feat(dream): flag memories derived from a dead source

[skakri]

Fixes #1444.

A memory derived from another stays live after its source is withdrawn, and nothing says so. dream now reports it as a deterministic dependent_of_dead_source finding, ranked with stale_reference.

Behavior

• Scope. Every live memory in the active repo that reaches a dead memory along derived_from edges is reported, directly or through other derived memories.
• Dead means obsolete, rejected, or anchored only to code that is gone (every binding anchor_status = 'gone'). Unanchored Task/Concept nodes are never dead by bindings.
• Suspect, never retired. Nothing is changed automatically, in line with truth-maintenance semantics. The finding names the dead source, the first memory on the chain when the link is indirect, and a live memory that supersedes the source when one exists.
• Self-clearing. The kind is evaluated on every run, so the finding resolves once the source stands again, the edge is removed, or the dependent is retired.
• Stable claims. The evidence names memories by id only, so renaming one does not re-open a dismissed finding. A new successor does, since that is new information.

Deliberate choices

• Edges are read by target_anchor, not target_node_id. target_anchor is the target's globally unique node id on every path (local add, sync drain, consolidation). The sync drain leaves target_node_id unset for this device to resolve, so filtering on it would hide every synced edge.
• A target missing from this store is unknown, never dead. Local memories are retired, not deleted. The rows that vanish in production are the sync drain's: a quarantined peer update, an edge that arrived before its target, a target consolidation left behind. Those are memories still alive elsewhere, and a reviewer must not be told to act on them.
• A rewrite is not a dependent. A memory recorded as both derived_from and supersedes its source is its replacement, not a dependent left standing on it.
• Chains stop at the repo boundary. The walk follows edges owned by the active repo. A target in another repo is judged by its own status, but its own edges are not followed.
• "All bindings gone" reads the stored anchor_status, the same checkout-scoped verdict stale_reference and memory_doctor use.

Tests

schema_bootstrap_tests/repo_memory/dead_source_findings.rs covers:

• a direct dependent, and transitive propagation that names the root;
• the successor pointer;
• restoring the source clears the finding;
• a source anchored only to gone code;
• a derived_from cycle terminates and flags nothing;
• a synced edge with no resolved target is still followed, to its real source;
• a missing target flags nothing;
• renaming the source keeps a dismissal;
• a rewrite that supersedes its source is not flagged.

Each test fails under a mutation of the code it guards. Cross-repo chains are not covered by a test yet.

The dream-review skill (both copies), docs/mcp-tools.md and the MCP dream description describe the new kind.

[github-actions]

SCIP oracle — resolution report

Heuristic→compiler edge resolution per corpus. Δ compares resolved-after to the main baseline (only when the corpus profile + tool version match).

corpus	tool	edges	resolved (heuristic → compiler)	precision	recall	monikers	Δ vs main
c-libuv	scip-clang	12691	33.5% → 45.0%	96.0%	55.3%	661	+0.0pp
cpp-yaml	scip-clang	7686	42.0% → 67.8%	82.0%	41.1%	1049	+0.0pp
py-django	scip-python	12882	30.3% → 73.8%	93.6%	83.4%	3701	+0.0pp
py-rich	scip-python	8383	27.8% → 92.4%	94.6%	81.3%	1196	+0.0pp
rust-time	rust-analyzer	9907	37.8% → 59.1%	81.2%	83.6%	574	+0.0pp
ts-rxjs	scip-typescript	6533	43.7% → 94.9%	86.2%	76.8%	631	+0.0pp

_{resolved = Exact/Syntactic + compiler upgrades + resolved-external, over edge candidates with a callee range. precision/recall are the oracle eval metrics.}

[github-actions]

Bencher Report

Project	rag-rat
Branch	feat/dream-dead-source
Testbed	hetzner-bigmem

⚠️ WARNING: No Threshold found!

Without a Threshold, no Alerts will ever be generated.

• L1 Hits (hits)
• Total read+write (reads/writes)
• LL Hits (hits)
• Instructions (instructions)
• Estimated Cycles (cycles)
• RAM Hits (hits)

Click here to create a new Threshold
For more information, see the Threshold documentation.
To only post results if a Threshold exists, set the --ci-only-thresholds flag.

Click to view all benchmark results

Benchmark	Estimated Cycles	cycles x 1e6	Instructions	instructions x 1e6	L1 Hits	hits x 1e6	LL Hits	hits x 1e6	RAM Hits	hits x 1e3	Total read+write	reads/writes x 1e6
rag_pipeline::pipeline::index cargo_resolver:resolver_config()	📈 view plot ⚠️ NO THRESHOLD	1,900.50 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,189.06 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,662.94 x 1e6	📈 view plot ⚠️ NO THRESHOLD	38.33 x 1e6	📈 view plot ⚠️ NO THRESHOLD	1,311.60 x 1e3	📈 view plot ⚠️ NO THRESHOLD	1,702.58 x 1e6
rag_pipeline::pipeline::query_cold cargo_resolver:resolver_built_config()	📈 view plot ⚠️ NO THRESHOLD	307.07 x 1e6	📈 view plot ⚠️ NO THRESHOLD	204.01 x 1e6	📈 view plot ⚠️ NO THRESHOLD	290.51 x 1e6	📈 view plot ⚠️ NO THRESHOLD	3.06 x 1e6	📈 view plot ⚠️ NO THRESHOLD	35.66 x 1e3	📈 view plot ⚠️ NO THRESHOLD	293.61 x 1e6
rag_pipeline::pipeline::query_warm cargo_resolver:resolver_index()	📈 view plot ⚠️ NO THRESHOLD	278.77 x 1e6	📈 view plot ⚠️ NO THRESHOLD	184.84 x 1e6	📈 view plot ⚠️ NO THRESHOLD	263.81 x 1e6	📈 view plot ⚠️ NO THRESHOLD	2.76 x 1e6	📈 view plot ⚠️ NO THRESHOLD	33.77 x 1e3	📈 view plot ⚠️ NO THRESHOLD	266.60 x 1e6

🐰 View full continuous benchmarking report in Bencher

[github-actions]

Bencher Report

Project	rag-rat
Branch	feat/dream-dead-source
Testbed	hetzner-bigmem

Click to view all benchmark results

Benchmark	Latency	Benchmark Result seconds (s) (Result Δ%)	Upper Boundary seconds (s) (Limit %)
index_time/full_rebuild_cargo	📈 view plot 🚷 view threshold	5.68 s (-7.41%) Baseline: 6.14 s	7.17 s (79.22%)

🐰 View full continuous benchmarking report in Bencher

[codecov]

Codecov Report

❌ Patch coverage is 98.07692% with 5 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...ootstrap_tests/repo_memory/dead_source_findings.rs	97.84%	3 Missing ⚠️
crates/rag-rat-dream/src/dead_sources.rs	98.19%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!