Image hero+gallery inventory — Phase 0 of #5 rework

Generated June 2026. Pre-rework snapshot. Read alongside docs/guides/post-332-follow-ups.md §5.

Scope

Eleven detail-panel surfaces that follow a hero+gallery pattern were audited for hero-file coverage, provenance consistency, and refactor remnants.

Per-surface inventory

Surface	Entities on disk	Hero coverage	Pattern
Missions	233 hero files + 58 gallery dirs (old) + 24 gallery dirs (new)	24 missing (Slice A/B/C)	missions/<id>.jpg for hero; gallery split between missions/<id>/ (old) and mission-galleries/<id>/ (new)
Fleet	236 entity dirs	100% (236 / 236)	fleet-galleries/<id>/01.jpg for both hero + first gallery slot
Moon sites	16	100%	moon-sites/<id>/01.jpg
Mars sites	25	100%	mars-sites/<id>/01.jpg
Earth objects	13	100%	earth-objects/<id>/01.jpg
Planets	8	100%	planets/<id>/01.jpg
Satellites	16	100%	satellites/<id>/01.jpg
Small bodies	8	100%	small-bodies/<id>/01.jpg
ISS modules	24	100%	iss-modules/<id>/01.jpg
Tiangong modules	6	100%	tiangong-modules/<id>/01.jpg
Sun	1 (single entity, 20 gallery files flat)	n/a	sun/<nn>.jpg

Missions is the only surface with a hero gap, and the gap is structural — a 24-entry cohort with three sequential follow-up steps all skipped.

The 24-mission hero gap — root cause

Slice A/B/C (commits 62c35a0da, 45f668cd4, f4ee2e15d on main — "feat(missions + fleet)") added 24 missions to the catalog via the new static/images/mission-galleries/<id>/ directory pattern. Three follow-up steps that the older static/images/missions/<id>/ flow includes were skipped:

1. Hero staging: the cover image at static/images/missions/<id>.jpg was never copied from the gallery. Card UI references that path; the onerror="cover-missing" handler hides the figure gracefully but the cards lose their visual identity.
2. Provenance registration: none of the gallery files in mission-galleries/ were added to static/data/image-provenance.json. This shows up as 491 orphan files on disk vs 0 broken-reference entries in the manifest (the manifest is internally consistent but silent on a whole subtree of the corpus).
3. Vision scoring: because the gallery isn't in provenance, scripts/score-images.ts never sees these files. They have no entry in static/data/image-vision.json and no pre-cropped variants (.1x1.jpg / .4x3.jpg / .16x9.jpg).

The 24 affected mission ids

apollo-1     apollo-soyuz    inspiration4   mariner10
otv-1        otv-2           otv-3          otv-4
otv-5        otv-6           otv-7          polaris-dawn
shenzhou-1   skylab-2        skylab-3       skylab-4
voskhod-1    voskhod-2       vostok-1       vostok-2
vostok-3     vostok-4        vostok-5       vostok-6

Provenance ↔ disk consistency

• static/data/image-provenance.json entries: 4241
• Broken references (path in JSON but file missing on disk): 0
• Disk source files (non-variant): 2300
• Disk source files registered in provenance: 1809
• Disk orphans (file exists but not in provenance): 491 — 100% in mission-galleries/, matching the Slice A/B/C cohort above.

The manifest itself is clean. The data hygiene gap is absence, not wrongness — 491 files live outside the only system that scores them.

Pattern inconsistencies worth flagging

• Hero-path convention: missions use <surface>/<id>.jpg (a top- level file). Every other surface uses <surface>/<id>/01.jpg (a slot inside the gallery dir). The mission outlier is what made the Slice A/B/C skip possible — the staging step has no equivalent on the other surfaces because the first gallery slot is the hero.
• Variant filename convention: old pattern uses 01.16x9.jpg (dot separator). Slice A/B/C mission-galleries/ uses 01-16x9.jpg (dash separator). Pipeline tooling that grep-matches one form silently skips the other.

Both inconsistencies are fixable in the rework (collapse missions onto the universal <surface>/<id>/01.jpg pattern, standardise on dot or dash) but are independent of the hero-quality problem and shouldn't be mixed into the same change without a separate review.

What Phase 1 needs to handle

Per the rubric locked in chat (see §"HERO CRITERIA"), the audit script needs to evaluate every entity × surface for:

1. Hero existence — does the file at the conventional hero path exist? (Catches the 24-mission cohort + any future skipped staging.)
2. Provenance coverage — is the hero registered? (Catches the orphan-corpus class.)
3. Hero quality — pulls existing score from image-vision.json, applies the eight-point hero rubric (subject / composition / context / render policy / no-deny-list / recognition / prefer-alternate / cross-surface), emits per-surface audit markdown.
4. Cross-surface consistency — flags assets that appear under both a mission and a fleet entry (BepiColombo, Juno, SLIM noted in chat) so the same canonical hero resolves for both.

Output: one docs/provenance/<surface>-hero-audit.md per surface covering auto-swap proposals, needs-curation queue, and accepted overrides — same format as the existing fleet-hero-audit.md but emitted from one shared script driven by a per-surface config.

Budget plan

• Phase 1 (audit infrastructure): $0 — no API calls, uses existing manifest data + algorithmic post-filter.
• Phase 2 (operator review): $0 — Marko reads the per-surface markdowns, marks them up.
• Phase 3 (targeted rescore with hero rubric): $1-4 worst case — bump SCORING_PROMPT_VERSION to v1.1, extend the prompt to return a hero_score field alongside score, re-run only on Phase 2's flagged candidates (estimated 50-150 images).
• Phase 4 (sourcing the missing): $0 — agency-first fetcher pattern (Wikimedia / NASA Image Library / agency archives), applied to the Phase 1 needs-curation queue + the 24-mission backlog from this report.
• Phase 5 (frontend wiring): $0 — replace hardcoded image-path references with pickHero(surface, id) that reads the audit- approved manifest + an optional <surface>-hero-overrides.json.

Total expected: < $5 USD. Hard ceiling per Marko: $20 USD.

---

This inventory file is the artefact for Phase 0. Phase 1 (the generalised audit script) will be drafted next; outline + sample output will land here before any code is committed.