feat(pinax): implement Phase 01 pager, buffer pool, and B+tree

.kanon-lint-ignore

@@ -30,10 +30,38 @@ STORAGE/no-migration-checksum:crates/lexis/src/ast.rs
 STORAGE/no-migration-checksum:crates/lexis/src/lib.rs
 
 # =============================================================================
-# crates/{hypomnema,phylaxis,pinax}/ — genuinely empty scaffolds
+# crates/pinax/ — Phase 01 landed; false positives, not real defects
 # =============================================================================
 
-# WHY: these three crates are empty by design at this phase — lib.rs is
+# WHY(forkwright/kanon#2975): same has_ddl bare-substring-match false
+# positive as the lexis entries above, same root cause, same upstream fix
+# tracked there. `database.rs`'s module doc names "CREATE TABLE" as Phase
+# 04 territory this facade does not implement yet (Decision 1: "one file
+# per database" today, no table catalog); `row.rs`'s module doc cites
+# `lexis`'s own exit-criterion wording, which likewise names "CREATE TABLE"
+# only to say the typed-row/`TableDef` format it gates is Phase 4/5
+# territory. Neither file runs a migration or executes DDL. Drop these two
+# entries once kanon#2975 lands.
+STORAGE/no-migration-checksum:crates/pinax/src/database.rs
+STORAGE/no-migration-checksum:crates/pinax/src/row.rs
+
+# WHY: lib.rs is a module-declaration + public re-export file with no
+# behavior of its own — same shape as the lexis/lib.rs entry above, not the
+# "empty scaffold" reason the hypomnema/phylaxis entries below still carry.
+# Phase 01 landed real behavior in `btree/`, `buffer_pool.rs`, `codec.rs`,
+# `database.rs`, `page.rs`, `pager.rs`, and `row.rs`, each with its own
+# colocated `#[cfg(test)] mod tests`; the TESTING/no-tests heuristic only
+# inspects lib.rs itself and a sibling `tests/` directory (which exists here
+# too — `tests/phase01_acceptance.rs` — but the heuristic still cannot see
+# module-level coverage), so it cannot see any of that. Drop once the
+# heuristic learns to walk sibling submodules.
+TESTING/no-tests:crates/pinax/src/lib.rs
+
+# =============================================================================
+# crates/{hypomnema,phylaxis}/ — genuinely empty scaffolds
+# =============================================================================
+
+# WHY: these two crates are empty by design at this phase — lib.rs is
 # module-doc-comment-only, zero functions, zero types, zero logic (see each
 # crate's [package.metadata.kanon] maturity = "scaffold" in Cargo.toml). A
 # #[cfg(test)] module here would have no real behavior to assert against;
@@ -42,9 +70,7 @@ STORAGE/no-migration-checksum:crates/lexis/src/lib.rs
 # is worse than an honest, exit-criteria-bound suppression. Each crate's own
 # Cargo.toml exit-criteria field names the concrete Phase that ends this:
 # hypomnema Phase 02 (WAL + virtual-WAL trait + causal changelog), phylaxis
-# Phase 03/06 (MVCC + encryption), pinax Phase 01 (pager + buffer pool +
-# B-tree) — see kanon/projects/pinax/ROADMAP.md. Drop each entry the moment
-# its crate gets real behavior and a real test.
+# Phase 03/06 (MVCC + encryption) — see kanon/projects/pinax/ROADMAP.md.
+# Drop each entry the moment its crate gets real behavior and a real test.
 TESTING/no-tests:crates/hypomnema/src/lib.rs
 TESTING/no-tests:crates/phylaxis/src/lib.rs
-TESTING/no-tests:crates/pinax/src/lib.rs

Cargo.toml

@@ -31,6 +31,16 @@ proptest = "1"
 # lets ARCHITECTURE/crate-index-conformance read real `path =` edges out of
 # each crate's Cargo.toml and cross-check them against CRATE-INDEX.toml.
 snafu = "0.8"
+# WHY: pinax's page-format checksum per PLAN.md Decision 2 — pure-Rust
+# XxHash3-64, the same algorithm Turso's page cache uses
+# (`turso/core/storage/checksum.rs:81`), chosen over CRC32C (32-bit is
+# collision-borderline at millions of pages) and blake3 (256-bit dominates a
+# 4 KiB page's overhead budget; cryptographic integrity is the encryption
+# layer's job, not corruption detection's).
+xxhash-rust = { version = "0.8", features = ["xxh3"] }
+# WHY: fixture temp-dirs for pager/buffer-pool/B-tree tests — kanon RUST.md
+# § Testing forbids `/tmp/<literal>` and `.parent()` off a tempdir.
+tempfile = "3"
 
