Lean (version 4.19.0, x86_64-unknown-linux-gnu, commit 6caaee842e94, Release)
ℹ [7/26] Replayed Effective.Examples
info: ././././Effective/Examples.lean:59:0: 'Effective.compile_correct' depends on axioms: [propext]
info: ././././Effective/Examples.lean:60:0: 'Effective.execute_correct' depends on axioms: [propext]
info: ././././Effective/Examples.lean:61:0: 'Jets.compileJet_correct' depends on axioms: [propext, Quot.sound]
info: ././././Effective/Examples.lean:62:0: 'Jets.executeChartJetCoefficient_transport' depends on axioms: [propext]
info: ././././Effective/Examples.lean:63:0: 'Effective.Sample.derivatives_correct' depends on axioms: [propext, Quot.sound]
info: ././././Effective/Examples.lean:64:0: 'Effective.Method.step_equivariant' depends on axioms: [propext]
info: ././././Effective/Examples.lean:65:0: 'Effective.Method.distribution_equivariant' depends on axioms: [propext, Quot.sound]
info: ././././Effective/Examples.lean:66:0: 'Effective.PathProof.compile' does not depend on any axioms
info: ././././Effective/Examples.lean:67:0: 'Effective.PathProof.complete' does not depend on any axioms
info: ././././Effective/Examples.lean:68:0: 'Effective.Path.diagonalLoop_ne_refl' does not depend on any axioms
info: ././././Effective/Examples.lean:69:0: 'Effective.Examples.sgd_expectation' depends on axioms: [propext]
info: ././././Effective/Examples.lean:70:0: 'Effective.Examples.momentum_expectation' depends on axioms: [propext]
info: ././././Effective/Examples.lean:71:0: 'Effective.Examples.newton_expectation' depends on axioms: [propext]
info: ././././Effective/Examples.lean:72:0: 'Jets.quarticHalley' depends on axioms: [propext]
info: ././././Effective/Examples.lean:73:0: 'Effective.DoubleDelooping.interchange_certificate' does not depend on any axioms
info: ././././Effective/Examples.lean:74:0: 'Effective.DoubleDelooping.no_section_through_terminal' does not depend on any axioms
ℹ [10/26] Replayed Effective.AutoencoderExample
info: ././././Effective/AutoencoderExample.lean:160:0: 'Effective.AutoencoderExample.exact_compiled_loss_curve' depends on axioms: [propext, Quot.sound]
info: ././././Effective/AutoencoderExample.lean:161:0: 'Effective.AutoencoderExample.exact_compiled_final_probe' depends on axioms: [propext]
info: ././././Effective/AutoencoderExample.lean:162:0: 'Effective.AutoencoderExample.swapped_compiled_replay' depends on axioms: [propext]
ℹ [14/26] Replayed Main
info: ././././Main.lean:34:0: 'Effective.Autoencoder.compiledTrain_correct' depends on axioms: [propext]
info: ././././Main.lean:35:0: 'Effective.Autoencoder.Model.distribution_relabel' depends on axioms: [propext, Quot.sound]
info: ././././Main.lean:36:0: 'Effective.Autoencoder.ModelPath.duplicateLoop_nontrivial' does not depend on any axioms
info: ././././Main.lean:37:0: 'Effective.Autoencoder.ChartExample.exactLossJet' depends on axioms: [propext, Quot.sound]
info: ././././Main.lean:38:0: 'Effective.Autoencoder.ChartExample.exactNewtonLineStep' depends on axioms: [propext]
Build completed successfully.
SGD: ABC (9/8, -2); two-draw mean (145/256, -3/2); histories 9; mass 1
Momentum: ABC (23/16, -3); two-draw mean (185/256, -2); histories 9; mass 1
Newton: ABC (3, -2); two-draw mean (3/2, -3/2); histories 9; mass 1
SGD along the coordinate-swap arrow: (-2, 9/8)
Quartic Taylor jet at 1: [1, 5, 8, 4, 1]
Quartic Halley step from 1: 29/49
The B²C₂ generator equals the identity 2-cell: false
Autoencoder 4→2→4: compiled rational replay with quantization after each update
  step 0: train mean loss 253/128; held-out mean loss 17765/16384
  step 24: train mean loss 2961942727595501/2251799813685248; held-out mean loss 63889366923370361/72057594037927936
  step 48: train mean loss 2506274285613929/9007199254740992; held-out mean loss 6575880403925209/72057594037927936
  step 72: train mean loss 1196538387416269/9007199254740992; held-out mean loss 401150585002289/9007199254740992
  step 96: train mean loss 645774757259623/9007199254740992; held-out mean loss 130921364073197/2251799813685248
  probe (1, 1/2, -1/2, -1): latent (7109/4096, 6993/4096); reconstruction (15161647/16777216, 34149/65536, -34149/65536, -15161647/16777216)
  replay from swapped initialization: latent (6993/4096, 7109/4096); reconstruction (15161647/16777216, 34149/65536, -34149/65536, -15161647/16777216)
Autoencoder local loss jet: [(16205 : Rat)/8192, (-121 : Rat)/512, (163 : Rat)/128, (7 : Rat)/8, 1]
  backprop directional derivative -121/512
  Newton distance 121/1304; sample loss 16205/8192 -> 5690182520071/2891414573056
Iris reconstruction (120 training, 30 held out, mean half-squared loss)
Training-mean baseline: train 0.139442, held out 0.128459
Untrained autoencoder: train 0.115408, held out 0.108481
Epoch 1: train 0.021482, held out 0.021393
Epoch 4: train 0.012202, held out 0.011594
Epoch 8: train 0.007521, held out 0.007558
Epoch 16: train 0.006241, held out 0.007093
