ScrolIQ — Grand Prize evidence for reading Herculaneum scrolls

What is actually differentThe contribution is evidence composition, not ownership of every underlying diagnostic.

Public Vesuvius work already covers substantial parts of scan QA, TIFXYZ integrity/repair, self-intersections, spiral evaluation and held-out ink validation. ScrolIQ's system-level contribution is to bind those measurements to the exact bytes and exact eligible volume they describe, preserve negative results, and prevent missing, stale, cross-volume or train-overlapping evidence from being promoted into a stronger claim.

Identity before confidence

CT root, surface hashes, external reports, checkpoint identity, seeds and evaluation arrays travel with the result. Same scroll name is not enough.

Falsification stays visible

Disjoint resampling, injection controls and held-out checks are allowed to fail. A failed gate narrows the claim instead of disappearing.

One proof before more breadth

The next milestone is one exact Grand Prize volume with a measurable held-out surface / fit / ink improvement and a single passport connecting the evidence. Protocol →

Open-problem diagnosticsScrolIQ maps evidence to the Vesuvius Challenge pipeline instead of treating one score as readiness.

Scan diagnostics

LIVE

Real-voxel health and coordinate-preserving spatial scan maps. Next: validated layer-separability and decohesion diagnostics.

Surface topology

CT support, competing sheets, topology risk, and surface-placement uncertainty.

Mesh connectivity

LIVE

scroliq-mesh audits native TIFXYZ and can bind official Villa CT-support and VC3D self-intersection reports to the exact surface; scroliq-obj covers triangular meshes. Both expose advisory findings or explicit --fail-on-findings CI gates. Sheet identity remains unproven.

Fiber connectivity

LIVE

scroliq-fiber reads VC3D vc3d_fiber JSON v1/v3/v4 or CSV, hashes exact inputs, validates persisted span/tracer metadata, and checks rendered continuity plus control-point/render consistency. Two public corpora are now complete: 136 / 136 PHercParis4 and 411 / 411 PHerc0139 fibers (547 / 547 total). The fallback-gap pattern replicated on both scrolls, but a pre-registered span-aware gap rule failed its independent adoption gate on PHerc0139 (KEEP), so the current rule remains. Independent CT tests also replicated: fiber-point material support reached AUC 0.775 on PHercParis4 and AUC 0.760 on PHerc0139, while median tangent direction lay only 3.3° and 3.5° out of the local CT sheet plane; wrong-frame controls passed. These are physical frame-consistency results, not declared dataset bindings: Paris4 does not name the volume, and PHerc0139 is a disclosed 7-voxel range near miss. A passport-bound Fiber IQ report still requires the exact --volume-root. Fiber evidence + limits →

Winding annotations

LIVE

scroliq-winding validates PointCollections structure, exact-file provenance, volume binding, winding semantics, and axial coverage, and checks winding numbers against the umbilicus before any fit. The PHercParis4 ray-order audit found 2 / 13,700 comparable-pair inversions, both under 0.1 voxel. Cross-collection attachment then tied 2,111 / 2,232 points to verified patches, flagged 6 / 16,074 constraints, and caught 194 / 200 injected +2 errors. The queue now exports as native VC3D PointCollections for inspection. Review workflow →

Spiral fitting

The held-out geometry evaluator and reproduction contract are ready; the missing evidence is one real GPU fit scored on constraints it never saw. November N5 →

Label quality

LIVE

Label–volume coverage join: all 70 published ink-detection labels sit on PHercParis4 while 14 top-quartile volumes have zero published ink labels and are flagged as label-coverage candidates for review. This is a coverage gap, not evidence of ink or surface readiness. Next: physical label offset, snapping candidates, review queues.

Ink reliability

LIVE

scroliq-ink-audit fails closed on declared train/evaluation overlap and audits checkpoint, seed, held-out-run, and falsification-control provenance. It is an evidence audit, not an ink verdict. Next: attach measured perturbation and cross-scroll results.

Data scale

LIVE

Cloud-native partial reads, decode provenance, exact coordinate bindings, file hashes, and machine-readable artifacts keep evidence inspectable across stages.

Grand Prize

TARGET

One evidence trail from full recto coverage to TIFXYZ, renders, validation, and VC3D.

Official 2026 Open Problems · ScrolIQ alignment roadmap

Evidence contractsWhat a passing diagnostic proves — and what it deliberately leaves unknown.

Mesh IQ

Proves: native TIFXYZ structure/provenance and selected grid-local geometry checks; validated upstream reports can additionally bind CT support and transverse self-intersection evidence. Does not prove: correct sheet identity.

Fiber IQ

Proves: exact-file/native-format provenance plus rendered-line continuity and control-point/render consistency checks when supplied. Adds physical evidence: frozen CT-support and tangent-direction tests now replicate across PHercParis4 and PHerc0139. Does not prove: a declared exact-volume binding, which CT sheet a fiber belongs to, horizontal/vertical fiber identity, or one-sheet continuity.

Winding IQ

Proves: volume-bound PointCollections input integrity, descriptive axial coverage, and (with an umbilicus) radial-order review candidates with a measured recall. Does not prove: patch attachment, consistent cycles, ±1/±2 winding errors, or spiral-fit accuracy.

Ink IQ

Proves: declared spatial separation, checkpoint/seed provenance, and falsification-control coverage. Does not prove: that a prediction is ink.

Passport

Cross-volume artifacts are excluded and failed audits block their stage. A passing component remains partial until the next direct evidence layer exists.

Mesh IQ usage · Winding IQ usage · Ink IQ usage

2027 Grand Prize campaignExact eligible inputs, held-out geometry, and ink falsification before whole-scroll commitment.

Exact-volume eligibility

13 prize targets are matched by exact volume ID. Same-scroll higher-resolution scans are explicitly excluded rather than silently substituted.

