World Alignment: unpaired cross-modal correspondence by relational identifiability

Method: scene states are sets of part states; relation fields are built
within each modality and are invariant to how each side labels its own
features; the cross-modal bridge is a coupling searched under an energy
that is a closed-form functional of one matrix; solving is spectral
initialisation followed by exact local refinement.

Evidence: in a procedurally generated closed world, blind recovery of a
hidden image-caption correspondence reaches 95.3% at 256 scenes against
0.39% chance, and the recovered pairs transfer to 200 held-out scenes at
93.0% exact retrieval with random-pair and shuffled-image controls at or
near chance. Cross-modal value correspondence is derived from disjoint
corpora rather than declared. On Visual Genome the field correlation
reaches 0.656 against the 0.9 that polynomial recovery needs, with the
deficit attributed away from segmentation and discretisation.

Protocol: no image-text pair enters any objective, optimiser,
initialisation, or model selection; hidden pairs score orderings only.

Co-Authored-By: Claude <noreply@anthropic.com>

1 files changed, 471 insertions, 0 deletions

diff --git a/artifacts/synth_v0/content_gate_simmim_k32.json b/artifacts/synth_v0/content_gate_simmim_k32.json
new file mode 100644
index 0000000..0766260
--- /dev/null
+++ b/artifacts/synth_v0/content_gate_simmim_k32.json
@@ -0,0 +1,471 @@
+{
+  "protocol": "Content-projected states, directions fitted without pairs on held-out scenes; the complete assignment gate is scored with hidden pairs.",
+  "dataset": "flickr",
+  "dims": 32,
+  "samples": 512,
+  "gate_a": {
+    "true": {
+      "mse": 1.844810211280161,
+      "mse_channels": [
+        1.844810211280161
+      ],
+      "mse_channel_mean": 1.844810211280161,
+      "conditional_kl": 137.15283750731368,
+      "m30_total": 31.120187924023057
+    },
+    "similarity_shift": {
+      "mse": 2.004996945997749,
+      "mse_channels": [
+        2.004996945997749
+      ],
+      "mse_channel_mean": 2.004996945997749,
+      "conditional_kl": 171.62256096780501,
+      "m30_total": 38.334506085556505
+    },
+    "random": {
+      "mse": {
+        "mean": 1.9999890327453613,
+        "std": 0.006502027157694101,
+        "min": 1.976330280303955,
+        "true_z": 23.8662166595459,
+        "true_rank_among_random": 0,
+        "count": 1000
+      },
+      "mse_channel_mean": {
+        "mean": 1.9999890327453613,
+        "std": 0.006502027157694101,
+        "min": 1.976330280303955,
+        "true_z": 23.8662166595459,
+        "true_rank_among_random": 0,
+        "count": 1000
+      },
+      "conditional_kl": {
+        "mean": 172.59413146972656,
+        "std": 0.8811267614364624,
+        "min": 169.89364624023438,
+        "true_z": 40.22270202636719,
+        "true_rank_among_random": 0,
+        "count": 1000
+      },
+      "m30_total": {
+        "mean": 38.51880645751953,
+        "std": 0.17819947004318237,
+        "min": 37.971012115478516,
+        "true_z": 41.51874542236328,
+        "true_rank_among_random": 0,
+        "count": 1000
+      }
+    }
+  },
+  "gate_b": {
+    "pairs": 130816,
+    "improving_pairs": 6729,
+    "improving_fraction": 0.05143866193737769,
+    "delta_mean": 0.001239330213353294,
+    "delta_min": -0.008128058275144848,
+    "identity_is_local_min_mse": false,
+    "worst_improving_swaps": [
+      {
+        "pair": [
+          44,
+          468
+        ],
+        "delta_mse": -0.008128058275144848,
+        "text_cosine": 1.5612499384259757,
+        "visual_cosine": -2.7689657138423684
+      },
+      {
+        "pair": [
+          44,
+          67
+        ],
+        "delta_mse": -0.007586740551768971,
+        "text_cosine": -0.43287228875729283,
+        "visual_cosine": -5.086753941409695
+      },
+      {
+        "pair": [
+          44,
+          62
+        ],
+        "delta_mse": -0.007125813375887883,
+        "text_cosine": -0.6928210192309991,
+        "visual_cosine": -3.434739305124902
+      },
+      {
+        "pair": [
+          44,
+          495
+        ],
+        "delta_mse": -0.007120108622317686,
+        "text_cosine": -0.9750438221195373,
+        "visual_cosine": -3.222959159022599
+      },
+      {
+        "pair": [
+          44,
+          85
+        ],
+        "delta_mse": -0.006695005565022447,
+        "text_cosine": -0.39104471209744573,
+        "visual_cosine": -3.5084679431578882
+      },
+      {
+        "pair": [
