refactor!: delete legacy APIs and shrink storage footprint

.cargo/config.toml

@@ -70,8 +70,8 @@ rustflags = ["-C", "split-debuginfo=unpacked"]
 # results, 8 threads, slow-timeout termination), the root fixture feature and
 # `search-eval`, which is what compiles the evaluator acceptance suites (the
 # root crate's dependency on the evaluator is optional so the Linux transport
-# partition does not build the eval-only lexical projection; a `--workspace`
-# run resolves the evaluator anyway). `.github/workflows/ci.yml` runs this
+# partition does not link it; a `--workspace` run resolves the evaluator
+# anyway). `.github/workflows/ci.yml` runs this
 # exact alias, and its support binaries/archives are built with `--profile
 # perf` so they share artifacts with it. Extra flags append: `cargo test-ci
 # --locked -E 'binary(=x)'`.

.gitattributes

@@ -1,9 +1,10 @@
 # Force LF line endings for test fixtures. Tree-sitter grammars
 # expect Unix line endings and produce wrong parse trees with CRLF.
 tests/fixtures/** text eol=lf
-# The embedded SQL fixture's definitions feed the exact configuration schema
-# digest; checkout line-ending conversion must not change those definitions.
-crates/tracedecay-global-db/src/configuration/store/tests/fixtures/configuration-registry-revision3.sql text eol=lf
+# Embedded SQL is executed verbatim and its stored definitions feed exact
+# schema-shape digests; a CRLF checkout rewrites those definitions and makes
+# a released shape read as foreign.
+*.sql text eol=lf
 # The search-quality candidate workload pins each corpus document against the
 # blob in the pinned source tree, and the packaged evaluator assets are pulled
 # into the binary with include_bytes! under a pinned digest. One corpus

.github/workflows/ci.yml

@@ -461,9 +461,8 @@ jobs:
   # `tracedecay-search-eval`) with the root fixture feature. Journeys also
   # enable `tracedecay/search-eval`, the only link from the root crate to the
   # evaluator library. It stays off every other partition: an unconditional
-  # dependency unifies `tracedecay-query/search-eval` into every test
-  # target of the package, and the transport suites then compile the eval-only
-  # lexical projection. Every cargo invocation inside a job (the test build,
+  # dependency links the evaluator into every test target of the package.
+  # Every cargo invocation inside a job (the test build,
   # the executables the suites spawn) is a cache hit against that job's
   # resolution. `scripts/linux-test-partitions.py check` proves, from `cargo
   # metadata`, that every test target in the workspace is selected by exactly
@@ -859,7 +858,7 @@ jobs:
       # Acceptance tests execute the ordinary evaluator binaries, and
       # `tracedecay/search-eval` is what compiles the suites that check the
       # CLI receipt against the library. This lane builds `--workspace`, so
-      # the evaluator and `tracedecay-query/search-eval` already resolve here
+      # the evaluator already resolves here
       # and the feature adds no compilation; the Linux partitions split the
       # package selection, which is why it is per-partition there. Match the
       # archive's features and perf profile to reuse dependency artifacts;

Cargo.toml

@@ -85,6 +85,7 @@ rusqlite = { version = "0.40.1", default-features = false, features = ["backup",
 # resolve the same object database and status implementation, so a feature
 # added for one is never missing for the other.
 gix = { version = "=0.87.1", default-features = false, features = ["revision", "blob-diff", "parallel", "sha1", "sha256", "status"] }
+schemars = "1.2.1"
 
 [profile.bench]
 opt-level = 3

README.md

@@ -102,7 +102,6 @@ the installed cache is loaded.
 tracedecay daemon install-service   # install + start the daemon (required before init)
 tracedecay init [path]              # enroll a project and publish its first generation
 tracedecay sync [path]              # explicit administrative refresh
-tracedecay sync --force [path]      # explicit full generation refresh
 tracedecay status [path]            # graph stats, freshness, savings, cost
 tracedecay tool                     # list every MCP tool
 tracedecay tool search "<query>"    # CLI symbol search
@@ -124,7 +123,7 @@ list and `tracedecay tool <name> --help` for one tool's parameters.
 
 The MCP server exposes tools grouped around normal coding workflows:
 
-- Discovery: `tracedecay_context`, `tracedecay_search`, `tracedecay_outline`, `tracedecay_files`
+- Discovery: `tracedecay_context`, `tracedecay_search`, `tracedecay_source_outline`, `tracedecay_files`
 - Graph traversal: `tracedecay_callers`, `tracedecay_callees`, `tracedecay_impact`, `tracedecay_affected`
 - Code health: `tracedecay_complexity`, `tracedecay_dead_code`, `tracedecay_unmounted_files`, `tracedecay_coupling`, `tracedecay_test_risk`
 - Git workflow: `tracedecay_diff_context`, `tracedecay_pr_context`, `tracedecay_changelog`, `tracedecay_test_map`

