v0.2 Evidence

v0.2 is the credibility milestone. The release has to show that AionDB is a working PostgreSQL-wire database foundation rather than a demo path, with a graph surface that holds up against Neo4j-class expectations. This page lists the evidence that has to exist before broad claims are made.

v0.2 does not claim production completeness. It only makes storage, WAL, driver behavior, type behavior, graph behavior, and baseline evaluation inspectable from a fresh checkout.

Evidence matrix

Required local checks

Run the contract checks before treating v0.2 as credible:

cargo test -p aiondb-buffer-pool frozen_layout_v1
cargo test -p aiondb-storage-engine storage_compat
cargo test -p aiondb-wal frozen_wal
cargo test -p aiondb-wal reader_replay_is_idempotent

Run the compatibility matrix when the required client tools are installed:

cargo xtask ecosystem-compat --list
cargo xtask ecosystem-compat --report tmp/clean_prod/ecosystem_compat.json

Run the graph correctness checks before making graph maturity claims:

cargo test -p aiondb-graph-api -p aiondb-graph-projection -p aiondb-graph
cargo test -p aiondb-storage-engine adjacency
cargo test -p aiondb-executor cypher_graph
cargo test -p aiondb-engine graph

Run a small baseline benchmark only after correctness checks pass:

cargo build --release -p aiondb-server --bin aiondb

BENCH_ENGINES=aiondb \
PGBENCH_SCALE=1 \
PGBENCH_CLIENTS=1 \
PGBENCH_DURATION=10 \
benchmarks/run.sh pgbench

The benchmark output is not a claim unless it includes the command, commit hash, durability settings, hardware, raw output, and correctness checks.

Storage and WAL acceptance

v0.2 should not move forward unless:

• every stable file kind has documented magic bytes or a documented manifest entry;
• unknown future storage majors are refused;
• v0.1 data directories have a tested upgrade path;
• the WAL tag table is dense, frozen, and asserted by tests;
• replay from the same start LSN is deterministic across reopens;
• corrupt WAL tails stop recovery cleanly instead of producing undefined state.

Driver and type acceptance

Driver evidence should name exact versions and protocol paths:

Type evidence should include:

• PG name and OID reported through catalog and row descriptions;
• text format round-trip;
• binary format status: supported, rejected, or not requested by the driver;
• null round-trip;
• cast behavior and overflow error class;
• unsupported types returning SQLSTATE 0A000 or another documented error class.

Graph acceptance

v0.2 treats graph as a real part of the engine, not an isolated demo. Neo4j-class maturity is the engineering target, with evidence for what is already comparable and what is still behind.

Required graph evidence:

• stable node, edge, label, property, path, and projection semantics;
• documented behavior for duplicate edges, null endpoints, self loops, multi-label entities, and path uniqueness;
• SQL/PGQ and Cypher-compatible subsets named precisely;
• algorithm correctness fixtures for traversal, shortest path, connected components, centrality, and community-style workloads where implemented;
• resource controls for heavy algorithms: memory budget, timeout, result limit, and abort behavior;
• migration notes for Neo4j label-property graphs, including unsupported Cypher features;
• graph-specific metrics for traversal depth, frontier size, index hit rate, memory use, timeout aborts, and result truncation.

Required Neo4j-class benchmark fields:

Allowed graph claim:

On this pinned workload, AionDB is faster/comparable/slower than Neo4j, with these query shapes and this raw output.

Disallowed graph claim:

AionDB is mature as Neo4j.

The second form is too broad unless every relevant surface is backed by benchmark, compatibility, migration, and operations evidence.

Baseline benchmark acceptance

The v0.2 benchmark baseline is meant to expose bottlenecks for later milestones. It should not claim that AionDB is faster than PostgreSQL, Neo4j, or distributed SQL systems.

Required baseline shapes:

• startup and first connection;
• simple insert path;
• point read by primary key;
• range scan over an indexed column;
• WAL append and flush path;
• checkpoint or snapshot publication where implemented;
• restart recovery after committed writes.

Required report fields:

Claim language

Use precise v0.2 language:

• "AionDB speaks PostgreSQL wire protocol."
• "This driver path has been tested against these operations."
• "This storage/WAL behavior is part of the v0.2 contract."
• "This graph workload has been compared with a pinned Neo4j version and raw output."
• "This benchmark is a baseline for a named commit and configuration."

Avoid broad language:

• "PostgreSQL compatible" without a compatibility matrix.
• "Faster than PostgreSQL" without a named benchmark and raw output.
• "Production ready" before the v1.0 support, security, backup, HA, and operations bars are met.
• "Supports Neo4j/CockroachDB-style workloads" before the later milestone evidence exists.

Exit checklist

Before tagging v0.2, the release should have:

• storage and WAL docs aligned with code;
• storage doctor and upgrade tests passing;
• WAL frozen tag tests passing;
• graph semantics, algorithm, benchmark, migration, and operations evidence recorded for supported Neo4j-class claims;
• at least one public driver matrix report;
• type mapping page reviewed against catalog output;
• benchmark baseline commands and raw output retained;
• limitations page updated for unsupported PostgreSQL, storage, graph, vector, and operations behavior;
• release notes that separate implemented behavior from roadmap intent.