 [workspace.lints.clippy]
 pedantic = { level = "warn", priority = -1 }

README.md

@@ -9,10 +9,11 @@ in Rust. The Tier-1 storage primitive: the answer to "what is." Replaces
 SQLite for fleet consumers whose state is tabular, transactional, and
 aggregation-heavy.
 
-**Status:** design phase - no code yet. This repo carries the public
-specification surface as it lands; the phased roadmap and state tracking
-live in the fleet planning home. First implementation phase is the
-pager / buffer pool / B-tree tier.
+**Status:** Phase 01 (pager / buffer pool / B-tree) implemented - a
+checksummed, copy-on-write, integer-keyed B+tree behind an LRU-evicting
+buffer pool. The phased roadmap and state tracking live in the fleet
+planning home; `lexis` (the strict six-type value system) landed first,
+as the vocabulary Phase 01 stores.
 
 ## Locked design decisions
 

crates/pinax/CRATE-SHAPE.toml

@@ -6,4 +6,4 @@
 shape = "unclassified"
 top_level = "layer"
 within_crate = "feature"
-purpose = "Facade: pager, buffer pool, B-tree, page format, SQL surface, async API, migration runner, CLI (Decision 14). Not yet implemented — reserves the workspace position."
+purpose = "Facade: pager, buffer pool, B-tree, page format land Phase 01 (Decision 1, Decision 2). SQL surface, async API, migration runner, CLI (Decision 14) land later phases."

crates/pinax/Cargo.toml

@@ -7,7 +7,7 @@ license.workspace = true
 homepage.workspace = true
 repository.workspace = true
 authors.workspace = true
-description = "(scaffold) Relational storage facade for the forkwright fleet: pager, buffer pool, B-tree, SQL surface, async API, migrations, and CLI."
+description = "Relational storage facade for the forkwright fleet: page format, pager, buffer pool, and B+tree land Phase 01; SQL surface, async API, migrations, and CLI land later phases."
 readme = "../../README.md"
 keywords = ["sql", "database", "storage"]
 categories = ["database"]
@@ -16,17 +16,23 @@ categories = ["database"]
 hypomnema = { path = "../hypomnema", version = "0.0.2" }
 lexis = { path = "../lexis", version = "0.0.2" }
 phylaxis = { path = "../phylaxis", version = "0.0.2" }
+snafu = { workspace = true }
+xxhash-rust = { workspace = true }
+
+[dev-dependencies]
+tempfile = { workspace = true }
 
 [lints]
 workspace = true
 
 # WHY: maturity flag is read by the kanon substrate registry so consumers can
-# tell at a glance which API surfaces are pre-stabilization. This crate is an
-# empty scaffold reserving the facade's position in the Decision 14
-# dependency graph; the pager + buffer pool + B-tree land Phase 01.
+# tell at a glance which API surfaces are pre-stabilization. Phase 01 lands
+# the page format, pager, buffer pool, and B+tree (Decision 1, Decision 2,
+# Decision 14); the SQL surface, async API, migration runner, and CLI are
+# later phases per ROADMAP.md.
 [package.metadata.kanon]
-maturity = "scaffold"
+maturity = "alpha"
 since = "2026-08-15"
 phase = "1"
-phase-description = "empty crate reserving the workspace position and dependency edges; no implementation yet"
-exit-criteria = "Phase 1 pager + buffer pool + B-tree land; see kanon/projects/pinax/ROADMAP.md Phase 01"
+phase-description = "page format (checksummed, configurable 4-64 KiB), pager, buffer pool with LRU eviction, and a copy-on-write B+tree keyed by i64 (Decision 1, Decision 2)"
+exit-criteria = "Phase 2 adopts this pager under a WAL + transaction boundary; see kanon/projects/pinax/ROADMAP.md Phase 02"