Showcase samples: make the 4 front-page images locate themselves on the globe

[rdhyee]

Context

The homepage has a Showcase section with four "charismatic" sample thumbnails:

• Diamond — IGSN:DIA0000YL (SESAR, Brazil, 2019)
• Fossil coral — IGSN:IEGIL000C (SESAR, Cayman Islands, ~10,000 BCE)
• Paracirrhites arcatus fish — ark:65665/337856f1a655e4ad78b1ef10a16dfb6e3 (Smithsonian, French Polynesia, 2006)
• Red-figure askoi — ark:28722/r2p24/vdm_19600211 (OpenContext, Murlo Italy, late 4th–early 3rd c. BCE)

Currently each thumbnail links to the sample's source record at DOI/ARK. There's no way from the homepage to see where on the globe the sample is, despite the four being chosen in part for geographic spread across four continents.

The hero above them is a pre-rendered animated WebP — a rotating Earth. It's decorative, not interactive.

Goal

Let the showcase act as an entry point into the Interactive Explorer, anchored at the location of the clicked sample. Three candidate designs, in increasing ambition:

Option A — Cheapest: deep-link each thumbnail into the Explorer (~30 min)

The Interactive Explorer already reads a pid URL hash parameter (see `readHash()` and `gs.selectedPid` in tutorials/progressive_globe.qmd). Wrapping each showcase thumbnail in a secondary "View on globe →" link would open the Explorer with that sample pre-selected, its card populated, and the camera flown to its coordinates.

• Pro: near-zero code; reuses existing deep-link machinery; preserves the current DOI/ARK link for the scholarly source record.
• Con: leaves the homepage itself still static; adds a second link-target per thumbnail (need to handle UX — click image = source record, click overlay button = globe).

Before shipping, verify all four pids exist as-is in samples_map_lite.parquet (the file the Explorer queries for sample-level data).

Option B — Mid: hover/click morph, stay on homepage (~1 day)

Replace the static "View on globe" link with an inline sample locator. On click, the thumbnail card expands to show a small static map (Leaflet or even just a marker on an equirectangular Natural Earth image) with a pin at the sample's coordinates, plus a "Open in Interactive Explorer" CTA.

• Pro: gives spatial context without forcing a page jump; homepage becomes slightly interactive.
• Con: new dependency (Leaflet) or a build-time step to generate the four static maps; doubles the visual language of the homepage.

Option C — Most ambitious: replace the WebP hero with a live mini-globe (~2–3 days)

Retire the pre-rendered animated WebP. Embed a small live Cesium mini-globe on the landing page, configured to:

• spin at the same cadence as the current WebP,

• render the four showcase samples as labeled pins,

• on click: fly to the sample and open its card, or teleport the user into the full Interactive Explorer with that pid.

• Pro: the homepage becomes the Explorer's teaser, not a separate thing to explain; the thumbnails and globe align in one interaction.

• Con: adds Cesium (~1.5 MB JS + terrain tiles) to the landing-page critical path; will measurably slow first-paint for every visitor, including ones who never click Explorer; accessibility / no-JS fallback story needed; the animated WebP is cached and immutable today (zero runtime cost).

Recommendation

Start with Option A because it's reversible and cheap — and almost certainly what most users want. It verifies the deep-link flow end-to-end (is the card populated correctly? does fly-to work reliably for all four pids across sources?) before we invest in B or C.

If A lands cleanly and users engage with it, the question for B vs C becomes: is the homepage's job to be the Explorer (C), or to advertise the Explorer (A/B)? Current site architecture treats these as distinct pages (hero in nav leads to Interactive Explorer as a separate destination), which argues for A. If we later want the hero to double as the product's first interactive moment, C.

Implementation notes for Option A

• Edit index.qmd showcase block. For each of the four thumbnails, either:
  • replace the DOI/ARK link target with /tutorials/progressive_globe.html#pid=<pid>&v=1, OR
  • keep the DOI/ARK link on the image and add a small "📍 locate on globe" caption/button below that links to the Explorer.
• Confirm each pid resolves in samples_map_lite.parquet:

  SELECT pid, label, source, latitude, longitude
  FROM read_parquet('https://data.isamples.org/isamples_202601_samples_map_lite.parquet')
  WHERE pid IN ('IGSN:DIA0000YL', 'IGSN:IEGIL000C',
                'ark:65665/337856f1a655e4ad78b1ef10a16dfb6e3',
                'ark:28722/r2p24/vdm_19600211')

• Smoke test: visit each constructed URL, confirm camera flies to the sample, card shows, source badge matches.

Open questions

1. Do we want click-on-image to still go to the scholarly source record (DOI/ARK) or to the Explorer? Default answer: keep image → source record, add secondary CTA for Explorer. Either is defensible.
2. For Option C: is the existing WebP good enough that we should leave it alone, or is a live globe worth the perf hit?
3. If we add 4 pins to a future live hero, do we also want sample thumbnails to appear as inset badges on the globe itself, or just stay in the showcase row below?