Exact-volume mesh cross-cut → · Frozen qualifier →

Blind geometry first

Each first-wave target gets the same deterministic 24-region sample: 18 fit regions and 6 held-out regions. Failed regions stay in the denominator.

Blind probe protocol →

Try to falsify the ink

Promising signals are checked on the correct surface, ±3 voxel normal offsets, an adjacent winding, perturbed geometry, and an independent checkpoint/fold.

Reproducibility ledger

Exact inputs, hashes, fixed seeds, held-out splits, failure counts, experiment artifacts, and human-input time are part of the evidence trail — not cleanup work at submission time.

No opaque winner score. The frozen 2026-09-30 two-axis qualifier produced a Pareto frontier containing PHerc0813 and PHerc1447, with PHerc0800 retained as a deliberately different geometry hypothesis. Re-derived on 2026-10-04 from the pre-registered stability-v2 scores, PHerc0813 is off the frontier in every v2 view; PHerc1447 is the only robust member and the segment-free slot is a band within noise (prize-frontier-v2). The September frontier stays frozen historical evidence.

Pre-registered stability: FAIL

Two disjoint 48-chunk samples were run under a protocol frozen before reading the data. On the 48 / 64 volumes dense enough for both arms, Spearman ρ = 0.750 against a pre-registered gate of 0.85; mean |Δ| = 2.94, top-10 overlap 7 / 10. The consequence was not a softer threshold: the leaderboard now reports rank bands. Frozen result + post-hoc diagnostics →

Exact-volume Grand Prize mesh evidence

Intersecting the pinned 13-volume prize manifest with the public mesh corpus leaves only 2 / 13 eligible targets with published meshes on the exact prize-listed scan. PHerc0800: 6 / 6 clean local Mesh IQ audits. PHerc1447: 15 / 15 carry an enclosed-hole review finding; 14 are review tier and 1 is multi-defect. All 21 / 21 published segments on those two targets are accounted for. The other 11 targets are evidence gaps, not failures. Offline exact-volume cross-cut →

Independent same-byte cross-check

Mesh IQ was run on the exact 1,818,055 SHA-256-verified bytes in TIFXYZ Doctor's pinned 10-case real-data benchmark: 10 / 10 agreement on enclosed-hole presence, 10 / 10 on single-component presence, and 7 / 7 hole-presence agreement on the PHerc0800 + PHerc1447 Grand Prize overlap. The first run also exposed our own isometry normalization error; after fixing it, all 10 / 10 Doctor-below-threshold cases have no Mesh IQ isometry finding, with nearly identical p95 stretch on the Villa controls. Rival benchmark roles are provenance labels, not geometry ground truth. Frozen same-byte evidence →

Cross-tool evidence, content-bound

On PHerc0139 segment 20260306000001-w051_2026030600, ScrolIQ and Windcheck independently name the same published coordinate bytes: full x/y/z SHA-256s, no mask, 379×480 grid, and 164,096 valid vertices agree. Mesh IQ reports 15 edge jumps and 173 normal reversals; Windcheck's same-original nonlocal census reports 3,333 transverse contacts. Binding stops at coordinate-exact, not semantic-exact, because the external release-index projection does not expose the original meta.json hash. Frozen binding-aware dossier →

VC3D jump-to-problem queue

Mesh IQ now emits native VC3D PointCollections instead of leaving reviewers with counts. On the content-bound PHerc0139 surface, the frozen queue contains 35 directly loadable XYZ sites: all 15 / 15 edge jumps plus the top 20 / 173 normal reversals. The same surface is independently bound to Windcheck's 3,333-contact nonlocal census, so local review coordinates and independent nonlocal evidence can be inspected without conflating them. VC3D PointCollections → · Frozen campaign →

Winding check with measured sensitivity

On public PHercParis4 annotations, the ray-order audit found 2 inversions / 13,700 comparable pairs. In a seeded injection control, deliberate ±3 winding errors were detected in 179 / 200 testable trials and ±5 errors in 171 / 200; ±2 is undetectable by construction. That measured blind spot is part of the result. Audit + injection control →

Two-scroll physical fiber evidence

Reproducible censuses now cover 547 / 547 public fibers: 136 PHercParis4 and 411 PHerc0139. Both corpora reproduce the same descriptive gap pattern: every gap candidate occurs in a fiber containing fallback-interpolated spans. A simple span-aware gap rule was then tested under frozen rules; PHercParis4 was INSUFFICIENT (24 fallback plants < 30) and independent PHerc0139 returned KEEP because flags fell only 1,622 → 1,088 (ratio 0.67, above the 0.5 adoption gate), despite 100% planted-break recall. Physical controls independently support the proposed CT frames: fiber-point material support is AUC 0.775 on PHercParis4 and AUC 0.760 on PHerc0139; median tangent direction is 3.3° and 3.5° out of the local CT sheet plane, with wrong-frame controls passing. These results do not declare an exact dataset binding or identify which sheet a fiber follows. The exact positive control remains the 2026-10-01-public-fiber-audit: 53,828 points with 11 gap and 26 sharp-turn candidates. Fiber audit evidence →

Scan-quality distribution64 volumes · 5-point bins

Legacy triage bands: S ≥ 70 · A 60–70 · B 45–60 · C < 45

Scan-quality analyticsUpdates with the current view

—

mean score

—

median score

—

top-quartile cutoff

—

unique scrolls

—

S / A / B / C volumes

—

with segment roots

Scan-quality surveyClick a column to sort · click a row for the full breakdown

Ranks are bands. The pre-registered stability test (protocol) returned FAIL: Spearman ρ = 0.750 between two disjoint 48-chunk samples, against a gate of 0.85. So each volume shows the best–worst rank it takes across the two runs and their pool, among the 48 volumes dense enough for the test; — marks a volume too sparse to rank. Scores are the published September campaign. Those scores came from a sample that can sit in one or two slabs of a scroll: across these 48 volumes they differ from the stability test’s pooled 96-chunk score by 6.9 points on average, so treat a single score as approximate too.

