0.5.204: the minor triple — edge v21.1.0 / persist v42.1.0; dimension reads are index-served (#571)

[emooreatx]

Adopts the minor triple from CIRISServer#571: edge v21.1.0 / persist v42.1.0 / verify v15.0.0 (unchanged). Pins ×4 persist, ×2 edge, plus the lens-core member crate; TARGET_* in the substrate gate; README pin prose; evidence rows re-stamped to v0.5.204; one copy of each crate (cargo tree -i). No code moved for the substrate — persist's only public API change is an added helper. Closes #571.

What reaches this node

• CIRISPersist#817 — dimension reads are index-served. Both dimension axes of list_attestations compiled to a per-row json_extract; V137 indexes the generated dimension column and the prefix filter compiles to a range on it. This is the substrate fix for graph_config::get_* is O(rows this node authored) per key — a keyed read that costs a full scan, and gets slower the more the node emits #557 (our report): graph_config's config: prefix scan is no longer O(rows this node authored). The snapshot cache stays — fifty getters costing one read is a property worth keeping whatever the read costs. The comment in graph_config::live_config_rows says so now.
• CIRISPersist#818 — a prefix filter compares bytes on every backend; sqlite's LIKE was case-insensitive. Every dimension this node reads by prefix (config:, consent:) is lowercase, so no row set changes here; the memory/sqlite/postgres parity is what improves.
• CIRISEdge#579 — the cohab lane injects verify beside persist. Our conformance workflow injects no pins (checked), so nothing to change on our side; noted for the recipe.

Also on main since v0.5.203, shipped by this version

• The stop is legible from the node's own log: serve marker, numbered drain, named origin (#568) #569 the stop is legible: serve marker, per-listener drain count, named stop origin, every teardown step timed/named/bounded; SIGTERM → exit in ~10 s (was never; CIRISEdge#578).
• The one-scan config test tolerates a snapshot TTL that elapsed during its own loop #570 per-engine config snapshot cache (the mac/win one-scan flake) and the ladder path filter covering the config plane.
• The Windows installer bundles the desktop JAR from the client package (#562) #566 the Windows installer builds from the client package; the job is gating.
• The test log capture is a process-global layer routing to a task-local sink (#542) #565 the test log capture is a process-global routing layer (Flake (macOS once, ubuntu once): retention_loop's steady-state pass captured no INFO line on main (a_pass_with_nothing_to_evict_is_audible_and_not_an_alarm) — the scoped capture, not the loop #542).

Verification

Filled from the log in the PR conversation: check / clippy -D warnings / fmt; the 22 gates separately; lib; test-anchor suites (anchor block verifies under the new pair); admission and config suites; lens-core; the sequential sweep of every integration binary; both harness ladders on the release wheel. Full OS matrix on the PR (ci:full).

Tagging v0.5.204 waits for Eric's go.

🤖 Generated with Claude Code

https://claude.ai/code/session_0132ELwj5JU6t4jEKpwH9bJ7

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[emooreatx]

Local verification on the release wheel (ciris_server-0.5.204-cp310-abi3-linux_x86_64.whl, 34 MB, --features test-anchor,python):

• cargo check --all-targets, clippy -D warnings, fmt clean.
• All 22 src-scraping gates green separately, including substrate_contract_gate (README ↔ Cargo.toml pins) and release_gates 42 with the moved TARGET_*.
• lib 494; lens-core all targets.
• test-anchor suites: anchor_block_verifies 1/1 — the vendored anchor block still verifies under persist v42.1.0 / verify v15.0.0 — trace_round_e2e 4/4, trust_root_qa 7/7.
• Sequential sweep of all 119 integration binaries: 0 red. Run because #818 changes what a dimension-prefix read matches (byte-exact instead of sqlite's case-folded LIKE); no row set in this repo's suites changed, every dimension we read by prefix being lowercase.
• chat ladder: SUCCESS (the standing RED-EXPECTED at one_sided, CIRISEdge#373). traceflow ladder: SUCCESS — full chain sealed → consented → shipped → arrived → materialized → summarized → scored.

cargo tree -i: one copy each of persist 42.1.0, edge 21.1.0, verify-core 15.0.0.