[rdhyee]

Parquet verification results

Queried isamples_202601_wide.parquet and samples_map_lite.parquet for each showcase pid. Summary: only 1 of 4 is directly linkable today.

Showcase	Exact pid in parquet?	Details
Coral IGSN:IEGIL000C	✅ Yes	SESAR, label "18LCI-5", coords (19.67, -80.10), Cayman Islands. Direct deep-link will work.
Diamond IGSN:DIA0000YL	❌ No	The specific pid is absent from the Jan 2026 snapshot, but SESAR has 10+ other IGSN:DIA* diamonds (Russia/Mirny, DRC/Mbuji-Mayi).
Fish ark:65665/337856f1a655e4ad78b1ef10a16dfb6e3	❌ No	That ark is a Smithsonian search/image ark, a different namespace from the sample records in the parquet (which use UUID-style ark:/65665/3f9748e2c-...). But Paracirrhites arcatus by label DOES exist: ark:/65665/377885683-222f-42da-a879-df7910849c42 (label "Paracirrhites arcatus AH7ZV35"). A different specimen, same species.
Askoi ark:28722/r2p24/vdm_19600211	❌ No	OpenContext pids in the parquet look like ark:/28722/k2zg70v1g; the r2p24/vdm_* format is a legacy/media ark. But 3 OpenContext samples exist at the exact Murlo coordinates (43.15, 11.40) — ark:/28722/k2x92md15, ark:/28722/k20s06419, ark:/28722/k2ms44w6v — labels "Object PC 19960081/19960007/19960101".

So what?

The showcase thumbnails link to display/media URLs, not to iSamples sample records. The DOI/ARK chosen for visual appeal isn't the canonical sample identifier. Three of four won't deep-link as-is.

Two paths forward:

Path 1: Update the showcase to use pids that exist in the parquet

Swap pids for matched records:

• Coral → keep IGSN:IEGIL000C (already works)
• Fish → switch to ark:/65665/377885683-222f-42da-a879-df7910849c42 (Paracirrhites arcatus AH7ZV35)
• Askoi → pick one of the three Murlo OpenContext samples (or a different archaeology sample with a known good pid)
• Diamond → pick from the 10+ SESAR DIA-prefix samples (though the thumbnail image is of DIA0000YL specifically — would need matching thumbnail)

Tradeoff: the visual language changes — the thumbnail image might not match the deep-linked sample exactly. User viewing the "arcatus" photo clicks through and sees a row for "AH7ZV35" which is a different individual.

Path 2: Keep the thumbnails as-is; deep-link to a nearby cluster

Instead of linking to the specific sample, link to the region or cluster that contains similar material. For the askoi: fly to Murlo (43.15, 11.40) and let the user pan/zoom; the user lands in the right neighborhood even if they can't click exactly that pot.

Tradeoff: less precise, but preserves visual authenticity of the showcase.

Path 3: Fix the underlying pid data

Figure out why the "canonical" pids used on isamples.org/samples (or wherever the showcase was curated) don't match the parquet. If it's a routing/crosswalk problem, fix it at the pipeline level. This is the right long-term answer but out of scope for a homepage link fix.

Recommended next step

Choose between Path 1 (update showcase pids) and Path 2 (link to neighborhoods). Path 1 gives sharper UX but requires new thumbnails or accepting mismatches; Path 2 preserves the showcase's charisma but deep-links to approximate locations. Either way, the work is ~30 min.

Reposting a summary of pids I'd use for Path 1 so it's copy-pasteable:

Coral    → IGSN:IEGIL000C
Fish     → ark:/65665/377885683-222f-42da-a879-df7910849c42
Askoi    → ark:/28722/k2x92md15   (or one of the other two Murlo samples)
Diamond  → IGSN:DIA000004  (or any of the 10 DIA-prefix samples)

[rdhyee]

Additional findings: thumbnails aren't in the parquet either

Pulled on the thread of "find adjacent beautiful images":

1. The parquet has a thumbnail_url column — but it's empty for all 6.7M samples

SELECT n as source, COUNT(*) AS total, COUNT(*) FILTER (WHERE thumbnail_url IS NOT NULL AND thumbnail_url <> '') AS with_thumb
FROM read_parquet('isamples_202601_wide.parquet')
WHERE otype = 'MaterialSampleRecord'
GROUP BY 1

Returns zero thumbnails across SESAR, OpenContext, GEOME, Smithsonian. So we can't query the parquet to find "samples near Murlo with pretty pictures" — that column is a schema placeholder that the PQG pipeline never fills. Worth filing separately against the iSamples data pipeline (a feature the schema advertises but doesn't deliver).

