.md) | Datasheet: motivation, composition, collection, uses, distribution, maintenance |
+| [`locks/README.md`](locks/README.md) | Hash-checked environments for the generator, the CPU models and the foundation models |
+
+## Repository
+
+| Folder | Contents |
+|---|---|
+| [`causaldemand/`](causaldemand) | The `causaldemand` command: download, score, elasticity, rescore, generate |
+| [`metrics/`](metrics/README.md) | The scorer (`panel_scorer.py`) and the elasticity diagnostic (`elasticity.py`) |
+| [`generator/`](generator/README.md) | The data generator, the counterfactual scenarios and the ground truth |
+| [`config/`](config/README.md) | One configuration per dataset |
+| `inputs/` | Generator inputs: the [calibration bundle](inputs/calibration_bundle/README.md), the [product descriptions and distances](inputs/product_distances/README.md), the [CDC influenza curves](inputs/cdc-ili/README.md) |
+| [`packaging/`](packaging/README.md) | Turns generator output into the released files |
+| [`models/`](models/README.md) | The reference models of the paper, as the code of the archived runs |
+| [`experiments/`](experiments/README.md) | The scripts behind every result of the paper |
+| [`results/`](results/README.md) | Scores of the paper's reference-model runs on the synthetic datasets (facial tissue and yogurt), with run records except for the household-switching runs |
+| [`notebooks/`](notebooks) | `quickstart.ipynb` and `full_run.ipynb` ([Notebooks](#notebooks)) |
+| `tests/` | Tests of the package; with `metrics/`, the default pytest run of the repository |
+
+## License and citation
+
+Code: Apache License 2.0 ([`LICENSE`](LICENSE)). Data: CC BY 4.0 ([`DATA_LICENSE.md`](DATA_LICENSE.md)), which also
+lists the sources and their terms. To cite, use [`CITATION.cff`](CITATION.cff): the paper, and the datasets by their
+DOI:
+
+```bibtex
+@inproceedings{hong2027causaldemand,
+  title     = {CausalDemand: Benchmarking Causal Recovery of Demand from Retail Transactions and Product Text},
+  author    = {Hong, Juwon and Hwang, Minha and Shankar, Venkatesh},
+  booktitle = {Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2027), Datasets and Benchmarks Track},
+  year      = {2027},
+  note      = {Submitted}
+}
+
+@misc{hong2026causaldemanddata,
+  title     = {CausalDemand},
+  author    = {Hong, Juwon and Hwang, Minha and Shankar, Venkatesh},
+  year      = {2026},
+  publisher = {Hugging Face},
+  doi       = {10.57967/hf/10744},
+  url       = {https://huggingface.co/datasets/jean-jsj/CausalDemand},
+  note      = {Datasets, version 1.0.0 (Hugging Face revision 3b0a380)}
+}
+```
+
+The first author maintains the benchmark; questions and problems go to the GitHub issues of this repository.