ACCURACY POLICY
What is exact, what is inferred, and what is ours
HOI4 War Planner uses a bundled vanilla 1.19.3 baseline. Some values come straight from retained game files, some behavior has to be inferred, and some outputs are intentionally planner-created analysis.
Evidence classes
Game-file exact means the retained 1.19.3 files directly support the value or relationship. Executable inferred means the behavior is constrained by evidence but is not fully provable from those files alone. Planner analytical means the site created the comparison, score, generated opponent set or other analytical layer.
Oracle testing can also mark a narrow result oracle-validated or oracle-divergent. Untested executable behavior remains unvalidated.
Why seeded comparisons are used
If two designs are tested under the same assumptions, it is much easier to tell which change mattered. Seeded simulations make those comparisons reproducible. A modeled win rate is still an analytical estimate, not a promise about one battle in HOI4.
How Counter works
Counter starts from the two Division Lab templates and keeps the selected tech, doctrine, MIO, equipment and battlefield assumptions. It tests bounded structural changes against the same target and battle-simulates the final recommendations.
The recommendation search and ranking are planner analytical. The cards show direct equipment cost per division as current IC/div → proposed IC/div, with the absolute and percentage change. Counter does not reconstruct your live production queue or decide whether your economy can afford the recommendation.
Oracle-validated land-combat core
For HOI4 1.19.3, controlled Oracle experiments O1–O25 support the core ordinary 1v1 resolver at the tested boundaries: a one-hour initial fire delay, the wider attack-point sampler, stochastic Defense/10 rounding with bounding, 10% defended and 40% undefended hit gates, ordinary 1–4 organization and 1–2 strength dice, and an additional 0.9 executable strength-damage scalar.
The armor/piercing series also validates the tested 100% / 80% / 65% / 50% piercing damage tiers, a 1–6 armored-on-soft organization die when the attacker is not pierced, and coherent composition of those mechanics in the higher-resolution O25 combined test. O24 remains preserved as a predeclared bisection14 mismatch; O25 repeated the same combat mechanics with finer measurement and resolved that lone mixed-channel result as an observation-resolution artifact.
These results are oracle-validated only within the controlled experiments; the implementation is still an executable reconstruction.
Known boundaries
The Oracle series does not establish every hardness and hard-attack mixture, tactic path outside the neutralized harness, multi-division targeting or coordination rule, reinforcement/combat-width interaction, terrain/weather combination, or broader RNG ordering detail. The site does not claim bit-for-bit hoi4.exe parity.
Gauntlet
Gauntlet uses 10,000 deterministic designs across eight terrain types while attacking and defending, producing 160,000 matchup checks.
Bottom line
The planner is a theorycrafting and comparison tool. It should tell you where a number came from, where the model is making an inference, and where the site is doing its own analysis. It does not claim bit-for-bit parity with hoi4.exe.