Use sentence case for all Sphinx documentation headings

Align heading style with Plone documentation style guide which follows
the Microsoft Writing Style Guide: only capitalize the first word and
proper nouns in headings. Affects 12 files across tutorials, how-to
guides, reference, and explanation sections.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: jensens

docs/sources/explanation/architecture.md

@@ -42,7 +42,7 @@ flowchart LR
     JW --> JS2["JSON string<br/>(direct write)"]
 ```
 
-### Step 1: Pickle parsing (decode.rs)
+### Step 1: pickle parsing (decode.rs)
 
 The decoder implements a pickle virtual machine that reads opcodes and builds
 up the `PickleValue` tree. It maintains a stack, a memo (for shared

docs/sources/explanation/optimization-journal.md

@@ -1,4 +1,4 @@
-# Optimization Journal
+# Optimization journal
 
 <!-- diataxis: explanation -->
 
@@ -31,7 +31,7 @@ of cloning avoids a deep copy of the entire PickleValue tree.
 
 **Impact:** ~1.8x faster encode path.
 
-### 2. Bypass serde_json intermediate
+### 2. bypass serde_json intermediate
 
 **Technique:** Added `src/pyconv.rs` for direct `PickleValue` to/from
 `PyObject` conversion, eliminating `serde_json::Value` as an intermediate
@@ -47,7 +47,7 @@ allocation and traversal.
 
 **Impact:** 2-3.5x faster across most categories.
 
-### 3. Decode path tuning
+### 3. decode path tuning
 
 **Technique:** Single-pass Dict decode, pre-allocated `Vec::with_capacity()`,
 set/frozenset move semantics (no `Vec` clone), `StackItem` removal, `@`
@@ -62,7 +62,7 @@ keys like `"title"`, `"description"`) skip all 15+ marker key comparisons.
 
 **Impact:** Codec competitive with CPython pickle decode.
 
-### 4. Encode marker fast path
+### 4. encode marker fast path
 
 **Technique:** Single-key dicts use direct key extraction plus match instead
 of 15 sequential `dict.get_item()` calls. 2-4 key dicts use a single-pass
@@ -81,7 +81,7 @@ prefixed keys and matches on the first character after `@`.
 
 **Impact:** Reduced `get_item()` calls for `@cls` + `@s` from ~16 to 2.
 
-### 5. Direct PyObject encoder
+### 5. direct PyObject encoder
 
 **Technique:** `pyconv.rs` encode path writes pickle opcodes straight from
 Python objects, bypassing the PickleValue tree allocation. Handles common
@@ -95,7 +95,7 @@ we eliminate the entire intermediate tree.
 
 **Impact:** Encode competitive with CPython C pickler.
 
-### 6. Shared ZODB memo
+### 6. shared ZODB memo
 
 **Technique:** Single Decoder instance processes both class and state pickles,
 preserving memo entries across the boundary.
@@ -124,7 +124,7 @@ requests concurrently.
 single-thread speedup, but significant throughput improvement in
 multi-threaded deployments.
 
-### 8. Single-pass PG decode
+### 8. single-pass PG decode
 
 **Technique:** `decode_zodb_record_for_pg()` combines decode + persistent
 reference extraction + null-byte sanitization in one function, eliminating two
@@ -141,7 +141,7 @@ two full traversals.
 
 ## v1.3.0
 
-### 9. Box Instance variant
+### 9. box instance variant
 
 **Technique:** Changed `Instance(InstanceData)` to `Instance(Box<InstanceData>)`
 in the `PickleValue` enum.
