v0.5.0

Minima More v0.5.0

The conformance + analytics release. Apache-2.0 licensed.

Highlights

Script conformance fixes (consensus-critical)

• HEX() now matches Java: HEX(255)=0xFF, HEX(128)=0x80,
  HEX(0)=0x00, HEX(256)=0x0100 (the leading 0x00 sign byte is removed).
• cleanScript() now matches Java exactly, including operator spacing.
  All 18 deployed Stables vault/contract scripts derive the exact Java
  addresses, including all 8 vault contracts.
• Long-hex crash fixed: token classifiers no longer use recursive
  std::regex; a 12,000-byte hex script parses, cleans and executes.
• Regression tests: test_script_conformance, test_consensus_vectors 12/12.
• Apache-2.0: LICENSE, NOTICE, SECURITY.md, and provenance in README.

Archive

• Chunked cross-block compression: 256 blocks/chunk (ZCHN), 3.83x on the
  full archive; legacy per-row ZBLK rows still read.
• Complete imports: the final cascade-depth tail is saved, giving
  2,376,119 blocks / 0 gaps / 0 missing chunks on the full archive.
• archive action:integrity: 2,376,119 blocks, 0 errors.
• Record battery: 260 heights, parent linkage, ID lookup, round-trip and
  chunk-boundary ranges pass (~2 ms average).
• IBD cooldown fixed: -ibdlimit N (default 30, 0 = unlimited) now
  actually expires; the old set never did.

minima-analytics sidecar (optional)

• Archive → DuckDB indexer; portable -backup / -restore.
• Read-only API: capabilities/catalog, blocks, txs, addresses, tokens, scripts,
  magic numbers, consistency, UTXO distribution, HODL waves, cohorts, metrics.
• Full index measured: 2,376,119 blocks, 12.5 min, 2.7 GB DuckDB,
  2.86 GB peak RSS.
• 8,858 tokens / 7,851 NFTs (decimals 0); 1,167 embedded NFT images
  decoded and served (/media.bin), plus URL media metadata.
• 320,015 script hashes; known-script registry seeded from
  minima-global/MiniDAPP.
• Magic numbers indexed per block (max_txpow_size, max_kissvm_ops,
  max_txns, min_work) with /v1/magic and change history.
• Burn, spent-age/ASOL, CDD/CDD-90/CYD, dormancy, address activity and UTXO
  percentiles. Minima has no fee market, so fee metrics are not included.
• Backup: full 4.4 GB DuckDB → 1.2 GB Parquet in 4.7 s, with
  analytics_manifest.json; restore verified at 2,376,119 blocks.

BI templates (optional)

• Evidence (analytics/integrations/evidence/, MIT) — historical
  reports: blocks/txs/burn, HODL waves, CDD, UTXO/cohorts, tokens/NFTs, scripts.
• Rill (analytics/integrations/rill/, Apache-2.0) — live/ops
  dashboards: tip/lag, recent blocks, burn, magic changes, script activity,
  alerts; attaches the DuckDB index.
• Shared metric ownership/refresh contract:
  analytics/integrations/common/metrics.yaml.

Artifacts

• minima-more-v0.5.0-linux-{x86_64,aarch64} — standard dynamic node
• minima-more-v0.5.0-linux-{x86_64,aarch64}-static — fully static node
• minima-more-analytics-v0.5.0-linux-{x86_64,aarch64} — sidecar + bundled
  libduckdb.so
• .sha256 for every tarball

Upgrade notes

• Archive DBs remain compatible; old per-row blobs are read transparently.
• -ibdlimit 0 is intended for trusted peers/benchmarks; default remains 30 min.
• Analytics requires DuckDB; the release tarball bundles libduckdb.so with an
  $ORIGIN rpath. Node builds are unchanged (MINIMA_BUILD_ANALYTICS=OFF).
• MMR enrichment reports comparable=false when the MMR snapshot is older than
  the archive tip; it is not a corruption signal.

