# Chapter 1: finite alert–demand worlds

These programs enumerate a controlled teaching model. They do not use sampled utility data or estimate an empirical alert effect. Six background states have integer weights of one. Dividing by total weight gives probability 1/6. Independent alert assignment crosses those backgrounds with A=0 and A=1, producing twelve equal states.

Run from this directory, or pass an absolute script path:

```sh
python alert-worlds.py
julia --startup-file=no alert-worlds.jl
Rscript alert-worlds.R
python verify.py
```

The calculations use Python's standard library, Julia Base, and base R. No external numerical package or internet connection is required. Python 3.14.7, Julia 1.12.7, and R 4.6.1 are the native reference environment; the generated `results/verification.json` records the versions actually executed. Python, R and Julia implement the quantities independently. The verifier executes them concurrently and requires matching metric names, finite values, cross-language agreement, analytic targets, state-by-state invariance, positive assigned-alert loads, and correct MW-to-kW scaling.

`data/manifest.json` describes the input and its SHA256. `results/verification.json` describes the checks. The continuous parameter-family and nonidentification proofs are in the chapter. Numerical checks at five parameter values do not replace those proofs. No Monte Carlo sample, browser language runtime, or live assessment is included.

The figure-generation program `figures.py` uses Matplotlib. It reads the same fixture; Matplotlib is only required when regenerating the scientific figures. The released SVG, PNG, and PDF figures can be viewed without it.

Authored teaching text, examples, code and figures: © 2026 Ian Helfrich, all rights reserved. Public access does not grant a blanket redistribution or adaptation license. Cited books and papers are linked, not included.


# Chapter 2: measurement, calibration, and denominators

Run the independent standard-library implementations and verifier:

```sh
python measurement.py
julia --startup-file=no measurement.jl
Rscript measurement.R
python verify-measurement.py
```

The 36-row `data/meter-states.csv` fixture crosses true energy, two independent meter-noise multipliers, and an independent billing perturbation. `data/measurement-manifest.json` specifies units, probabilities, calibration/noise settings and the separate controlled denominator examples. `data/measurement-dictionary.csv` records the fields' meanings and support. These are invented teaching records, not observed tariffs or actual meter tests.

The verifier matches 301 independently computed quantities, 189 analytic targets, and 108 row identities at three noise amplitudes and three calibrations. It checks classical attenuation, a repeated-reading covariance ratio, the failure of a copied reading or shared calibration, projection and conditional-mean risks, averaging, MWh-to-kWh scaling, ratio growth, and weighted aggregation. `results/measurement-verification.json` records native versions and tolerances; `results/measurement-reference.json` contains all checked population quantities. No Monte Carlo draws or finite-sample unbiasedness claims are involved.

`measurement-figures.py` uses Matplotlib to regenerate the original web and vector print diagrams. Browser controls enumerate the declared model in JavaScript; they do not run Python, Julia or R. The manuscript supplies the proofs and restrictions that give these numerical results their interpretation.

# Chapter 3: observation clocks and revisions

```sh
python clocks.py
julia --startup-file=no clocks.jl
Rscript clocks.R
python verify-clocks.py
```

All three implementations use their standard libraries (Julia includes Dates). No internet or external numerical package is required to run the released fixture calculations. The verifier runs them concurrently and checks 288 quantities against explicit expected targets, not just each other. It tests nine origins, three collection delays, inclusive boundary selection, unavailable periods, future-only value changes, a future appended record, input-order invariance, duplicate/nonfinite rejection, negative-delay rejection, and refusal to forecast an empty input. It also checks the target-score identity and the finite information-gain identity. Native versions, tolerances and input hashes are in `results/clocks-verification.json`.

`data/gdp-release-ledger.csv` contains three attributed BEA real-GDP growth headlines for 2024 Q1, with original publication times. `data/gdp-source-receipts.json` records raw-document hashes and bounded source reads; the PDF source documents are linked, not included. `data/clocks-manifest.json` defines annualization, rounding, reference period, clocks, keys, rights and the declared collection-delay scenarios. Publication plus a modelled delay is not evidence of actual historical receipt time. The JSON copies are derived from the CSV inputs for browser use; their rows are checked against the native source.

The six records in `data/controlled-release-ledger.csv` and the four states in `data/information-worlds.csv` are separately authored teaching fixtures. Their forecast and proof calculations are not empirical GDP results. A later training revision changes the incorrect old-origin mean from 3.5 to 2; future-only mutation leaves the correct mean at 3.5. First and revised targets are scored separately. No nowcast is fitted, and the larger C02 real-time case remains in development.

`clocks-figures.py` regenerates the original diagrams with Matplotlib. The interactive release audit uses JavaScript and has a complete static comparison in the PDF and web print view. It does not execute Python, Julia or R in the browser.