SECURITY.md

@@ -31,7 +31,7 @@ tracedecay builds a **local** code graph stored in the active project store. Rep
 - Call relationships and dependency edges
 - FTS5 search index
 - Cross-session memory: durable facts, named entities, code-area notes, decisions, and feedback events in the holographic fact store. Those rows are local-only project data.
-- A response cache for `tracedecay_read` (`read_cache` table): the rendered output served to the agent, stored as a BLOB keyed by file path, mode, and arguments. For full/line-range reads this rendered output contains source text. Rows are freshness-gated by file mtime and swept after a period of inactivity.
+- A response cache for mode-aware source reads (`read_cache` table): the rendered output served to the agent, stored as a BLOB keyed by file path, mode, and arguments. For full/line-range reads this rendered output contains source text. Rows are freshness-gated by file mtime and swept after a period of inactivity.
 
 Aside from the `read_cache`, the graph itself does **not** persist raw source code. It stores structural metadata only. The active project store is local-only. There is no cloud sync, remote database, or server-side storage.
 
@@ -148,7 +148,7 @@ The Windows-elevation `unsafe` documented in earlier versions was removed alongs
 ## Best Practices
 
 - Add `.tracedecay/` to your `.gitignore` to avoid committing local store markers or repo-local databases.
-- If your project contains sensitive code, be aware that the database stores symbol names and signatures, and the `read_cache` table can hold rendered source text from `tracedecay_read` responses. Keeping repo-local store directories ignored and treating profile-sharded stores as private user data keeps both out of version control.
+- If your project contains sensitive code, be aware that the database stores symbol names and signatures, and the `read_cache` table can hold rendered source text from source-read responses. Keeping repo-local store directories ignored and treating profile-sharded stores as private user data keeps both out of version control.
 - Keep tracedecay updated (`tracedecay upgrade`) to receive security fixes.
 - Review the [CHANGELOG](CHANGELOG.md) before upgrading to understand what changed.
 

benchmark_data/run_benchmarks.py

@@ -342,14 +342,20 @@ def avg_ms(samples: list[int]) -> float | None:
             ]
             out["find_symbol_avg_ms"] = avg_ms(find_us)
 
-            body_us = [
-                d
-                for q in queries
-                if (
-                    d := handler_us(mcp.call_tool("tracedecay_body", {"symbol": q, "limit": 1}))
+            # Name-to-source is an exact lookup followed by `tracedecay_source_body`.
+            body_us: list[int] = []
+            for q in queries:
+                fresp = mcp.call_tool(
+                    "tracedecay_find_exact_symbol", {"name": q, "limit": 1, "format": "json"}
                 )
-                is not None
-            ]
+                try:
+                    matches = json.loads(fresp["result"]["content"][0]["text"])["matches"]
+                    node_id = matches[0]["id"]
+                except (KeyError, IndexError, TypeError, json.JSONDecodeError):
+                    continue
+                bus = handler_us(mcp.call_tool("tracedecay_source_body", {"node_id": node_id}))
+                if bus is not None:
+                    body_us.append((handler_us(fresp) or 0) + bus)
             out["get_function_source_avg_ms"] = avg_ms(body_us)
 
             impact_us: list[int] = []

benchmark_data/transport-boundary/README.md

@@ -1,26 +1,10 @@
-# Transport-boundary direct test fixture
+# Transport-boundary compile baseline
 
-> **Historical evidence only.** Preserve the real fixture and provenance in
-> this directory. Current requirements come only from the
-> `docs/plans/tracedecay-v2/` hierarchy; exact commands, test names/counts,
-> snapshots, receipts, attestations, PR packets, and gate fields below are not
-> rebuild instructions. Validate current transport behavior directly.
-
-This directory supplies a checked-in real fixture to the direct product test
-for the callable application transport boundary. It is not an acceptance packet,
-snapshot, gate manifest, or separate evidence authority. Product acceptance
-comes from direct tests and normal CI, with pending work reported as pending.
-
-`goldens/application-surface-parity.json` is consumed by
-`tests/api_application_parity.rs`. Together they require:
-
-- identical request/result contracts and binding identities for Git preview,
-  Git apply, feedback diagnostics, feedback get, feedback expand, and feedback
-  list across CLI, MCP, and HTTP dispatch;
-- all four feedback reads to reach a callable owner and conceal an unknown
-  request handle as `not_found_or_not_authorized`, never `unavailable`;
-- HTTP and SSE to project the same canonical feedback payload; and
-- concealed HTTP problems to omit binding identity.
+> **Historical evidence only.** Current requirements come only from the
+> `docs/plans/tracedecay-v2/` hierarchy; exact commands and fields below are
+> not rebuild instructions.
+> `crates/tracedecay/tests/product_surface_suite/api_application_parity.rs`
+> verifies transport parity directly.
 
 The compile workload remains executable, but its measurement is explicitly
 pending execution. Static validation does not run Cargo: