| name | energy-counterfactual-pair-and-hierarchy-closure |
| description | Use when the research task is in energy — batteries, electrochemistry, energy-system modelling and identification — at study design, analysis or writing. Counterfactual pair, nested closure, native resolution: the accounting an energy study is graded on |
Counterfactual pair, nested closure, native resolution: the accounting an energy study is graded on
Energy-systems work is graded on accounting. Three habits outsiders skip.
The counterfactual pair. No headline number stands alone: solve or measure the system twice, once with the studied mechanism active and once with it removed (network constraints relaxed, policy off, technology absent, do-nothing ideal), under identical cost and physical accounting, and report both arms and their difference. Design the counterfactual run at the same time as the main run; it is half the result, not a sensitivity check.
Nested closure. Report at every level the system defines - asset, then building, zone, community or country, then whole system - and show the closure: components must sum to the reported total, with the residual in physical units and an overlay of the two series for one representative window. State every rate or fraction with its numerator, its denominator and both in physical units, naming the population the denominator covers: which assets, which hours, which sites.
Native resolution. Plot the full horizon at the data's own timestep and the full extent at its own spatial unit, decomposing the total into its components at each timestep, rather than reporting only period aggregates or one mean map.
Then report one quantified effect for every input variable and data layer you were supplied, including those that turn out not to matter: an explicit null is a result, a variable that vanishes from the report reads as an omitted mechanism. Report the full cross-product of scenarios and name the best and worst cell, not the mean.
Why this is here
Four of four Energy tasks demand a paired counterfactual and three demand hierarchical closure - and both were the specific misses. The constrained-vs-unconstrained contrast existed only as two rows inside a wide scenario table and the judge wrote 'there is no explicit unconstrained case'; the highest-weight criterion of the worst task (0.40) was the children-sum-to-parent identity, computed to 1e-16 but used as one clause of a data-audit argument with no section, figure or day-level overlay, scoring 0. The per-layer null-effect rule recovers the criterion lost when a supplied input layer was discarded and then reported as a data limitation.