@@ -164,7 +164,7 @@ utilization.
 
 ## v1.3.1
 
-### 10. Thin LTO + codegen-units=1
+### 10. thin LTO + codegen-units=1
 
 **Technique:** Set `lto = "thin"` and `codegen-units = 1` in the Cargo release
 profile.
@@ -185,7 +185,7 @@ increase is modest and the performance gain is consistent.
 
 ## v1.4.0
 
-### 11. Direct pickle-to-JSON string path
+### 11. direct pickle-to-JSON string path
 
 **Technique:** `decode_zodb_record_for_pg_json()` converts ZODB pickle records
 entirely in Rust, producing a JSON string directly. The GIL is released for
@@ -204,7 +204,7 @@ allocation.
 **Impact:** 1.3x faster full pipeline on real-world data, plus improved
 multi-threaded throughput from GIL release.
 
-## v1.5.0 -- Encode Performance Rounds 1-4
+## v1.5.0 -- encode performance rounds 1-4
 
 ### Round 1
 
@@ -232,7 +232,7 @@ this replaces ~20 string comparisons with a single hash probe.
 **Impact:** Measurable improvement across all encode benchmarks, especially
 for integer-heavy and large-dict payloads.
 
-#### 13. Profile-Guided Optimization (PGO)
+#### 13. profile-guided optimization (PGO)
 
 **Technique:** Two-pass instrumented build. First pass collects execution
 profiles using real FileStorage data (5 MB Wikipedia database, 1,692 records)
@@ -259,7 +259,7 @@ can exploit.
 
 ### Round 2
 
-#### 14. Direct known-type encoding
+#### 14. direct known-type encoding
 
 **Technique:** For `@dt`, `@date`, `@time`, `@td`, `@dec` markers: write
 pickle opcodes directly to the output buffer instead of allocating
@@ -279,7 +279,7 @@ sequence of buffer writes.
 
 **Impact:** special_types **9.2x faster** than Python pickle.
 
-#### 15. Thread-local buffer reuse
+#### 15. thread-local buffer reuse
 
 **Technique:** `ENCODE_BUF` thread-local `Vec<u8>` retains capacity across
 calls. On each encode, the buffer is cleared (length reset to 0) but the
@@ -305,7 +305,7 @@ sharing), and the memory cost is bounded (one buffer per thread).
 
 ### Round 3
 
-#### 16. Direct JSON string writer
+#### 16. direct JSON string writer
 
 **Technique:** `json_writer.rs` writes JSON tokens directly from the
 `PickleValue` AST to a `String` buffer. Fast-path string escaping (scan for
@@ -333,7 +333,7 @@ floats to strings without going through `format!()`, avoiding a temporary
 
 ### Round 4
 
-#### 17. Class pickle cache
+#### 17. class pickle cache
 
 **Technique:** Thread-local cache per `(module, name)` pair. Linear search
 over ~6 entries (faster than `HashMap` for small sets). First call builds and

docs/sources/explanation/why-json.md

@@ -1,4 +1,4 @@
-# Why Convert Pickle to JSON?
+# Why convert pickle to JSON?
 
 <!-- diataxis: explanation -->
 

docs/sources/how-to/index.md

@@ -1,4 +1,4 @@
-# How-To Guides
+# How-to guides
 
 <!-- diataxis: how-to -->
 

docs/sources/how-to/run-benchmarks.md

@@ -79,21 +79,21 @@ All benchmark commands accept:
 
 Profile-Guided Optimization (PGO) produces the most accurate performance numbers by optimizing based on actual benchmark workloads.
 
-### 1. Install LLVM tools
+### 1. install LLVM tools
 