Verify

tar -xzf minima-more-v0.5.0-linux-<arch>[-static].tar.gz
sha256sum -c minima-more-v0.5.0-linux-<arch>[-static].tar.gz.sha256
./minima-more-v0.5.0-linux-<arch> -manifest

tar -xzf minima-more-analytics-v0.5.0-linux-<arch>.tar.gz
./minima-more-analytics-v0.5.0-linux-<arch> -help

License: Apache-2.0. See LICENSE, NOTICE, and SECURITY.md.

v0.2.0

Minima More v0.2.0

This release hardens the node, fixes legacy data import (the user-reported
blocker), and cuts mainnet memory dramatically. It also adds the capability
manifest and the in-node MCP server.

Highlights

• Legacy imports now work: MegaMMR (megammr.mmr and Minima More
  .megammr), raw archives (.raw.dat), and legacy H2 .gzip archive dumps.
• ~70–95 MB less RSS on a synced mainnet node via role-based SQLite caches
  and returning free glibc arenas to the OS.
• In-node MCP (stdio + HTTP/SSE) generated from a capability manifest over
  all 111 commands.
• SSRF hardening for outbound fetches and a tightened RPC read-only allowlist.

Security

• Bounded fetcher for outbound URLs (RPCClient::sendGET): HTTP/HTTPS only,
  rejects loopback/private/link-local/CGNAT/ULA hosts, max 3 redirects, 15 s
  timeout, 1 MiB response cap.
• RPC read-only allowlist tightened:
  • removed vault (seed phrase exposure), plus getaddress, keys,
    runscript, megammr;
  • removed peers (its url: action was an SSRF vector).

Import & data migration

• MegaMMR import fixed (root cause): MegaMMRBackup::readDataStream
  required pre-set MegaMMR/IBD objects, and loadObjectSlow() swallowed the
  resulting exception — so every megammr action:import silently failed.
  • Now allocates its own objects (matching the Java reference).
  • Auto-detects versioned .megammr vs legacy raw megammr.mmr; the
    reply reports legacyraw.
  • megammr action:integrity validates either form and handles a missing IBD.
• Legacy H2 archive import: archive action:import file:x.gzip now reads the
  gzipped H2 SQL script (SCRIPT … COMPRESSION GZIP, SYNCBLOCK /
  CASCADEDATA, X'…' blobs), converts it to the raw archive stream, and reuses
  the normal resync flow. This is a focused archive-schema reader, not a general
  H2 engine.
• Raw archive import verified for both implementations; raw:true forces raw
  parsing for renamed files.

Memory & performance

Measured on one host: AWS c6g.xlarge (4 vCPU / 8 GB / Neoverse-N1).

• Role-based SQLite cache: 4 MiB (p2p/userprefs), 8 MiB (hot state),
  16 MiB (archive) instead of 16 MiB × 6 connections.
  • Mainnet anonymous RSS 224 MB → 160 MB at the same block height.
  • Env: MINIMA_SQLITE_CACHE_KIB, MINIMA_SQLITE_MMAP_MB.
• malloc_trim(0) after key generation + periodic checker/cleanup:
  • steady-state mainnet RSS 261–282 MB → ~196 MB automatically;
  • memreport trim:true reaches ~169 MB (freed glibc arenas returned to
    the OS).
• Bounded startup key precompute: 16 keys (≤2 GB), 32 (≤8 GB), 64 (larger);
  MINIMA_INIT_KEYS overrides.
• Archive cache on non-archive nodes: 16 MiB/128 MiB → 4 MiB/32 MiB.
• Optional jemalloc: -DMINIMA_USE_JEMALLOC=ON; A/B showed −35 MB
  steady-state (219 → 180 MB).
• memreport command + -memreport flag: process RSS (total/anonymous/file),
  per-database sizes incl. WAL, byte attribution for tree/cascade/RAM TxPoW/MMR,
  glibc arena/in-use/free, trim:true, and dump:true (jemalloc profile).

Command manifest & in-node MCP

• Capability manifest generated from the registry: node header (version,
  network, mode, serves) + 111 commands with category, annotations
  (readOnly / destructive / idempotent / sensitive) and typed params.
• capabilities command, -manifest / -manifest md.
• In-node MCP: initialize, tools/list, tools/call, ping,
  notifications.
  • -mcp stdio (logs routed to stderr; stdout carries only MCP frames).
  • -mcp http: POST /mcp and GET /mcp SSE on the RPC port, after the RPC
    authorisation check; read-only credentials are refused.
• Generated docs: COMMAND_REFERENCE.md (regenerate with
  gen_command_reference; test_docs_manifest prevents drift).