Label-coverage candidate: a volume in the top quartile of this scan-health survey with zero published ink-detection labels. It is a review-priority flag, not evidence that ink is present or that a usable surface is available.

rank band	volume	score	tier	signal	texture	dynamic	penalties	ink labels	segments	note
1–2	PHerc0172 7.910µm · 53keV	78.3	S	37.0	25.2	16.5	−0.3	0	53	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0172/volumes/20241024131839-7.910um-53keV-masked.zarr` nonzero fraction 0.8317 gradient energy 10.080 dynamic range 165 chunks decoded 24 chunk score spread 21–89 (std 19.3) dead slices 0 signal 37.0/40 — nonzero voxel fraction 0.832 · texture 25.2/30 — gradient energy 10.08 · dynamic 16.5/20 — range 165 · saturation penalty −0.3 · 53 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
—	PHercParis4 45.532µm · 110keV	77.2	S	31.5	30.0	15.8	−0.1	70	169
full root `community-uploads/forrest/volcomp/PHercParis4/volumes/20260310173927-45.532um-11.0m-110keV-masked.zarr` nonzero fraction 0.7085 gradient energy 12.480 dynamic range 158 chunks decoded 21 chunk score spread 0–90 (std 30.5) dead slices 0 signal 31.5/40 — nonzero voxel fraction 0.709 · texture 30.0/30 — gradient energy 12.48 · dynamic 15.8/20 — range 158 · saturation penalty −0.1 · 70 published ink-detection labels on this scroll · 169 segment roots in open data
4–8	PHerc0139 2.403µm · 77keV	76.3	S	40.0	17.4	20.0	−1.1	0	114	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20250820105138-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9952 gradient energy 6.979 dynamic range 213 chunks decoded 24 chunk score spread 62–82 (std 4.9) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.995 · texture 17.4/30 — gradient energy 6.98 · dynamic 20.0/20 — range 213 · saturation penalty −1.1 · 114 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
11–24	PHerc0813 9.362µm · 113keV	76.2	S	40.0	21.4	15.0	−0.2	0	0	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0813/volumes/20250821151723-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.9272 gradient energy 8.559 dynamic range 150 chunks decoded 24 chunk score spread 30–88 (std 14.3) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.927 · texture 21.4/30 — gradient energy 8.56 · dynamic 15.0/20 — range 150 · saturation penalty −0.2 · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
3	PHerc0139 9.362µm · 113keV	75.8	S	40.0	20.1	15.8	−0.0	0	114	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20250728140407-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.9193 gradient energy 8.034 dynamic range 158 chunks decoded 24 chunk score spread 45–90 (std 11.5) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.919 · texture 20.1/30 — gradient energy 8.03 · dynamic 15.8/20 — range 158 · 114 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
6–9	PHerc0139 2.403µm · 77keV	75.7	S	40.0	16.6	20.0	−0.9	0	114	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20250822062710-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9901 gradient energy 6.626 dynamic range 200 chunks decoded 24 chunk score spread 67–85 (std 4.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.990 · texture 16.6/30 — gradient energy 6.63 · dynamic 20.0/20 — range 200 · saturation penalty −0.9 · 114 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
6–14	PHerc1203 9.362µm · 113keV	74.7	S	40.0	21.1	13.8	−0.2	0	0	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc1203/volumes/20250820131727-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.9232 gradient energy 8.437 dynamic range 138 chunks decoded 24 chunk score spread 20–87 (std 18.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.923 · texture 21.1/30 — gradient energy 8.44 · dynamic 13.8/20 — range 138 · saturation penalty −0.2 · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
2–33	PHerc0846B 9.362µm · 113keV	72.0	S	37.0	20.7	14.4	−0.2	0	0	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0846B/volumes/20250804142305-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.8333 gradient energy 8.282 dynamic range 144 chunks decoded 24 chunk score spread 0–87 (std 23.4) dead slices 0 signal 37.0/40 — nonzero voxel fraction 0.833 · texture 20.7/30 — gradient energy 8.28 · dynamic 14.4/20 — range 144 · saturation penalty −0.2 · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
15–20	PHerc0139 2.403µm · 77keV	70.7	S	40.0	13.9	17.0	−0.2	0	114	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20260319133554-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9958 gradient energy 5.555 dynamic range 170 chunks decoded 24 chunk score spread 58–80 (std 5.4) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.996 · texture 13.9/30 — gradient energy 5.55 · dynamic 17.0/20 — range 170 · saturation penalty −0.2 · 114 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
21–29	PHercParis4 2.400µm · 137keV	70.4	S	40.0	16.1	14.3	−0.1	70	169
full root `community-uploads/forrest/volcomp/PHercParis4/volumes/20260323153942-2.400um-0.2m-137keV-masked.zarr` nonzero fraction 0.9400 gradient energy 6.443 dynamic range 143 chunks decoded 24 chunk score spread 14–79 (std 13.8) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.940 · texture 16.1/30 — gradient energy 6.44 · dynamic 14.3/20 — range 143 · saturation penalty −0.1 · 70 published ink-detection labels on this scroll · 169 segment roots in open data
1–4	PHerc0814 9.362µm · 113keV	69.4	A	31.6	23.0	15.4	−0.7	0	65	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0814/volumes/20250804134230-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.7120 gradient energy 9.215 dynamic range 154 chunks decoded 24 chunk score spread 0–90 (std 32.6) dead slices 0 signal 31.6/40 — nonzero voxel fraction 0.712 · texture 23.0/30 — gradient energy 9.21 · dynamic 15.4/20 — range 154 · saturation penalty −0.7 · 65 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