+          0,
+          44
+        ],
+        "delta_mse": -0.006491390155252355,
+        "text_cosine": 0.9397540805698859,
+        "visual_cosine": -4.1194203048096005
+      },
+      {
+        "pair": [
+          337,
+          495
+        ],
+        "delta_mse": -0.00629538176804794,
+        "text_cosine": -0.24228143709408076,
+        "visual_cosine": -3.392061341909287
+      },
+      {
+        "pair": [
+          44,
+          53
+        ],
+        "delta_mse": -0.0061662610559064905,
+        "text_cosine": -0.9026227847738374,
+        "visual_cosine": -3.1552342831674447
+      },
+      {
+        "pair": [
+          36,
+          44
+        ],
+        "delta_mse": -0.0058999521938150334,
+        "text_cosine": -0.4296819615253912,
+        "visual_cosine": -2.974524633398935
+      },
+      {
+        "pair": [
+          411,
+          495
+        ],
+        "delta_mse": -0.005676624840686077,
+        "text_cosine": -0.2635911374880898,
+        "visual_cosine": -3.181174947422931
+      },
+      {
+        "pair": [
+          44,
+          203
+        ],
+        "delta_mse": -0.005614514039757846,
+        "text_cosine": -0.40713098871940995,
+        "visual_cosine": -4.383080219726325
+      },
+      {
+        "pair": [
+          67,
+          337
+        ],
+        "delta_mse": -0.005610249606507696,
+        "text_cosine": -1.3503613818617461,
+        "visual_cosine": -3.7797333937646833
+      },
+      {
+        "pair": [
+          337,
+          468
+        ],
+        "delta_mse": -0.005598396188044702,
+        "text_cosine": -1.4080000855393644,
+        "visual_cosine": -3.0878435021236004
+      },
+      {
+        "pair": [
+          26,
+          44
+        ],
+        "delta_mse": -0.005509493680000466,
+        "text_cosine": -0.11621733050711933,
+        "visual_cosine": -4.134339810712093
+      },
+      {
+        "pair": [
+          384,
+          411
+        ],
+        "delta_mse": -0.005409247301562223,
+        "text_cosine": -0.7906092572622423,
+        "visual_cosine": -2.870932810458305
+      },
+      {
+        "pair": [
+          44,
+          229
+        ],
+        "delta_mse": -0.005404972263575057,
+        "text_cosine": -0.7432228189130635,
+        "visual_cosine": -3.6985020700171165
+      },
+      {
+        "pair": [
+          44,
+          197
+        ],
+        "delta_mse": -0.005308725763337642,
+        "text_cosine": -0.07460056918828972,
+        "visual_cosine": -4.784314440312638
+      },
+      {
+        "pair": [
+          411,
+          446
+        ],
+        "delta_mse": -0.0052749546309531405,
+        "text_cosine": 0.6776748670955213,
+        "visual_cosine": -3.4307650195756514
+      },
+      {
+        "pair": [
+          0,
+          337
+        ],
+        "delta_mse": -0.005215021377308596,
+        "text_cosine": -0.9118954938025817,
+        "visual_cosine": -2.775284153246821
+      },
+      {
+        "pair": [
+          337,
+          385
+        ],
+        "delta_mse": -0.0051436506694668505,
+        "text_cosine": 1.7781635256151704,
+        "visual_cosine": -2.574617913091009
+      }
+    ]
+  },
+  "descent_from_true": {
+    "start_accuracy": 1.0,
+    "final_accuracy": 0.009765625,
+    "final_objective": 1.5688702240501762,
+    "final_energy": {
+      "mse": 1.5688702240501762,
+      "mse_channels": [
+        1.5688702240501762
+      ],
+      "mse_channel_mean": 1.5688702240501762,
+      "conditional_kl": 124.80006548339843,
+      "m30_total": 28.09775354478004
+    },
+    "accepted_moves": 1011,
+    "moved_fraction": 0.990234375,
+    "mse_local_min_certificate": true,
+    "trajectory": [
+      {
+        "step": 0,
+        "objective": 1.836682153005016,
+        "accuracy": 0.99609375
+      },
+      {
+        "step": 50,
+        "objective": 1.7540454121483582,
+        "accuracy": 0.810546875
+      },
+      {
+        "step": 100,
+        "objective": 1.7155301515371122,
+        "accuracy": 0.65234375
+      },
+      {
+        "step": 150,
+        "objective": 1.6859674597753478,
+        "accuracy": 0.501953125
+      },
+      {
+        "step": 200,
+        "objective": 1.6626857122849417,
+        "accuracy": 0.384765625
+      },
+      {
+        "step": 250,
+        "objective": 1.6442835779918294,
+        "accuracy": 0.294921875
+      },
+      {
+        "step": 300,
+        "objective": 1.6299289104772625,
+        "accuracy": 0.224609375
+      },
+      {
+        "step": 350,
