Files
duocthu/ingestion/data/verified/formula_regions_2d.json
T

185 lines
3.1 KiB
JSON

{
"note": "2D (stacked-fraction) formula regions, every one confirmed by rendering the page and reading it. bbox is the fraction bar itself; numerator and denominator sit above and below it.",
"verified_on": "2026-08-01",
"method": "residual-ink fraction_bar_candidate, then visual inspection of all 23 candidates",
"regions": [
{
"physical_page": 43,
"bar_bbox": [
133.44,
308.16,
222.72,
308.16
]
},
{
"physical_page": 92,
"bar_bbox": [
430.08,
603.36,
465.6,
603.36
]
},
{
"physical_page": 92,
"bar_bbox": [
362.88,
704.64,
386.4,
704.64
]
},
{
"physical_page": 92,
"bar_bbox": [
400.32,
704.64,
422.88,
704.64
]
},
{
"physical_page": 92,
"bar_bbox": [
456.96,
704.64,
480.0,
704.64
]
},
{
"physical_page": 92,
"bar_bbox": [
505.44,
704.64,
522.72,
704.64
]
},
{
"physical_page": 202,
"bar_bbox": [
371.52,
151.68,
489.6,
151.68
]
},
{
"physical_page": 325,
"bar_bbox": [
445.92,
562.08,
544.32,
562.56
]
},
{
"physical_page": 325,
"bar_bbox": [
439.68,
623.52,
560.16,
623.52
]
},
{
"physical_page": 349,
"bar_bbox": [
384.48,
336.0,
491.52,
336.48
]
},
{
"physical_page": 1042,
"bar_bbox": [
97.92,
492.0,
286.56,
492.0
]
},
{
"physical_page": 1043,
"bar_bbox": [
94.56,
536.16,
140.16,
536.16
]
},
{
"physical_page": 1043,
"bar_bbox": [
145.92,
536.16,
244.8,
536.16
]
},
{
"physical_page": 1132,
"bar_bbox": [
63.36,
498.72,
239.52,
498.72
]
},
{
"physical_page": 1402,
"bar_bbox": [
120.96,
711.84,
201.6,
711.84
]
},
{
"physical_page": 1402,
"bar_bbox": [
180.48,
770.4,
205.92,
770.4
]
},
{
"physical_page": 147,
"bar_bbox": [
307.9,
672.0,
428.5,
672.0
],
"source_prints_no_bar": true,
"note": "ADENOSIN infusion-rate formula. The source page prints three plain lines with no fraction bar at all, so no geometric detector can find it — confirmed by rendering the region and reading it. Left in prose it reads as a multiplication chain. Quarantined on the strength of the reading, and flagged for human confirmation of the intended division."
}
],
"rejected": [
{
"physical_page": 4,
"reason": "decorative underlines on the Ministry decision page"
},
{
"physical_page": 63,
"reason": "ruled box around a treatment-protocol paragraph"
},
{
"physical_page": 845,
"reason": "table header cell border"
},
{
"physical_page": 878,
"reason": "table header cell border"
},
{
"physical_page": 1667,
"reason": "rule above the colophon on the last page"
}
],
"recall_limit": "The fraction-bar signal cannot find a fraction the source never typeset. ADENOSIN (physical page 147) is one confirmed case, found only because a prose-leak gate matched its text. The true number of bar-less formulas in the book is UNMEASURED."
}