Charismatic samples audit: swap or annotate front-page features

[rdhyee]

Summary

Raymond observed that most of the four "charismatic" featured samples in the ## Showcase: Real Samples from the Collection section of index.qmd are illustrative, not actual records in the iSamples Zenodo export. This audit verifies each against the Jan 2026 wide parquet (zenodo_wide_2026-01-09.parquet, 20M entities).

TL;DR: 1 of 4 is in-collection. None have thumbnails yet (sidecar not merged). The section header overstates.

Audit results

#	Image (local asset)	Visible label on page	Linked identifier	Classification	In Zenodo wide?	Thumbnail?	Recommendation
1	assets/IGSN_10.58052_DIA0000YL.png	Diamond, Brazil, 2019-06-11	doi.org/10.58052/DIA0000YL → IGSN:DIA0000YL	not-in-collection	No (no DIA0000YL, range is DIA000004-DIA000009+)	N/A	Swap
2	assets/IGSN_10.58052_IEGIL000C.png	Fossil coral, Cayman Islands, 10000 BCE	doi.org/10.58052/IEGIL000C → IGSN:IEGIL000C	in-collection-no-image	Yes (SESAR, real label 18LCI-5, 19.67N 80.10W)	None (SESAR 0% thumbnail coverage)	Keep PID, fix caption + needs SESAR sidecar
3	assets/ark_65665_337856f1a655e4ad78b1ef10a16dfb6e3.png	Paracirrhites arcatus, French Polynesia, 2006-03-10	ark:65665/337856f1... (Smithsonian NAAN)	not-in-collection	No (hash 337856f1... not in 1.18M Smithsonian records)	N/A	Swap
4	assets/ark_28722_r2p24_vdm_19600211.png	Red-figure askoi, late-4C-early-3C BCE, Murlo, Italy	ark:28722/r2p24/vdm_19600211 (OpenContext VdM)	not-in-collection	No (VdM range in export is vdm_20060001-vdm_20175070, 1755 records — 19600211 is pre-2006 catalog)	N/A	Swap

Net: 1 real (coral), 3 illustrative. Even the one real sample has a descriptive caption ("Fossil coral") that doesn't match its actual iSamples label (18LCI-5).

Thumbnail coverage (system-wide)

Source	MaterialSampleRecord count	With thumbnail_url	%
SESAR	4,688,386	0	0.00%
OPENCONTEXT	1,064,831	0	0.00%
SMITHSONIAN	322,161	0	0.00%
GEOME	605,554	0	0.00%

The enrichment sidecar pattern (Raymond's endorsed approach, 2026-04-17) has not yet been merged into the published wide parquet — so even the in-collection coral (#2) cannot be rendered with its own real thumbnail. Until sidecars ship, any showcase must use either (a) images pulled live from source repositories (opencontext.org, geosamples.org, collections.nmnh.si.edu), or (b) remain illustrative with a disclosure.

Recommended swaps (real, in-collection candidates)

For #1 Diamond → real SESAR IGSN:DIA* diamond

• IGSN:DIA000004 — label: Russia_Mirny_03072014_52768, Mirny diamond mine, Russia (66.4N, 129.2E). Iconic locality.
• IGSN:DIA000009 — label: DRC_Mbuji-Mayi (Miba)_03072014_55864, DRC (-6.2N, 23.0E). Geographic diversity.

For #3 Fish (Paracirrhites arcatus) → same species or sibling, real PID

• GEOME ark:/21547/CXs2MParis0001 — Paracirrhites arcatus MParis0001, French Polynesia (-17.48, -149.88). Same species, same region.
• Smithsonian ark:/65665/3f9768717-39ce-4422-a86e-932be0555df7 — Centropyge loriculus AG0IV08 (Flame angelfish), French Polynesia (-17.47, -149.93). Real Smithsonian ARK, charismatic reef fish.

For #4 Red-figure askoi → real OpenContext Murlo vessel

• ark:/28722/r2p24/vdm_20155002 — label: Pottery VdM20155002, Murlo (Poggio Civitate) excavation. Same site, real PID, in VdM 2015 catalog.
• ark:/28722/r2p24/vdm_20155120 — label: Object VdM 20155120, same site.
• Search did not find "Red-figure" or "askos" explicitly in VdM labels — if visual fidelity to a red-figure vessel matters more than site-provenance, we may need to pull from a different OC project (e.g., Etruscan collections).