27–31	PHerc0139 2.403µm · 77keV	68.9	A	40.0	13.2	16.0	−0.3	0	114	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20260319133050-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9986 gradient energy 5.268 dynamic range 160 chunks decoded 24 chunk score spread 61–77 (std 4.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.999 · texture 13.2/30 — gradient energy 5.27 · dynamic 16.0/20 — range 160 · saturation penalty −0.3 · 114 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
17–18	PHerc0332 2.399µm · 78keV	68.6	A	40.0	13.3	15.6	−0.3	0	0	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0332/volumes/20251211183505-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.9029 gradient energy 5.337 dynamic range 156 chunks decoded 24 chunk score spread 0–80 (std 19.4) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.903 · texture 13.3/30 — gradient energy 5.34 · dynamic 15.6/20 — range 156 · saturation penalty −0.3 · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
—	PHerc0009B 2.401µm · 77keV	68.1	A	40.0	13.0	16.0	−0.9	0	30	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0009B/volumes/20250820154339-2.401um-0.3m-77keV-masked.zarr` nonzero fraction 0.9438 gradient energy 5.209 dynamic range 160 chunks decoded 20 chunk score spread 20–89 (std 16.9) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.944 · texture 13.0/30 — gradient energy 5.21 · dynamic 16.0/20 — range 160 · saturation penalty −0.9 · 30 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
22–29	PHerc0841 2.403µm · 77keV	68.1	A	40.0	13.0	15.3	−0.2	0	6	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc0841/volumes/20260319124803-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9530 gradient energy 5.191 dynamic range 153 chunks decoded 24 chunk score spread 19–75 (std 12.1) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.953 · texture 13.0/30 — gradient energy 5.19 · dynamic 15.3/20 — range 153 · saturation penalty −0.2 · 6 segment roots in open data · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
25–32	PHerc1203 2.403µm · 77keV	67.7	A	40.0	12.9	14.9	−0.1	0	0	label-coverage candidate
full root `community-uploads/forrest/volcomp/PHerc1203/volumes/20260319130212-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.9794 gradient energy 5.166 dynamic range 149 chunks decoded 24 chunk score spread 53–83 (std 5.9) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.979 · texture 12.9/30 — gradient energy 5.17 · dynamic 14.9/20 — range 149 · saturation penalty −0.1 · label-coverage candidate — top-quartile scan health with zero published ink-detection labels; prioritization flag only, not evidence of ink or surface readiness
4–9	PHerc0358 9.362µm · 113keV	66.3	A	37.2	18.0	11.2	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0358/volumes/20250821151737-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.8374 gradient energy 7.186 dynamic range 112 chunks decoded 24 chunk score spread 0–89 (std 29.7) dead slices 0 signal 37.2/40 — nonzero voxel fraction 0.837 · texture 18.0/30 — gradient energy 7.19 · dynamic 11.2/20 — range 112 · saturation penalty −0.1
—	PHercParis4 45.532µm · 74keV	66.3	A	31.6	20.1	14.8	−0.2	70	169
full root `community-uploads/forrest/volcomp/PHercParis4/volumes/20260310170716-45.532um-11.0m-74keV-masked.zarr` nonzero fraction 0.7119 gradient energy 8.032 dynamic range 148 chunks decoded 21 chunk score spread 0–90 (std 28.6) dead slices 0 signal 31.6/40 — nonzero voxel fraction 0.712 · texture 20.1/30 — gradient energy 8.03 · dynamic 14.8/20 — range 148 · saturation penalty −0.2 · 70 published ink-detection labels on this scroll · 169 segment roots in open data
—	PHerc1299 2.399µm · 78keV	66.0	A	40.0	11.6	14.4	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc1299/volumes/20260309130042-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.9072 gradient energy 4.651 dynamic range 144 chunks decoded 20 chunk score spread 14–78 (std 15.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.907 · texture 11.6/30 — gradient energy 4.65 · dynamic 14.4/20 — range 144 · saturation penalty −0.1
7–22	PHerc0306B 8.640µm · 116keV	65.8	A	33.3	18.0	14.6	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0306B/volumes/20250521133212-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.7487 gradient energy 7.207 dynamic range 146 chunks decoded 24 chunk score spread 1–87 (std 25.0) dead slices 0 signal 33.3/40 — nonzero voxel fraction 0.749 · texture 18.0/30 — gradient energy 7.21 · dynamic 14.6/20 — range 146 · saturation penalty −0.1
—	PHerc1451 2.399µm · 78keV	65.7	A	39.1	12.6	14.1	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc1451/volumes/20260319101107-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.8806 gradient energy 5.041 dynamic range 141 chunks decoded 21 chunk score spread 9–77 (std 19.6) dead slices 0 signal 39.1/40 — nonzero voxel fraction 0.881 · texture 12.6/30 — gradient energy 5.04 · dynamic 14.1/20 — range 141 · saturation penalty −0.1
—	PHerc1545 9.362µm · 113keV	65.6	A	33.9	19.3	12.6	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc1545/volumes/20250821151648-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.7620 gradient energy 7.700 dynamic range 126 chunks decoded 24 chunk score spread 6–88 (std 32.6) dead slices 0 signal 33.9/40 — nonzero voxel fraction 0.762 · texture 19.3/30 — gradient energy 7.70 · dynamic 12.6/20 — range 126 · saturation penalty −0.1
13–33	PHerc0483B 8.640µm · 116keV	64.8	A	36.2	16.4	12.3	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0483B/volumes/20251124083638-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.8136 gradient energy 6.561 dynamic range 123 chunks decoded 24 chunk score spread 1–82 (std 25.0) dead slices 0 signal 36.2/40 — nonzero voxel fraction 0.814 · texture 16.4/30 — gradient energy 6.56 · dynamic 12.3/20 — range 123 · saturation penalty −0.1
13–17	PHerc0846A 9.362µm · 113keV	64.6	A	34.2	13.4	17.0	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0846A/volumes/20250728152254-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.7696 gradient energy 5.376 dynamic range 170 chunks decoded 24 chunk score spread 24–87 (std 14.8) dead slices 0 signal 34.2/40 — nonzero voxel fraction 0.770 · texture 13.4/30 — gradient energy 5.38 · dynamic 17.0/20 — range 170
