# Metadata search Add Comics keeps provider groups and canonical `comicvine:`, `metron:` and `gcd:` identities. Descriptions are plain text at the search presentation boundary; stored library descriptions are not rewritten. Related publications are separate volumes, never automatic merges, provider switches or shared issue lists. ComicVine description links explicitly introduced by “preceded by”, “continued from”, “continued in” or “continues as” become predecessor/continuation evidence. Only recognized ComicVine hosts and exact `4050-` volume paths are admitted. The volume-number inference and relationship extraction share the same description module. Unlinked prose can retain its legacy numbering inference but cannot create a relationship. No title/numbering similarity establishes identity or continuation. Search expands at most two targets per direct result and four targets overall, at depth one, within the same provider. Exact IDs already returned are not fetched again; failures retain primary results. Title punctuation/whitespace normalization is used locally for ranking, without removing words or rewriting provider queries. The ordering is exact title/alias, other direct token matches, explicit related results, then direct results with no query tokens. Token overlap and optional year break ties; provider order is retained for equivalent candidates. Publisher, volume number and issue count remain visible facts, not inferred query constraints. `metadata-search/v2` adds `relations`, `search_origin`, `relation_reason` and `rank_components`. Add Comics and Library Import request `presentation=v2&artwork=true`; Library Import also requests `expand_relations=false`. Its compact result covers use the same bounded lazy artwork ticket. Clicking an available cover opens a view-only lightbox; the original provider URL is used only when already supplied and validated. A resolved artwork-ticket image is a compact preview, not a full-resolution cover. Selecting any candidate always uses its exact provider-qualified ID. ## Artwork and request bounds Primary search makes **zero artwork enrichment calls**. Visible cards use the authenticated artwork endpoint, with server-issued ten-minute search tickets: four identities per batch, twelve per search, one active batch per process. Each candidate needs at most one metadata lookup and one image GET; no full-volume fetch, issue enumeration, pagination or automatic retries are used for artwork. GCD's normal persisted request ledger charges its issue lookup; Metron rate-limit state is honored, with no artwork sleep while quota is unavailable. The in-memory LRU cache holds 128 provider-qualified identity/hint entries, at most 8 MiB of JPEG thumbnails: successful entries expire after 24 hours and unavailable entries after ten minutes. At most 32 search tickets are retained. An unavailable first-issue cover is a normal placeholder; later issues are not crawled for another image. No relation persistence or schema migration is needed: evidence is re-derived from the search description without an extra source call. Images use allowlisted HTTPS hosts/paths, public DNS validation and pinned TLS, no redirects, cookies, proxies or credentials. Input is limited to 2 MiB and 16 million pixels, decoded as JPEG/PNG/WebP and re-encoded to a metadata-free JPEG of at most 64 KiB. The browser receives only the thumbnail, never provider tokens or provider image request URLs. Slow/failed artwork cannot fail primary search; new searches discard stale artwork responses. ## Provider contracts - **ComicVine:** HTML descriptions and explicit linked volume evidence already consumed by Pullarr's numbering inference. Structured continuation fields are not assumed. The [API documentation](https://comicvine.gamespot.com/api/documentation) was inaccessible during the release audit; deterministic existing/raw-shaped fixtures verify the integration. Live authenticated acceptance is optional. - **GCD:** [official API documentation](https://github.com/GrandComicsDatabase/gcd-django/wiki/API) and [serializers](https://github.com/GrandComicsDatabase/gcd-django/blob/0b1a2c948c595e31cdb4864c3c78fc25c0cd4797/apps/api/serializers.py) expose `active_issues` in series results and `cover` in issue details. One exact member lookup supplies art; both issue ID and parent series are verified. The series serializer exposes no predecessor/successor relation. Public issue 100000 confirmed the `files1.comics.org` cover path during the audit. - **Metron:** [series serializers](https://github.com/Metron-Project/metron/blob/8b2cc756aee824e03f0a7ff909032c689bd3127c/api/v1_0/serializers/series.py) have no series image. The [issue-list serializer](https://github.com/Metron-Project/metron/blob/8b2cc756aee824e03f0a7ff909032c689bd3127c/api/v1_0/serializers/issue.py) includes `image` and parent series. Artwork reads only the first issue-list page. Detail `associated` series are shown as undirected `related_series` evidence (at most two); they are neither inferred continuations nor automatic expansion targets. Normal series-list searches omit this detail-only field.