chore(card-data): refresh MTGJSON token & subtype catalogs - #6849

Investigator 1/3: Investigator 1 of 3 reviewed 34 assigned PRs (all in phase-rs/phase, an MTG rules-engine repo) by a MEMBER author. Four PRs were hydrated in full patch mode (6835, 6832, 6829, 6821, 6807) and reviewed adversarialy at the hunk level; five were preview-mode with file lists only but truncated hunks (6804, 6800, 6785, 6636, 6778); 24 were metadata-only. The full patches show no supply-chain or backdoor behavior: no network calls, no dependency/package/script changes, no CI/release tampering, no obfuscation, no permission broadening, no credential/secret access, no installation-lifecycle abuse. PR 6835 (current PR) is a pure JSON metagame-feed data refresh (date bumps in client/public/feeds/*.json). PR 6832 is a Rust parser refactor (oracle.rs) reifying CopyChosenHost relation synthesis with fail-closed guards and standard test mocks (Math.random spy with mockRestore). PR 6829 is a security-hardening refactor making AI actions use engine-validated proposals (preventing unauthorized/outdated AI actions). PR 6821 is a one-line vintage date stamp. PR 6807 converts temporal delayed-trigger parsing to native IR with snapshot fixtures. The preview-mode PRs are consistent with their described parser/AI/client/engine work (file paths match titles) but hunks were truncated, so they are marked unknown rather than clean. The 24 metadata-only PRs cannot be judged and are unknown. No credible malicious patch evidence exists in the directly-reviewed hunks. Confidence is medium rather than high because 29 of 34 PRs in this shard were not fully hunk-reviewable (truncated preview or metadata-only); the overall safe verdict rests on the five full-patch PRs that were concretely reviewed plus the absence of any suspicious signal across the entire shard. Activity is single-repo by a MEMBER (not a burst of unsolicited cross-repo PRs), a positive consistency signal. Investigator 2/3: Investigator 2 of 3 reviewed the assigned shard of matthewevans's PR history in phase-rs/phase (an MTG rules engine). The contributor is a MEMBER of the repo and all activity is concentrated in this single project (no cross-repo burst; unsolicitedPrRatio 0). The fully-hydrated patches (PRs 6819, 6826, 6811, 6801, 6809) were inspected adversarially at the file/hunk level: they are clean, rules-citation-documented Rust engine/parser/AI/client work. No new network paths, no credential/secret access, no obfuscated code, no package-script or dependency-chain abuse, no CI/release tampering, no permission broadening, and no telemetry disguised as fixes were found. PR 6801 adds new GameActions (ActivateManaSource, BackToManaPayment) and a ManaSourceSelectionUI modal — a legitimate sacrificial-mana-source payment feature with i18n and wire-protocol additions, no exfiltration surface. PR 6826's new targeted_exchange.rs is a bounded reducer-backed AI safety veto (rejects self-destructive exchanges) — benign, no external I/O. PR 6819 reifies exile-land-play authorization through zone moves; benign game-logic refactor with tests. PRs 6811/6809 are parser IR-native emission refactors with snapshot tests; benign. PR 6619 (CI: disable incremental builds for sccache) and PR 6791 (MTGJSON catalog refresh) are benign chores. The metadata-only/preview PRs (e.g., CR 733 resolved-command journaling series: attachments, transforms, stack pushes, continuous effects, delayed triggers) are thematically consistent with the hydrated determinism/replay work and showed no anomalous signals in titles/descriptions, though their patches were not hydrated so they are marked unknown rather than clean. Because 70/100 PRs are omitted and many are metadata-only, confidence is held at medium rather than high; however, the highest-risk hydrated patches clearly rule out malicious behavior and the contributor is a trusted member working in one repo. Investigator 3/3: Investigator 3 of 3 reviewed 33 assigned candidate PRs, all in phase-rs/phase (and its forks). The contributor is a MEMBER of this dominant repository and the activity is a coherent, long-running effort: a methodical oracle-parser IR refactoring series ("Plan 05b", PRs 6776/6742/6733/6724/6720/6715/6709/6703), engine/casting bug fixes, a scoped BO3 concession feature, and automated metagame-feed refreshes. Four PRs carried full patches and were reviewed adversarially at the hunk level. PR 6820 (crates/engine/tests/.../cr733_resolved_frame_transition.rs) adds only a regression test asserting a stale-owner prompt is rejected atomically — no execution, network, dependency, or lifecycle change. PR 6830 (crates/engine-wasm/src/lib.rs) changes AiProposalRegistry::insert from proposals.clear() to proposals.retain by semantic_owner, a bounded per-owner token fix with added tests; no new network path, secret access, or obfuscation. PR 6806 (crates/engine/src/parser/oracle.rs + snapshot tests) refactors loyalty-ability parsing to emit native AbilityIr via a clean parse_loyalty_ability_ir helper; byte-identical full-pool parity is claimed; purely parser logic, no I/O. PR 6817 (62 files, large) adds scoped BO3 concession controls; visible patches cover test infrastructure and terminal-result commitment logic (recipient-bound finalStateCommitment, host-session fencing) which is game-integrity plumbing, not data exfiltration, with no hidden fetches or credential access in reviewed hunks (server/persistence hunks were truncated). Five preview-mode PRs (6803 scryfall-fetch.sh, 6792, 6750, 6749, 6633) show consistent data-fetch/parser/engine-refactor content matching their descriptions, though truncated patches limit full attestation. 24 PRs are metadata-only and could not be patch-verified; they belong to the same coherent parser-refactor series with detailed, technical bodies. No credible malicious or backdoor patch evidence was found in the reviewed shard: no credential exfiltration, hidden network calls, installation-lifecycle abuse, suspicious dependency/script changes, obfuscated code, permission broadening, or CI/release tampering. Confidence is medium rather than high because 24 of 33 shard PRs are metadata-only and several preview/full patches were truncated, so patch-level safety rests primarily on the four fully-reviewable PRs plus the coherent metadata; the merged verdict across all three investigators should reconcile the omitted 70 PRs. Patch-level safety guard downgraded the result to caution: Agent returned safe without patch-level clean reasoning for #6810.