35–36	PHercParis3 2.400µm · 78keV	64.6	A	40.0	11.4	13.3	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHercParis3/volumes/20260427095331-2.400um-0.2m-78keV-masked.zarr` nonzero fraction 0.9937 gradient energy 4.542 dynamic range 133 chunks decoded 24 chunk score spread 44–82 (std 8.1) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.994 · texture 11.4/30 — gradient energy 4.54 · dynamic 13.3/20 — range 133 · saturation penalty −0.1
20–24	PHerc0814 2.399µm · 78keV	64.4	A	39.6	11.6	13.6	−0.4	0	65
full root `community-uploads/forrest/volcomp/PHerc0814/volumes/20260309142202-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.8916 gradient energy 4.649 dynamic range 136 chunks decoded 24 chunk score spread 1–75 (std 18.1) dead slices 0 signal 39.6/40 — nonzero voxel fraction 0.892 · texture 11.6/30 — gradient energy 4.65 · dynamic 13.6/20 — range 136 · saturation penalty −0.4 · 65 segment roots in open data
28–39	PHerc0343 8.640µm · 116keV	63.9	A	38.9	13.6	11.3	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0343/volumes/20250521140437-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.8756 gradient energy 5.460 dynamic range 113 chunks decoded 24 chunk score spread 0–75 (std 16.7) dead slices 0 signal 38.9/40 — nonzero voxel fraction 0.876 · texture 13.6/30 — gradient energy 5.46 · dynamic 11.3/20 — range 113
13–26	PHerc0175A 8.640µm · 116keV	63.2	A	33.2	16.0	14.0	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0175A/volumes/20250521115057-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.7472 gradient energy 6.393 dynamic range 140 chunks decoded 24 chunk score spread 12–87 (std 22.1) dead slices 0 signal 33.2/40 — nonzero voxel fraction 0.747 · texture 16.0/30 — gradient energy 6.39 · dynamic 14.0/20 — range 140
39–42	PHerc0814 1.129µm · 59keV	63.1	A	39.1	8.1	16.9	−1.0	0	65
full root `community-uploads/forrest/volcomp/PHerc0814/volumes/20260521123630-1.129um-0.2m-59keV-masked.zarr` nonzero fraction 0.8795 gradient energy 3.240 dynamic range 169 chunks decoded 24 chunk score spread 25–69 (std 12.2) dead slices 0 signal 39.1/40 — nonzero voxel fraction 0.879 · texture 8.1/30 — gradient energy 3.24 · dynamic 16.9/20 — range 169 · saturation penalty −1.0 · 65 segment roots in open data
34	PHercParis4 1.129µm · 78keV	63.0	A	40.0	8.3	14.9	−0.2	70	169
full root `community-uploads/forrest/volcomp/PHercParis4/volumes/20260608103018-1.129um-0.2m-78keV-masked.zarr` nonzero fraction 0.9066 gradient energy 3.318 dynamic range 149 chunks decoded 24 chunk score spread 45–75 (std 8.3) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.907 · texture 8.3/30 — gradient energy 3.32 · dynamic 14.9/20 — range 149 · saturation penalty −0.2 · 70 published ink-detection labels on this scroll · 169 segment roots in open data
19–31	PHerc0175B 8.640µm · 116keV	62.8	A	33.7	15.3	13.9	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0175B/volumes/20250521125822-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.7576 gradient energy 6.136 dynamic range 139 chunks decoded 24 chunk score spread 0–76 (std 20.1) dead slices 0 signal 33.7/40 — nonzero voxel fraction 0.758 · texture 15.3/30 — gradient energy 6.14 · dynamic 13.9/20 — range 139 · saturation penalty −0.1
—	PHerc0009B 2.401µm · 77keV	62.1	A	40.0	9.8	12.7	−0.5	0	30
full root `community-uploads/forrest/volcomp/PHerc0009B/volumes/20260319104112-2.401um-0.3m-77keV-masked.zarr` nonzero fraction 0.9691 gradient energy 3.933 dynamic range 127 chunks decoded 18 chunk score spread 25–72 (std 13.1) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.969 · texture 9.8/30 — gradient energy 3.93 · dynamic 12.7/20 — range 127 · saturation penalty −0.5 · 30 segment roots in open data
16–25	PHerc1447 8.640µm · 116keV	61.7	A	37.6	13.2	10.8	−0.0	0	15
full root `community-uploads/forrest/volcomp/PHerc1447/volumes/20250521151220-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.8464 gradient energy 5.293 dynamic range 108 chunks decoded 24 chunk score spread 1–72 (std 18.1) dead slices 0 signal 37.6/40 — nonzero voxel fraction 0.846 · texture 13.2/30 — gradient energy 5.29 · dynamic 10.8/20 — range 108 · 15 segment roots in open data
5–19	PHerc0125 9.362µm · 113keV	61.1	A	35.8	13.3	12.1	−0.2	0	0
full root `community-uploads/forrest/volcomp/PHerc0125/volumes/20250821151825-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.8052 gradient energy 5.331 dynamic range 121 chunks decoded 24 chunk score spread 6–83 (std 21.3) dead slices 0 signal 35.8/40 — nonzero voxel fraction 0.805 · texture 13.3/30 — gradient energy 5.33 · dynamic 12.1/20 — range 121 · saturation penalty −0.2
37–41	PHerc0139 2.399µm · 78keV	61.1	A	40.0	9.0	12.2	−0.1	0	114
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20260102150214-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.9018 gradient energy 3.619 dynamic range 122 chunks decoded 24 chunk score spread 5–72 (std 16.9) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.902 · texture 9.0/30 — gradient energy 3.62 · dynamic 12.2/20 — range 122 · saturation penalty −0.1 · 114 segment roots in open data
—	PHerc1667 2.399µm · 78keV	60.5	A	40.0	8.4	12.3	−0.2	0	46
full root `community-uploads/forrest/volcomp/PHerc1667/volumes/20251217075048-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.9225 gradient energy 3.355 dynamic range 123 chunks decoded 24 chunk score spread 5–76 (std 17.2) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.922 · texture 8.4/30 — gradient energy 3.36 · dynamic 12.3/20 — range 123 · saturation penalty −0.2 · 46 segment roots in open data
38–40	PHercParis4 2.400µm · 78keV	60.5	A	40.0	10.5	12.3	−2.4	70	169
full root `community-uploads/forrest/volcomp/PHercParis4/volumes/20260411134726-2.400um-0.2m-78keV-masked.zarr` nonzero fraction 0.9449 gradient energy 4.212 dynamic range 123 chunks decoded 24 chunk score spread 12–75 (std 13.1) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.945 · texture 10.5/30 — gradient energy 4.21 · dynamic 12.3/20 — range 123 · saturation penalty −2.4 · 70 published ink-detection labels on this scroll · 169 segment roots in open data