2. Images exist upstream, but each source is a different hunt

Source	Where images live	Accessibility
OpenContext	/subjects/<uuid>.json → oc-gen:has-obs[*] → linked media records via /search/?type=media	Structured but requires traversal. Poggio Civitate (Murlo) has 209,234 media items upstream.
Smithsonian NMNH	iDigBio indexes fish specimens with hasMedia:true → returns dwc:associatedMedia UUIDs → resolvable to images. Paracirrhites arcatus has at least 5 media records.	Workable via iDigBio API.
SESAR	Typically no images (geological samples).	Not a useful source for beautiful thumbnails.
GEOME	Some specimen photos via FIMS/BageLive.	Sparse.

3. Automating "beautiful adjacent" at scale isn't cheap

To auto-suggest replacements I'd need to, for each showcase item:

• Query the relevant source's media API
• Filter to records that map to a pid present in our Jan 2026 parquet
• Rank by some "beautiful" proxy (photo present, resolution ≥ N, aspect close to square, curation notes)
• Download candidates and eyeball

That's a day's work for what amounts to a small, hand-curated gallery. Not a great ROI.

Proposed path

Two concrete, shippable options:

Option A (cheap, ~1 hr): Curate replacement pids manually

I'd narrow the candidate set to a handful per source, and you (or whoever curated the originals) pick the final 4 based on visual appeal. Starter candidates I already confirmed exist in the parquet with coordinates:

• Murlo area archaeology: ark:/28722/k2x92md15 / ark:/28722/k20s06419 / ark:/28722/k2ms44w6v (Poggio Civitate 'Object PC 1996...' series, 43.15,11.40). Images exist upstream at OpenContext — not yet downloaded.
• Paracirrhites arcatus: ark:/65665/377885683-222f-42da-a879-df7910849c42 — iDigBio has media for this species; haven't confirmed this specific specimen has a photo.
• Diamond: 10+ IGSN:DIA* Russia/Mirny + DRC diamonds. SESAR doesn't serve photos, so we'd reuse the existing DIA0000YL image (acknowledging it's representative, not exact).
• Coral: keep current (IGSN:IEGIL000C already matches).

Option B (cheaper, ~30 min): Link showcase to "neighborhood"

Keep the beautiful thumbnails untouched. For each, make the click-through link go to a representative location on the globe rather than a specific sample:

Showcase	Link target (approx coord or nearby pid)
Diamond	SESAR diamond cluster near Russia/Mirny, any IGSN:DIA* pid
Coral	IGSN:IEGIL000C (exact match, no compromise needed)
Fish	ark:/65665/377885683-222f-42da-a879-df7910849c42 (P. arcatus, different specimen)
Askoi	One of the three Murlo objects at (43.15, 11.40)

In Option B the URL caveats itself: a footnote like "click to see this sample's neighborhood on the interactive globe" softens the mismatch.

Recommendation

Option B ships today with no new imagery. Option A is the right next step if/when we're willing to invest in curated thumbnails — but only after we decide whether the iSamples data pipeline should be populating thumbnail_url in the first place (the deeper structural fix).

[rdhyee]

Scope decision

After the flow trace in #131, narrowing this to the iSamples-indexed pool:

• Stick with iSamples ingest for the showcase — don't try to reach beyond what's in our parquet.
• Openness to natural expansion: if during curation we find the existing iSamples ingest already contains a sample adjacent to the showcase theme but missing from our 4-picks, we can swap in a replacement. We're not opening the door to pulling samples that aren't in our indexed data (that's a separate ingest-scope conversation, not this issue).

Thumbnail availability after #131 fix

Once the Option-1 LEFT JOIN from #131 ships, the wide parquet gains ~47.7 K OpenContext thumbnails. Other sources (SESAR, GEOME, Smithsonian) won't have thumbnails in the near term — that's deferred until we either harvest ourselves or build a "data-dump intake" process for the PIs at each org.

What to do with the 4 showcase items

Three of four currently don't resolve to a parquet pid. Options ranked:

1. Coral — keep IGSN:IEGIL000C (exact match exists, will deep-link perfectly).
2. Askoi — swap to one of the 3 Murlo OpenContext samples (ark:/28722/k2x92md15 etc.). After Data pipeline: thumbnail_url column is empty across all 6.7M samples #131 lands, these will have thumbnails populated from Eric's data. We can pick one that is visually comparable to the askoi.
3. Fish — use ark:/65665/377885683-222f-42da-a879-df7910849c42 (Paracirrhites arcatus AH7ZV35). No thumbnail today; Smithsonian thumbnails would require the "data dump / self-harvest" path.
4. Diamond — pick from the 10+ IGSN:DIA* samples in SESAR. No thumbnails; SESAR rarely has photos. We could either keep the DIA0000YL image as a stand-in (labeled "representative") or drop the diamond slot.

This issue can be addressed in phases:

• Phase A (after Data pipeline: thumbnail_url column is empty across all 6.7M samples #131 Option 1): swap coral link-target fix + the Murlo swap for the askoi. Fish and diamond remain visually-only with no deep-link.
• Phase B: if/when we have thumbnails for SESAR + Smithsonian, make all 4 deep-link.

Moving this issue to blocked-by #131.