 ```bash
 rustup component add llvm-tools
 ```
 
-### 2. Instrumented build
+### 2. instrumented build
 
 Build with profiling instrumentation enabled:
 
 ```bash
 RUSTFLAGS="-Cprofile-generate=/tmp/pgo-data" maturin develop --release
 ```
 
-### 3. Generate profiles
+### 3. generate profiles
 
 Run both benchmark types to capture representative workload data:
 
@@ -102,13 +102,13 @@ python benchmarks/bench.py synthetic --iterations 5000
 python benchmarks/bench.py filestorage benchmarks/bench_data/Data.fs
 ```
 
-### 4. Merge profile data
+### 4. merge profile data
 
 ```bash
 llvm-profdata merge -o /tmp/pgo-data/merged.profdata /tmp/pgo-data/*.profraw
 ```
 
-### 5. Final PGO build
+### 5. final PGO build
 
 ```bash
 RUSTFLAGS="-Cprofile-use=/tmp/pgo-data/merged.profdata" maturin develop --release

docs/sources/index.md

@@ -63,7 +63,7 @@ Designed as the codec layer for a PostgreSQL JSONB storage backend.
 
 ::::
 
-## Quick Start
+## Quick start
 
 1. {doc}`Install zodb-json-codec <how-to/install>`
 2. {doc}`Encode and decode your first pickle <tutorials/getting-started>`

docs/sources/reference/btree-format.md

@@ -1,12 +1,12 @@
-# BTree Format
+# BTree format
 
 <!-- diataxis: reference -->
 
 BTrees from the `BTrees` package store their state as deeply nested
 tuples in pickle. The codec flattens these into human-readable,
 JSONB-queryable JSON while preserving full roundtrip fidelity.
 
-## Supported BTree Classes
+## Supported BTree classes
 
 All classes matching the pattern `BTrees.{PREFIX}BTree.{PREFIX}{Type}`
 are recognized, where `{Type}` is one of `BTree`, `Bucket`, `TreeSet`,
@@ -28,7 +28,7 @@ or `Set`.
 | `QQ` | unsigned long | unsigned long |
 | `fs` | FileStorage | (internal) |
 
-### Node Types
+### Node types
 
 BTree
 : Tree node for maps. Small trees use 4-level tuple nesting for inline
@@ -44,7 +44,7 @@ TreeSet
 Set
 : Leaf node for sets. Same structure as Bucket but stores keys only.
 
-## `@kv` -- Key-Value Pairs
+## `@kv` -- key-value pairs
 
 Used for map-type BTree nodes (BTree, Bucket). Contains an array of
 `[key, value]` pairs.

docs/sources/reference/json-format.md

@@ -1,12 +1,12 @@
-# JSON Format
+# JSON format
 
 <!-- diataxis: reference -->
 
 This page describes how Python types are represented in JSON by
 zodb-json-codec. All representations are **roundtrip-safe**: encoding to
 JSON and decoding back produces identical pickle bytes.
 
-## Native JSON Types
+## Native JSON types
 
 These Python types map directly to JSON without any markers:
 
@@ -20,12 +20,12 @@ These Python types map directly to JSON without any markers:
 | `list` | array | `[1, 2, 3]` |
 | `dict` (string keys) | object | `{"key": "value"}` |
 
-## Structural Markers
+## Structural markers
 
 These markers preserve Python types that have no direct JSON equivalent.
 Each uses a single-key dict with a `@`-prefixed key.
 
-### `@t` -- Tuple
+### `@t` -- tuple
 
 ```json
 {"@t": [1, 2, 3]}

docs/sources/reference/project-structure.md

@@ -1,11 +1,11 @@
-# Project Structure
+# Project structure
 
 <!-- diataxis: reference -->
 
 Source layout and module responsibilities for the zodb-json-codec Rust
 crate and Python package.
 
-## Directory Layout
+## Directory layout
 
 ```
 src/
@@ -34,9 +34,9 @@ benchmarks/
   bench.py          # Performance benchmarks vs CPython pickle
 ```
 
-## Rust Modules
+## Rust modules
 
-### `lib.rs` -- PyO3 Module
+### `lib.rs` -- PyO3 module
 
 Defines the Python-facing functions (`#[pyfunction]`) that are exported
 as the `zodb_json_codec._rust` extension module. Each function
@@ -56,7 +56,7 @@ Also defines `InstanceData` (module, name, state, plus optional
 `dict_items` and `list_items` for subclass support). The `Instance`
 variant is boxed to keep the enum size at 48 bytes.
 