7–10	PHerc0257 9.362µm · 113keV	59.2	B	37.7	11.7	9.9	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0257/volumes/20250821151750-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.8486 gradient energy 4.687 dynamic range 99 chunks decoded 24 chunk score spread 0–84 (std 22.1) dead slices 0 signal 37.7/40 — nonzero voxel fraction 0.849 · texture 11.7/30 — gradient energy 4.69 · dynamic 9.9/20 — range 99 · saturation penalty −0.1
30–37	PHerc0268 8.640µm · 116keV	58.6	B	36.9	11.6	10.3	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0268/volumes/20251110183117-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.8302 gradient energy 4.628 dynamic range 103 chunks decoded 24 chunk score spread 6–80 (std 20.2) dead slices 0 signal 36.9/40 — nonzero voxel fraction 0.830 · texture 11.6/30 — gradient energy 4.63 · dynamic 10.3/20 — range 103 · saturation penalty −0.1
30–32	PHerc0846A 2.403µm · 77keV	58.4	B	36.0	11.8	14.5	−3.8	0	0
full root `community-uploads/forrest/volcomp/PHerc0846A/volumes/20260319102732-2.403um-0.2m-77keV-masked.zarr` nonzero fraction 0.8095 gradient energy 4.727 dynamic range 145 chunks decoded 24 chunk score spread 1–78 (std 24.2) dead slices 0 signal 36.0/40 — nonzero voxel fraction 0.809 · texture 11.8/30 — gradient energy 4.73 · dynamic 14.5/20 — range 145 · saturation penalty −3.8
—	PHercMANBp 1.129µm · 59keV	58.4	B	40.0	6.8	11.7	−0.2	0	0
full root `community-uploads/forrest/volcomp/PHercMANBp/volumes/20260427100434-1.129um-0.2m-59keV-masked.zarr` nonzero fraction 0.9996 gradient energy 2.717 dynamic range 117 chunks decoded 6 chunk score spread 45–68 (std 8.6) dead slices 0 signal 40.0/40 — nonzero voxel fraction 1.000 · texture 6.8/30 — gradient energy 2.72 · dynamic 11.7/20 — range 117 · saturation penalty −0.2
6–14	PHerc0191 9.362µm · 113keV	58.0	B	33.2	13.9	11.0	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0191/volumes/20250821151635-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.7473 gradient energy 5.561 dynamic range 110 chunks decoded 24 chunk score spread 7–89 (std 29.5) dead slices 0 signal 33.2/40 — nonzero voxel fraction 0.747 · texture 13.9/30 — gradient energy 5.56 · dynamic 11.0/20 — range 110 · saturation penalty −0.1
12–29	PHerc0800 8.640µm · 116keV	56.9	B	33.1	12.2	11.7	−0.0	0	6
full root `community-uploads/forrest/volcomp/PHerc0800/volumes/20250521135224-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.7438 gradient energy 4.860 dynamic range 117 chunks decoded 24 chunk score spread 0–74 (std 23.1) dead slices 0 signal 33.1/40 — nonzero voxel fraction 0.744 · texture 12.2/30 — gradient energy 4.86 · dynamic 11.7/20 — range 117 · 6 segment roots in open data
36–37	PHerc0483A 8.640µm · 116keV	56.5	B	30.0	13.0	13.6	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0483A/volumes/20250521140913-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.6739 gradient energy 5.219 dynamic range 136 chunks decoded 24 chunk score spread 11–81 (std 24.3) dead slices 0 signal 30.0/40 — nonzero voxel fraction 0.674 · texture 13.0/30 — gradient energy 5.22 · dynamic 13.6/20 — range 136 · saturation penalty −0.1
—	PHerc0343P 8.640µm · 116keV	55.9	B	28.3	14.4	13.4	−0.1	0	16
full root `community-uploads/forrest/volcomp/PHerc0343P/volumes/20250521134555-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.6362 gradient energy 5.749 dynamic range 134 chunks decoded 24 chunk score spread 0–87 (std 30.9) dead slices 0 signal 28.3/40 — nonzero voxel fraction 0.636 · texture 14.4/30 — gradient energy 5.75 · dynamic 13.4/20 — range 134 · saturation penalty −0.1 · 16 segment roots in open data
—	PHerc0500P2 2.215µm · 111keV	55.9	B	40.0	5.9	10.1	−0.1	0	138
full root `community-uploads/forrest/volcomp/PHerc0500P2/volumes/20250526151718-2.215um-0.4m-111keV-masked.zarr` nonzero fraction 1.0000 gradient energy 2.370 dynamic range 101 chunks decoded 13 chunk score spread 43–62 (std 5.8) dead slices 0 signal 40.0/40 — nonzero voxel fraction 1.000 · texture 5.9/30 — gradient energy 2.37 · dynamic 10.1/20 — range 101 · saturation penalty −0.1 · 138 segment roots in open data
—	PHercMAN5 2.399µm · 78keV	55.8	B	36.6	7.6	12.9	−1.3	0	0
full root `community-uploads/forrest/volcomp/PHercMAN5/volumes/20260311104824-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.8235 gradient energy 3.048 dynamic range 129 chunks decoded 24 chunk score spread 8–71 (std 20.0) dead slices 0 signal 36.6/40 — nonzero voxel fraction 0.824 · texture 7.6/30 — gradient energy 3.05 · dynamic 12.9/20 — range 129 · saturation penalty −1.3
10–38	PHerc1451 8.640µm · 116keV	54.3	B	26.4	12.7	15.2	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc1451/volumes/20250521151225-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.5943 gradient energy 5.091 dynamic range 152 chunks decoded 24 chunk score spread 8–86 (std 23.8) dead slices 0 signal 26.4/40 — nonzero voxel fraction 0.594 · texture 12.7/30 — gradient energy 5.09 · dynamic 15.2/20 — range 152
11–24	PHerc0490B 8.640µm · 116keV	53.9	B	29.9	11.1	12.9	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0490B/volumes/20250521151215-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.6722 gradient energy 4.452 dynamic range 129 chunks decoded 24 chunk score spread 12–86 (std 19.2) dead slices 0 signal 29.9/40 — nonzero voxel fraction 0.672 · texture 11.1/30 — gradient energy 4.45 · dynamic 12.9/20 — range 129
43–44	PHerc0139 1.129µm · 59keV	53.4	B	36.1	3.3	14.2	−0.2	0	114
full root `community-uploads/forrest/volcomp/PHerc0139/volumes/20260413113053-1.129um-0.2m-59keV-masked.zarr` nonzero fraction 0.8115 gradient energy 1.339 dynamic range 142 chunks decoded 24 chunk score spread 6–64 (std 12.6) dead slices 0 signal 36.1/40 — nonzero voxel fraction 0.811 · texture 3.3/30 — gradient energy 1.34 · dynamic 14.2/20 — range 142 · saturation penalty −0.2 · 114 segment roots in open data
17–23	PHerc0211 9.362µm · 113keV	52.7	B	31.6	11.0	10.2	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHerc0211/volumes/20250821151803-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.7106 gradient energy 4.390 dynamic range 102 chunks decoded 24 chunk score spread 8–86 (std 27.6) dead slices 0 signal 31.6/40 — nonzero voxel fraction 0.711 · texture 11.0/30 — gradient energy 4.39 · dynamic 10.2/20 — range 102 · saturation penalty −0.1