Correctness

• JSON writer: put(key, "literal") no longer emits unquoted values
  ("type":simple); const char* is now quoted/escaped.
• Node flags: -mcp (and -memreport) are stripped before
  ParamConfigurer, fixing the “Unknown argument” startup failure.
• Benchmark harness: Java classpath resolution walks up to
  reference/minima-core (fixes NoClassDefFoundError: org/h2/Driver); Java pid
  resolved after -daemon fork; MODE=full|archive|megammr support.

Benchmarks & docs

• One-host results on AWS c6g.xlarge (4 vCPU / 8 GB / Neoverse-N1): live
  mainnet, core crypto/MMR, lifecycle, and storage engines.
• README rewritten around those numbers; all SVG charts and the hero infographic
  regenerated.
• Checked-in results: reference-bench/results/aws8-{mainnet,core,sweep,engines}.

Tests

• test_capabilities 11/11
• test_mcp_server 23/23
• test_docs_manifest 2/2
• test_megammr_import 6/6
• test_h2_import 5/5
• Existing gates still pass: test_consensus_vectors, test_mmr,
  test_winternitz, test_mempool_index, test_sql_batch.

Upgrade notes

• No on-disk format change for wallet/chain; existing data directories carry
  over.
• New flags/commands: -manifest, -mcp stdio|http, memreport, capabilities.
• To import legacy data: megammr action:import file:… and
  archive action:import file:….gzip (or .raw.dat).
• No config migration required; new environment knobs are optional.

Static builds

A fully static variant is included for both architectures
(minima-v0.2.0-linux-<arch>-static): it links SQLite, OpenSSL, zlib and a
minimal static libcurl (OpenSSL + zlib only) so it has no runtime library
dependencies. (Static glibc still uses the host’s NSS config for DNS, as usual.)

Known issues / in progress

• The H2 importer covers the SQL script export
  (SCRIPT … COMPRESSION GZIP); a raw H2 database file (.mv.db) requires
  running SCRIPT TO on the legacy node first.
• Archive exportraw on a node whose archive has no syncblock rows beyond the
  cascade can produce a cascade-only file (data-dependent; logic matches the
  Java reference).

Install

Download the tarball for your architecture and verify the checksum:

# standard (dynamic libs) or -static (fully statically linked)
tar -xzf minima-v0.2.0-linux-<arch>[-static].tar.gz
sha256sum -c minima-v0.2.0-linux-<arch>[-static].tar.gz.sha256
./minima-v0.2.0-linux-<arch>[-static] -manifest

Full changelog (since v0.1.0)

57048ad feat(archive): import legacy H2 .gzip archive dumps
6ec50bd fix(import): legacy + Minima More MegaMMR/archive import
dc4f324 perf(memory): return freed glibc arenas to the OS (malloc_trim)
8ba97ef bench(harness): add MODE=full|archive|megammr support
21ac279 perf(sqlite): shrink the archive cache on non-archive nodes
0040d0b feat(memory): add live memreport command + -memreport flag
3037439 perf: role-based SQLite cache budget, bounded key precompute, optional jemalloc
013fb02 feat(rfc-0011): generated capability manifest + in-node MCP server
7747340 docs(readme): rewrite around the one-host 8 GB Graviton results; regenerate all figures
7e4acc6 bench(results): core crypto/MMR C++ vs Java on AWS c6g.xlarge (aarch64)
d58de20 bench(results): live-mainnet C++ vs Java on AWS c6g.xlarge (4 vCPU/8GB, Neoverse-N1)
a30f64b bench(env): locate reference/minima-core by walking up, not fixed ../..
c1bc6c9 fix(json): quote/escape const char* values in JSON writer
8120f1a bench(mainnet): resolve real node pid after -daemon fork so java metrics populate
42043bd security(ssrf): bounded fetcher in RPCClient::sendGET
3030072 security(ssrf): drop 'peers' from RPC read-only allowlist
ccbc1b6 security: remove vault (and getaddress/keys/runscript/megammr) from RPC read-only allowlist

v0.1.0

Full Changelog: https://github.com/Totem-Edge/minima-more/commits/v0.1.0