-### `decode.rs` -- Pickle Decoder
+### `decode.rs` -- pickle decoder
 
 Implements a subset of the pickle virtual machine sufficient for ZODB
 records (protocol 2-3, partial protocol 4). Reads pickle bytes and
@@ -71,7 +71,7 @@ Key functions:
 Safety limits: memo capped at 100,000 entries, binary allocations capped
 at 256 MB, LONG text at 10,000 characters.
 
-### `encode.rs` -- Pickle Encoder
+### `encode.rs` -- pickle encoder
 
 Converts a `PickleValue` AST back to pickle bytes in protocol 3 (the
 maximum supported by zodbpickle). Handles all value types including
@@ -80,7 +80,7 @@ maximum supported by zodbpickle). Handles all value types including
 Recursion depth is limited to 1,000 levels. Integers are encoded with
 minimal byte length for signed little-endian representation.
 
-### `pyconv.rs` -- Direct PyObject Bridge
+### `pyconv.rs` -- direct PyObject bridge
 
 The fast path for the Python dict API. Converts between `PickleValue`
 AST and Python objects directly, bypassing the `serde_json::Value`
@@ -98,7 +98,7 @@ Provides both standard and PG-specific variants:
   BTree-aware decode.
 - `collect_refs_from_pickle_value` -- extract persistent reference OIDs.
 
-### `json.rs` -- JSON String Path
+### `json.rs` -- JSON string path
 
 Converts between `PickleValue` AST and `serde_json::Value` for the JSON
 string API (`pickle_to_json`, `json_to_pickle`). Also provides the
@@ -114,15 +114,15 @@ Key functions:
 - `pickle_value_to_json_string_pg` -- direct string output for PG
   (uses `json_writer.rs`).
 
-### `json_writer.rs` -- Direct JSON String Writer
+### `json_writer.rs` -- direct JSON string writer
 
 A low-level JSON token writer that appends directly to a `String`
 buffer. Used by the PG JSON path to avoid allocating intermediate
 `serde_json::Value` nodes entirely. Writes JSON tokens (object open/
 close, array open/close, strings, numbers, booleans, null) as raw
 characters.
 
-### `known_types.rs` -- Known Type Handlers
+### `known_types.rs` -- known type handlers
 
 Intercepts common Python REDUCE patterns at the PickleValue/JSON
 boundary and produces compact typed markers instead of generic `@reduce`
@@ -139,7 +139,7 @@ output. Handles both directions:
 Full timezone support: naive, fixed-offset (`datetime.timezone`),
 pytz (including named zones with full constructor args), and zoneinfo.
 
-### `btrees.rs` -- BTree State Handling
+### `btrees.rs` -- BTree state handling
 
 Classifies BTree classes by module/name and flattens their deeply nested
 tuple state into queryable JSON. Handles both directions:
@@ -151,7 +151,7 @@ tuple state into queryable JSON. Handles both directions:
 - **Classification**: `classify_btree` -- identify BTree class and node
   kind from module/name strings.
 
-### `zodb.rs` -- ZODB Record Handling
+### `zodb.rs` -- ZODB record handling
 
 Handles the ZODB two-pickle record format. Provides:
 
@@ -164,19 +164,19 @@ Handles the ZODB two-pickle record format. Provides:
 Also contains `#[cfg(test)]` encode functions for ZODB record
 roundtrip testing.
 
-### `opcodes.rs` -- Pickle Opcode Constants
+### `opcodes.rs` -- pickle opcode constants
 
 Defines constants for all pickle opcodes from protocol 0 through 5. The
 codec focuses on protocol 2-3 (ZODB standard) but includes protocol 4-5
 opcodes for partial forward compatibility.
 
-### `error.rs` -- Error Types
+### `error.rs` -- error types
 
 Defines `CodecError` with variants for all failure modes: unexpected
 EOF, unknown opcode, stack underflow, invalid data, JSON errors, and
 invalid UTF-8. Implements conversion to Python `ValueError` via PyO3.
 
-## Data Flow
+## Data flow
 
 The following diagram shows the three conversion paths through the
 codebase: