diff --git a/docs/PRD.md b/docs/PRD.md
index 571c329..6e16dcb 100644
--- a/docs/PRD.md
+++ b/docs/PRD.md
@@ -39,7 +39,7 @@ is an explicit re-platforming, not a refactor.
 | HTTP gateway | Axum + Tokio | `net/http` + `chi` (or `gin`) | High — Go's bread and butter |
 | gRPC | tonic | `google.golang.org/grpc` | High — Go is the reference impl |
 | Object store | Apache Arrow `object_store` | `aws-sdk-go-v2/service/s3` + thin wrapper | High |
-| Parquet I/O | parquet-rs (arrow-rs) | `apache/arrow-go/v15/parquet` | Medium — arrow-go lags arrow-rs but covers our needs |
+| Parquet I/O | parquet-rs (arrow-rs) | `apache/arrow-go/v18/parquet` | Medium — arrow-go lags arrow-rs but v18 covers our needs |
 | Query engine | DataFusion | **Hard problem** (see §Hard problems) | Low — no like-for-like Go equivalent |
 | Vector index (HNSW) | `hora` / hand-rolled | `coder/hnsw` or `Bithack/go-hnsw` (in-process) | High — HNSW is a self-contained algorithm |
 | Vector backend (Lance) | `lance` (Rust) | **Hard problem** — likely dropped, Parquet-only | Medium |
@@ -76,7 +76,7 @@ binaries built from one workspace.
 | gRPC | `google.golang.org/grpc` | Reference implementation |
 | Protobuf | `protoc-gen-go` + `buf` | Standard tooling |
 | Object store | `aws-sdk-go-v2` | Mature, covers S3 + MinIO + RustFS |
-| Parquet | `apache/arrow-go/v15` | Columnar I/O + Arrow interop |
+| Parquet | `apache/arrow-go/v18` | Columnar I/O + Arrow interop (v18.5.2 — March 2026) |
 | SQL engine | **Open** — see §Hard problems §1 | Biggest open decision |
 | Vector index | `coder/hnsw` | Pure-Go HNSW, in-process, no external service |
 | TOML config | `pelletier/go-toml/v2` | Standard |
diff --git a/docs/RESEARCH_LOG_2026-04-28.md b/docs/RESEARCH_LOG_2026-04-28.md
new file mode 100644
index 0000000..0fe1c42
--- /dev/null
+++ b/docs/RESEARCH_LOG_2026-04-28.md
@@ -0,0 +1,133 @@
+# Research Log — 2026-04-28
+
+Survey of upstream Go ecosystem state before Phase G0 begins. Goal:
+verify the SPEC's library choices reflect the *currently maintained*
+upstream and surface anything that has shifted since the SPEC was
+drafted earlier today.
+
+## Sources used
+
+- Context7 — `resolve-library-id` for `go-duckdb` and `arrow-go`
+- WebSearch — six targeted queries on Go HNSW, MCP SDK, DuckDB
+  bindings, Anthropic SDK, HTTP frameworks, parquet maturity
+- WebFetch — direct GitHub READMEs for `coder/hnsw`,
+  `modelcontextprotocol/go-sdk`, `duckdb/duckdb-go`
+
+## Findings — three SPEC-changing pivots
+
+### 1. DuckDB Go binding ownership transferred
+
+The original SPEC named `marcboeker/go-duckdb` as the query-engine
+library. The package has been formally **deprecated** in favor of the
+official maintainer transfer.
+
+| | Before (SPEC v1) | After |
+|---|---|---|
+| Repo | `github.com/marcboeker/go-duckdb` | `github.com/duckdb/duckdb-go/v2` |
+| Maintainer | Marc Boeker (community) | DuckDB team + Marc Boeker (joint, official) |
+| Latest | (varies) | v2.10502.0 (April 2026) |
+| DuckDB engine | varies | v1.5.2 |
+| Migration | n/a | `gofmt -w -r '"github.com/marcboeker/go-duckdb/v2" -> "github.com/duckdb/duckdb-go/v2"' .` |
+
+**Why this matters:** the upstream migration happened at v2.5.0; using
+the old import means we'd be on a deprecated branch from day one.
+SPEC §3.1 now names the official path.
+
+### 2. Official Go MCP SDK exists (Google collaboration)
+
+The original SPEC named `mark3labs/mcp-go` (community implementation).
+An official SDK now exists.
+
+| | Before (SPEC v1) | After |
+|---|---|---|
+| Repo | `mark3labs/mcp-go` | `github.com/modelcontextprotocol/go-sdk` |
+| Maintainer | mark3labs (community) | MCP org + Google (official) |
+| Stars | (smaller) | 4.4k |
+| Version | n/a | v1.5.0 stable |
+| Spec compat | various | targets MCP 2025-11-25, backward-compat to 2024-11-05 |
+| OSSF Scorecard | n/a | yes |
+
+**Why this matters:** the MCP server (`cmd/mcp`) is one of the most
+visible Go binaries we'll ship — it's what AI agents talk to. Using
+the official SDK aligns with the MCP spec's evolution and gets us the
+Google-tested code path.
+
+### 3. arrow-go is on v18, not v15
+
+The original SPEC referenced `arrow-go/v15`. Apache Arrow Go has
+shipped through **v18.5.2** (March 2026).
+
+| | Before (SPEC v1) | After |
+|---|---|---|
+| Module path | `apache/arrow-go/v15` | `apache/arrow-go/v18` |
+| Latest | n/a | v18.5.2 |
+| Recent fixes | n/a | parquet decryption, large string handling, complex type read |
+
+**Why this matters:** v18 has parquet encryption fixes that are
+relevant for our PII-masked safe views (per Rust ADR-017's federation
+and the production cutover work). Skipping three major versions is
+unnecessary risk.
+
+## Findings — validations (SPEC unchanged)
+
+### `coder/hnsw` — keep
+
+220 stars, 45 commits, active CI, recent PRs/issues. Documented
+import speed of 796.85 MB/s. No deprecation signals. In-memory
+alternative to Pinecone/Weaviate, fits the no-external-vector-DB
+constraint.
+
+### `chi` for HTTP routing — keep
+
+Confirmed as the "clean architecture, stdlib `net/http`, zero deps"
+pick. Fiber is faster (36k req/s vs ~34k for chi/gin/echo) but uses
+fasthttp, which is off the standard library path — wrong fit for our
+"boring is good" Go ethos. Fiber stays as a documented alternative if
+hot-path performance ever proves chi insufficient.
+
+### `marcboeker/go-duckdb` → `duckdb/duckdb-go/v2`
+
+Already covered above as pivot #1.
+
+## New things worth noting (not SPEC-changing yet)
+
+### Anthropic Go SDK is official
+
+`github.com/anthropics/anthropic-sdk-go` is the official Anthropic Go
+client library. We currently route Claude calls through OpenCode/Zen,
+so this isn't on the Phase G0 critical path. **Worth knowing for**:
+direct Claude API calls in `aibridge` if we ever want to bypass
+opencode (e.g., for the overseer correction loop in dev mode without
+a Zen subscription).
+
+Added as a noted option in the SPEC `aibridge` row.
+
+### DuckDB-Wasm exists
+
+`github.com/duckdb/duckdb-wasm` brings DuckDB to the browser via
+WebAssembly with Arrow / Parquet / CSV / JSON support. **Not in scope**
+for Phase G0 but a future option if the UI ever needs client-side
+analytical queries against fetched parquet (offline analytics over a
+permit cache, etc.).
+
+### IVF as an HNSW alternative
+
+If during Phase G3 the HNSW recall validation gate (G3.2.C) shows
+problems, IVF (Inverted File Index) is the next-best alternative:
+faster index builds, lower memory, better filtered-search performance
+than HNSW. No first-class Go IVF library was found — would need to
+wrap FAISS via cgo or hand-roll. Documented as a fallback only.
+
+## What I checked but found nothing actionable
+
+- HTMX-specific Go libraries (`htmx-go`, `gotempl`, etc.) — none has
+  emerged as a clear standard. Sticking with `html/template` + raw
+  HTMX as the SPEC plans.
+- Langfuse Go client — OSS support varies. SPEC's "hand-roll if
+  needed" stays.
+
+## Outcome
+
+SPEC §1 + §2 + §3.1 updated to reflect the three pivots. PRD's locked
+stack table updated to match. No phase changes, no acceptance gate
+changes, no new hard problems.