42–43	PHerc1667 1.129µm · 59keV	51.8	B	40.0	2.7	10.5	−1.3	0	46
full root `community-uploads/forrest/volcomp/PHerc1667/volumes/20260323082859-1.129um-0.2m-59keV-masked.zarr` nonzero fraction 0.9524 gradient energy 1.088 dynamic range 105 chunks decoded 24 chunk score spread 42–64 (std 8.4) dead slices 0 signal 40.0/40 — nonzero voxel fraction 0.952 · texture 2.7/30 — gradient energy 1.09 · dynamic 10.5/20 — range 105 · saturation penalty −1.3 · 46 segment roots in open data
—	PHerc0500P2 9.362µm · 113keV	51.1	B	28.0	12.6	10.6	−0.1	0	138
full root `community-uploads/forrest/volcomp/PHerc0500P2/volumes/20250820143440-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.6296 gradient energy 5.042 dynamic range 106 chunks decoded 24 chunk score spread 0–88 (std 32.3) dead slices 0 signal 28.0/40 — nonzero voxel fraction 0.630 · texture 12.6/30 — gradient energy 5.04 · dynamic 10.6/20 — range 106 · saturation penalty −0.1 · 138 segment roots in open data
16–28	PHerc0490A 8.640µm · 116keV	49.2	B	25.6	9.1	14.5	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0490A/volumes/20250521151210-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.5759 gradient energy 3.651 dynamic range 145 chunks decoded 24 chunk score spread 8–74 (std 22.0) dead slices 0 signal 25.6/40 — nonzero voxel fraction 0.576 · texture 9.1/30 — gradient energy 3.65 · dynamic 14.5/20 — range 145
45	PHerc0500P2 20250821110041-0.500um-masked	45.5	B	40.0	2.1	3.4	−0.0	0	138
full root `community-uploads/forrest/volcomp/PHerc0500P2/volumes/20250821110041-0.500um-masked.zarr` nonzero fraction 1.0000 gradient energy 0.834 dynamic range 34 chunks decoded 24 chunk score spread 41–61 (std 6.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 1.000 · texture 2.1/30 — gradient energy 0.83 · dynamic 3.4/20 — range 34 · 138 segment roots in open data
46	PHerc0500P2 0.550µm · 65keV	45.5	B	40.0	2.1	3.4	−0.0	0	138
full root `community-uploads/forrest/volcomp/PHerc0500P2/volumes/20250821110041-0.550um-0.1m-65keV-masked.zarr` nonzero fraction 1.0000 gradient energy 0.834 dynamic range 34 chunks decoded 24 chunk score spread 41–61 (std 6.0) dead slices 0 signal 40.0/40 — nonzero voxel fraction 1.000 · texture 2.1/30 — gradient energy 0.83 · dynamic 3.4/20 — range 34 · 138 segment roots in open data
—	PHerc0009B 8.640µm · 116keV	44.5	C	23.8	8.6	12.2	−0.0	0	30
full root `community-uploads/forrest/volcomp/PHerc0009B/volumes/20250521125136-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.5346 gradient energy 3.429 dynamic range 122 chunks decoded 24 chunk score spread 8–78 (std 21.8) dead slices 0 signal 23.8/40 — nonzero voxel fraction 0.535 · texture 8.6/30 — gradient energy 3.43 · dynamic 12.2/20 — range 122 · 30 segment roots in open data
—	PHercMANBp 2.399µm · 78keV	42.3	C	23.5	8.9	9.9	−0.1	0	0
full root `community-uploads/forrest/volcomp/PHercMANBp/volumes/20251216152116-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.5296 gradient energy 3.542 dynamic range 100 chunks decoded 8 chunk score spread 6–78 (std 32.7) dead slices 0 signal 23.5/40 — nonzero voxel fraction 0.530 · texture 8.9/30 — gradient energy 3.54 · dynamic 9.9/20 — range 100 · saturation penalty −0.1
41–44	PHerc1218 8.640µm · 116keV	41.6	C	21.7	7.6	12.3	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc1218/volumes/20250521120456-8.640um-1.2m-116keV-masked.zarr` nonzero fraction 0.4888 gradient energy 3.041 dynamic range 123 chunks decoded 24 chunk score spread 2–70 (std 21.2) dead slices 0 signal 21.7/40 — nonzero voxel fraction 0.489 · texture 7.6/30 — gradient energy 3.04 · dynamic 12.3/20 — range 123
—	PHerc0500P2 4.317µm · 111keV	38.9	C	20.2	5.3	16.2	−2.9	0	138
full root `community-uploads/forrest/volcomp/PHerc0500P2/volumes/20250528085330-4.317um-1.2m-111keV-masked.zarr` nonzero fraction 0.4548 gradient energy 2.130 dynamic range 162 chunks decoded 24 chunk score spread 3–64 (std 17.0) dead slices 0 signal 20.2/40 — nonzero voxel fraction 0.455 · texture 5.3/30 — gradient energy 2.13 · dynamic 16.2/20 — range 162 · saturation penalty −2.9 · 138 segment roots in open data
12–21	PHerc0826 9.362µm · 113keV	37.2	C	20.0	6.8	10.3	−0.0	0	0
full root `community-uploads/forrest/volcomp/PHerc0826/volumes/20250821151701-9.362um-1.2m-113keV-masked.zarr` nonzero fraction 0.4509 gradient energy 2.737 dynamic range 103 chunks decoded 24 chunk score spread 0–80 (std 29.2) dead slices 0 signal 20.0/40 — nonzero voxel fraction 0.451 · texture 6.8/30 — gradient energy 2.74 · dynamic 10.3/20 — range 103
26–40	PHerc0841 9.366µm · 113keV	31.4	C	20.5	3.7	7.2	−0.0	0	6
full root `community-uploads/forrest/volcomp/PHerc0841/volumes/20250821151531-9.366um-1.2m-113keV-masked.zarr` nonzero fraction 0.4614 gradient energy 1.460 dynamic range 72 chunks decoded 24 chunk score spread 0–87 (std 25.1) dead slices 0 signal 20.5/40 — nonzero voxel fraction 0.461 · texture 3.7/30 — gradient energy 1.46 · dynamic 7.2/20 — range 72 · 6 segment roots in open data
47	PHerc0343P 2.215µm · 111keV	30.5	C	40.0	4.5	11.1	−25.0	0	16
full root `community-uploads/forrest/volcomp/PHerc0343P/volumes/20260304131111-2.215um-0.4m-111keV-masked.zarr` nonzero fraction 1.0000 gradient energy 1.790 dynamic range 111 chunks decoded 24 chunk score spread 21–64 (std 12.3) dead slices 0 signal 40.0/40 — nonzero voxel fraction 1.000 · texture 4.5/30 — gradient energy 1.79 · dynamic 11.1/20 — range 111 · saturation penalty −25.0 · 16 segment roots in open data
48	PHercMANB 2.399µm · 78keV	13.5	C	32.8	4.0	6.8	−30.0	0	0
full root `community-uploads/forrest/volcomp/PHercMANB/volumes/20260323091048-2.399um-0.2m-78keV-masked.zarr` nonzero fraction 0.7371 gradient energy 1.581 dynamic range 68 chunks decoded 24 chunk score spread 1–71 (std 24.5) dead slices 11 signal 32.8/40 — nonzero voxel fraction 0.737 · texture 4.0/30 — gradient energy 1.58 · dynamic 6.8/20 — range 68 · dead slices: 11