Sidecar rollout implications

Only one of the four current features (coral) is in-collection-no-image. So the sidecar story here is narrower than I feared: even if we swap #1/#3/#4 to real PIDs, none of them would render from the published parquet today — every one would be in-collection-no-image.

Options:

1. Short term (ship this week): Swap the 3 illustrative PIDs to real ones per above; keep local /assets/*.png renderings but update captions + alt text to name the real PID. Add a footnote: "Images are curator-selected illustrations; data records resolve via the linked identifiers."
2. Medium term (sidecar): Prioritize SESAR + GEOME + Smithsonian sidecars so these 4 samples can render from upstream (SESAR has IGSN landing pages with images; Smithsonian NMNH has collection pages; GEOME/OpenContext are partners we can request from). This promotes all four from in-collection-no-image → in-collection-with-image.
3. Long term: After sidecar merge, the Showcase becomes queryable — pick top-N by photographic quality across all sources automatically.

Verification method

import duckdb
WIDE = 'https://data.isamples.org/current/wide.parquet'  # or local
con = duckdb.connect()
con.sql(f\"SELECT pid, label, n as source, thumbnail_url FROM read_parquet('{WIDE}') WHERE pid = 'IGSN:IEGIL000C'\").df()

Also cross-checked via altids / alternate_identifiers arrays and place_name / label ILIKE fuzzy searches. Source: ~/Data/iSample/pqg_refining/zenodo_wide_2026-01-09.parquet (20M rows, 6.7M MaterialSampleRecord).

Source file to edit

index.qmd lines 22-32 (the layout-ncol=4 block with 4 image/link pairs).

Related

• Sidecar pattern discussion: per Raymond's 2026-04-17 endorsement (per-source parquet sidecars keyed by pid, LEFT-JOINed into wide at build time).
• Landscape draft on Project 7: PVTI_lADOBCDimM4BS_FQzgq58y0.

---

Filed as part of the iSamples Grant Closeout (May 2026) site cleanup. Report generated by audit run 2026-04-24.

[rdhyee]

Evidence / provenance

A parallel draft (Codex, 2026-04-24) independently audited the same four front-page samples and reached the same conclusion — preserved here for the record.

The four featured samples

#	Label on front-page link	Image URL	PID shown
1	Diamond, 2019-06-11, Brazil	assets/IGSN_10.58052_DIA0000YL.png	IGSN 10.58052/DIA0000YL
2	Fossil coral, 10000 BCE, Cayman Islands	assets/IGSN_10.58052_IEGIL000C.png	IGSN 10.58052/IEGIL000C
3	Paracirrhites arcatus, 2006-03-10, French Polynesia	assets/ark_65665_337856f1a655e4ad78b1ef10a16dfb6e3.png	ARK ark:65665/337856f1a655e4ad78b1ef10a16dfb6e3
4	Red-figure askoi, 4th-3rd c. BCE, Murlo, Italy	assets/ark_28722_r2p24_vdm_19600211.png	ARK ark:28722/r2p24/vdm_19600211

Local verification (Codex audit)

• index.qmd, index_alt.qmd, and docs/index.html agree on the same four front-page samples.
• Against OpenContext PQG parquet (11.6M rows, 1.1M MaterialSampleRecord): no rows match any of the four identifiers in pid, alternate_identifiers, thumbnail_url, or by distinctive labels (Paracirrhites, Red-figure askoi, Fossil coral, Diamond hits are unrelated OC records).
• data/oc_geocodes_classified.parquet is location-only — cannot verify sample PIDs.
• No local SESAR parquet was available to verify the two IGSN samples; the two ARK samples did not resolve in the local OpenContext PQG snapshot.

The independent parallel audit (this issue's body) reached the matching conclusion with proposed real-PID substitutes. Two independent reads confirm: 1 of 4 real (fossil coral), 3 of 4 illustrative with all images hardcoded as local /assets/*.png files.