diff --git a/docs/SPEC.md b/docs/SPEC.md
index ac69c3d..1d56020 100644
--- a/docs/SPEC.md
+++ b/docs/SPEC.md
@@ -23,15 +23,15 @@ Effort scale (one engineer-week = ~40h focused work):
 | Crate | Rust deps that mattered | Go target | Library | Effort | Risk |
 |---|---|---|---|---|---|
 | `gateway` | axum, tokio, tonic, tower | `cmd/gateway` | `chi` + stdlib `net/http` + `google.golang.org/grpc` | **L** | low — Go's strongest domain |
-| `catalogd` | parquet-rs, arrow, sqlite | `cmd/catalogd` | `arrow-go/v15`, `mattn/go-sqlite3` | **L** | low |
+| `catalogd` | parquet-rs, arrow, sqlite | `cmd/catalogd` | `apache/arrow-go/v18`, `mattn/go-sqlite3` | **L** | low |
 | `storaged` | object_store, aws-sdk | `cmd/storaged` | `aws-sdk-go-v2`, `minio-go` for MinIO-specific paths | **M** | low |
-| `queryd` | datafusion, arrow | `cmd/queryd` | `marcboeker/go-duckdb` (cgo) | **HARD** | high — see §3 |
+| `queryd` | datafusion, arrow | `cmd/queryd` | **`duckdb/duckdb-go/v2`** (cgo, official) | **HARD** | high — see §3 |
 | `ingestd` | csv, json, lopdf, postgres | `cmd/ingestd` | stdlib `encoding/csv`, `encoding/json`, `pdfcpu/pdfcpu`, `jackc/pgx/v5` | **L** | low |
-| `vectord` | hora, arrow, hnsw | `cmd/vectord` | `coder/hnsw`, `arrow-go/v15` | **L** | medium — re-validate HNSW recall |
+| `vectord` | hora, arrow, hnsw | `cmd/vectord` | `coder/hnsw`, `apache/arrow-go/v18` | **L** | medium — re-validate HNSW recall |
 | `vectord-lance` | lance | **DROPPED** | n/a | n/a | n/a — Parquet+HNSW only |
-| `journald` | parquet, arrow | `cmd/journald` | `arrow-go/v15` | **M** | low |
-| `aibridge` | reqwest | library | `net/http` + connection pool | **S** | low |
-| `validator` | parquet, custom | library | `arrow-go/v15` parquet reader | **M** | low — port the 24 unit tests as gates |
+| `journald` | parquet, arrow | `cmd/journald` | `apache/arrow-go/v18` | **M** | low |
+| `aibridge` | reqwest | library | `net/http` + connection pool · `anthropics/anthropic-sdk-go` available for direct Claude calls (currently routed via opencode) | **S** | low |
+| `validator` | parquet, custom | library | `apache/arrow-go/v18` parquet reader | **M** | low — port the 24 unit tests as gates |
 | `truth` | tomli, custom DSL | library | `pelletier/go-toml/v2` | **M** | low |
 | `proto` | tonic-build | `proto/` + `protoc-gen-go` | `buf` + `protoc-gen-go-grpc` | **S** | low |
 | `shared` | serde, anyhow | library | stdlib `encoding/json`, `errors` | **S** | low |
@@ -47,7 +47,7 @@ one engineer; 6–8 weeks for two).
 
 | TS surface | Current location | Go target | Library | Effort | Risk |
 |---|---|---|---|---|---|
-| `mcp-server/index.ts` | Bun, :3700 | `cmd/mcp` | `mark3labs/mcp-go` (Go MCP SDK) | **L** | medium — MCP semantics |
+| `mcp-server/index.ts` | Bun, :3700 | `cmd/mcp` | **`modelcontextprotocol/go-sdk`** (official Go SDK, v1.5.0, Google-collab) | **L** | medium — MCP semantics |
 | `mcp-server/observer.ts` | Bun, :3800 | `cmd/observer` | stdlib `net/http`, `slog` | **M** | low |
 | `mcp-server/tracing.ts` | Bun, Langfuse client | library | `go.opentelemetry.io/otel` + Langfuse Go client (or hand-roll) | **M** | low — Langfuse Go OSS support varies |
 | `auditor/*.ts` | TS, runs as systemd | `cmd/auditor` | stdlib + `gitea API client` | **L** | medium — auditor cross-lineage logic is intricate |
@@ -64,7 +64,13 @@ one engineer; 6–8 weeks for two).
 
 ### §3.1 — Query engine (DuckDB via cgo)
 
-**Library:** `marcboeker/go-duckdb` — Go bindings via cgo.
+**Library:** `github.com/duckdb/duckdb-go/v2` — official Go bindings via
+cgo. (Replaces the legacy `marcboeker/go-duckdb`, which was deprecated
+when the DuckDB team and Marc Boeker jointly relocated maintenance to
+the DuckDB org at v2.5.0. Migration is a one-line `gofmt -r` rewrite of
+import paths.) Current version v2.10502.0 (April 2026), DuckDB v1.5.2
+compat. Statically links default extensions: ICU, JSON, Parquet,
+Autocomplete.
 
 **API shape** (replaces the DataFusion `SessionContext` pattern):
 ```go