Scan-health score (one diagnostic)

Signal presence (nonzero voxel fraction)	40
Texture / gradient energy	30
Dynamic range	20
Saturation penalty	−10
Dead-slice penalty	−25

The dead-slice detector is deliberately strict: a densely populated chunk holding a zero plane whose both neighbors are densely populated. A naive rule false-positives on mask geometry — that version was caught and corrected before shipping.

Honest scope. This is a triage signal, not a readability claim. It does not detect ink and does not predict which scroll will read first. It tells you which volumes have the healthiest voxels, so segmentation and labeling effort goes where the data is strongest. The weights are a judgment call, published with every score.

Campaign request: up to 12 × 128³ chunks per volume, spread across the shard grid. Sparse volumes may decode fewer; the achieved count and sampling provenance stay attached to every result. Decoded with the vendored libvolcomp via zarr-pyramid-audit.

One suite, two halves

zarr-pyramid-audit

Don't train on lies. Corruption detection for OME-Zarr pyramids — header-only audits, a publish-time gate, and a sampled chunk-content probe. Caught a live defective pyramid in the open-data S3 bucket.

repo · dashboard

scrollq-health

Runs the integrity audit and the quality score on any volume and issues one verdict: TRAIN / CAUTION / DO NOT TRAIN. Fails closed: unreadable or missing integrity evidence is DO NOT TRAIN, never a pass.

scrollq-health --root <volume>