Everything below computes client-side in plain JavaScript from the paper’s measured figures (embedded from exp/results/summary.json). Nothing is played back: change any input and the outputs change. Classes: MEASURED figures are the paper’s; outputs derived from them are CALCULATED; the nesting walk is a SIMULATED analyzer (a live REA run remains future work, as the paper states).
Pick a same-fixture pair from the virtual test lab (item COMP-01: both tools ran on byte-identical inputs), or enter your own artifact size to apply the paper’s measured Table 1 growth for the upstream default and the lite shield’s measured added bytes. Outputs are CALCULATED from MEASURED inputs.
Builds ThoxShield’s GGUF nesting breaker (nested ARRAY frames) in memory and walks it with a budget enforced during traversal — the walk aborts the moment depth exceeds the budget, exactly the lab’s SIM-02 method. REA’s published budget is 32; the breaker depth is 35.
ThoxShield refuses a shield that would exceed the cap: max(4,096 B, 5% of the artifact). Enter an artifact size and a format’s measured lite addition to see the verdict the hardener would reach. CALCULATED.