+        "objective": 1.618714428659824,
+        "accuracy": 0.1640625
+      },
+      {
+        "step": 400,
+        "objective": 1.6094114795326397,
+        "accuracy": 0.11328125
+      },
+      {
+        "step": 450,
+        "objective": 1.6011428657007496,
+        "accuracy": 0.083984375
+      },
+      {
+        "step": 500,
+        "objective": 1.595125917571092,
+        "accuracy": 0.068359375
+      },
+      {
+        "step": 550,
+        "objective": 1.5898438045117038,
+        "accuracy": 0.0546875
+      },
+      {
+        "step": 600,
+        "objective": 1.5850386783075012,
+        "accuracy": 0.041015625
+      },
+      {
+        "step": 650,
+        "objective": 1.5817489479077362,
+        "accuracy": 0.033203125
+      },
+      {
+        "step": 700,
+        "objective": 1.5790571929400437,
+        "accuracy": 0.017578125
+      },
+      {
+        "step": 750,
+        "objective": 1.5768918645924703,
+        "accuracy": 0.015625
+      },
+      {
+        "step": 800,
+        "objective": 1.5732990009406131,
+        "accuracy": 0.015625
+      },
+      {
+        "step": 850,
+        "objective": 1.571967851497371,
+        "accuracy": 0.013671875
+      },
+      {
+        "step": 900,
+        "objective": 1.5704626565804258,
+        "accuracy": 0.013671875
+      },
+      {
+        "step": 950,
+        "objective": 1.5697695375884466,
+        "accuracy": 0.013671875
+      },
+      {
+        "step": 1000,
+        "objective": 1.5689690074752038,
+        "accuracy": 0.009765625
+      }
+    ]
+  },
+  "descent_from_random": [
+    {
+      "start_accuracy": 0.00390625,
+      "final_accuracy": 0.0,
+      "final_objective": 1.5586234547779543,
+      "final_energy": {
+        "mse": 1.5586234547779543,
+        "mse_channels": [
+          1.5586234547779543
+        ],
+        "mse_channel_mean": 1.5586234547779543,
+        "conditional_kl": 126.79909755706429,
+        "m30_total": 28.477066420968768
+      },
+      "accepted_moves": 957,
+      "moved_fraction": 0.9921875,
+      "mse_local_min_certificate": true
+    },
+    {
+      "start_accuracy": 0.0,
+      "final_accuracy": 0.001953125,
+      "final_objective": 1.572073356332826,
+      "final_energy": {
+        "mse": 1.572073356332826,
+        "mse_channels": [
+          1.572073356332826
+        ],
+        "mse_channel_mean": 1.572073356332826,
+        "conditional_kl": 128.55083397529833,
+        "m30_total": 28.85431350772532
+      },
+      "accepted_moves": 792,
+      "moved_fraction": 0.998046875,
+      "mse_local_min_certificate": true
+    },
+    {
+      "start_accuracy": 0.001953125,
+      "final_accuracy": 0.001953125,
+      "final_objective": 1.5586501101673844,
+      "final_energy": {
+        "mse": 1.5586501101673844,
+        "mse_channels": [
+          1.5586501101673844
+        ],
+        "mse_channel_mean": 1.5586501101673844,
+        "conditional_kl": 126.27871960900215,
+        "m30_total": 28.3730441421352
+      },
+      "accepted_moves": 861,
+      "moved_fraction": 0.9921875,
+      "mse_local_min_certificate": true
+    },
+    {
+      "start_accuracy": 0.0,
+      "final_accuracy": 0.0,
+      "final_objective": 1.5667351390718218,
+      "final_energy": {
+        "mse": 1.5667351390718218,
+        "mse_channels": [
+          1.5667351390718218
+        ],
+        "mse_channel_mean": 1.5667351390718218,
+        "conditional_kl": 124.38017684176847,
+        "m30_total": 28.00950564649734
+      },
+      "accepted_moves": 854,
+      "moved_fraction": 0.994140625,
+      "mse_local_min_certificate": true
+    },
+    {
+      "start_accuracy": 0.001953125,
+      "final_accuracy": 0.00390625,
+      "final_objective": 1.5579046963474938,
+      "final_energy": {
+        "mse": 1.5579046963474938,
+        "mse_channels": [
+          1.5579046963474938
+        ],
+        "mse_channel_mean": 1.5579046963474938,
+        "conditional_kl": 124.79006043459924,
+        "m30_total": 28.07382147961484
+      },
+      "accepted_moves": 930,
+      "moved_fraction": 1.0,
+      "mse_local_min_certificate": true
+    }
+  ],
+  "verdict": {
+    "true_z_mse": 23.8662166595459,
+    "improving_fraction": 0.05143866193737769,
+    "identity_strict_local_min": false,
+    "descent_keeps": 0.009765625,
+    "true_mse": 1.844810211280161,
+    "best_random_descent": 1.5579046963474938,
+    "best_random_accuracy": 0.00390625,
+    "counterfeit_found": true
+  }
+}
\ No newline at end of file