Releases: madgodinc/mgi-mind
Release list
v2.5.0
v2.4.0
v2.4.0 — multi-tenant confinement gate closed (Д7) + tamper-evident a…
v2.3.0
v2.3.0 — multi-hop fact-graph traversal + local profile snapshot
v1.7.0
v1.7.0 — /kv blob store, /audit Track-3 trail, reindex-on-model-change
Media assets (brain graph art)
Heavy media kept out of the git tree to keep clones light. brain-demo.gif is the animated brain-graph visualization shown in the README. Stable download URL for inline README embedding.
v1.6.4 — Windows fix + contributor docs
Closes #23 (Windows stack overflow) and ships the developer-facing scaffolding.
Issue #23 — Windows stack overflow
tokio::main runs block_on on the process's main thread, which uses the OS default stack budget. Windows defaults to 1 MB; the v1.5 background re-test loop's futures overflow it.
Fix: re-launch main on std::thread::Builder with 8 MB stack + tokio runtime with thread_stack_size(8 * 1024 * 1024).
All six CI jobs go green:
- ✅ Linux fmt/clippy/test
- ✅ Windows fmt/clippy/test
- ✅ macOS fmt/clippy/test
- ✅ Linux integration (real Qdrant)
- ✅ Windows integration (bundled Qdrant)
- ✅ cargo-audit
mgimind doctor detects pending cardinality proposals
After extraction lands hundreds of predicate cardinality proposals, the user had to remember to register them. doctor now surfaces the count + --fix bulk-registers via the same path migrate-v14 cardinality --apply uses.
Project documentation
- CONTRIBUTING.md — project layout, three architecture rules, branch model.
- CODE_OF_CONDUCT.md — Contributor Covenant 2.1 compression.
- SECURITY.md — vulnerability disclosure policy.
- .github/ISSUE_TEMPLATE/ — bug + feature + Discussions routing.
- docs/design/adr/ — four ADRs (Cardinality enum, Mechanism 1 invariant, §10 q5 guarantees, install-mode anchors).
- benchmarks/v0.14.3-gpu/ — R@5 = 99.2% baseline committed for STALE comparison.
- docs/blog/2026-06-04-validity-model.md — draft technical post.
- .editorconfig — consistent indent / EOL.
Tests
290 unit + 6 integration tests pass on every platform (Linux, Windows, macOS).
Migration
None. Only Windows fix + scaffolding; no semantic changes.
v1.6.3 — bench-stale CLI + contributor docs
Evening polish on top of v1.6.2.
mgimind migrate-v14 cardinality --apply— bulk-register High-confidence proposals (Mad's base mid-extraction shows 1096 high / 17 low across 1113 predicates).mgimind bench-stale+mgimind bench-stale-sweep— CLI scaffold for issue #16 calibration tooling.- CONTRIBUTING.md, CODE_OF_CONDUCT.md, .github/ISSUE_TEMPLATE/ — first-time contributor surface.
290 unit + 6 integration tests, 0 failed.
v1.6.2 — facts inspection CLI
Two CLI additions on top of v1.6.1.
mgimind stats --json— machine-parseable variant for monitoring scripts.mgimind facts list/show— first-class KG inspection from the terminal.
290 unit + 6 integration tests, 0 failed.
v1.6.1 — CLI surfaces
Polish release. Four CLI fixes that expose v1.5 / v1.6 surfaces from the terminal.
- `mgimind outcome <signal_type>` — terminal access to mind_outcome.
- `mgimind audit list --op X --since-hours N` — filter audit by op type and time.
- `mgimind doctor` + `mgimind config install-mode` — show install-mode weights.
- `mgimind stats` — fact graph distribution + in-doubt + inherited counts.
No new MCP tools, no schema changes, no formula changes.
290 unit + 6 integration tests, 0 failed.
v1.6.0
A polish release that closes three TBD items declared in v1.5's "Honest limits" section. No new public surface — same 37 MCP tools, same CLI flags, same install modes. Smaller per-tick overhead and sharper coverage of the §10 q5 guarantees.
Step 1 — batched payload reads in retest_fact_step82
v1.5 made four separate `get_points` round-trips per fact. At BACKGROUND_PER_TICK_CAP=50 that was 200 round-trips per tick.
New `storage::read_point_payload_strings(client, collection, id, &keys)` returns a HashMap in one call. `retest_fact_step82` now does one Qdrant fetch + four HashMap lookups. 4× reduction in round-trips per fact, same semantics.
Step 2 — `cited_by` chain following
The v1.5 self-citation guard always blocked because the closure was `|_| None`.
v1.6 wires real lookup via `fetch_citing_confidences` — a single batched `get_points` against MEMORIES_COLLECTION. Closure reads from the pre-fetched map synchronously.
- Citing memory with cached confidence ≥ 0.5 → `cited_by` signal contributes `0.2 × external slot weight`.
- Citing memory below 0.5 or missing → guard blocks.
Mechanism 1 invariant preserved.
Step 3 — integration tests on `spawn_background_retest_loop`
Closes the audit-flagged gap: v1.5 had no test that ran the spawned task end-to-end. Four new tests at the registry / scheduling level.
Build hygiene
Six modules gain module-level `#![allow(dead_code)]` with comments. Build output is warning-free.
Tests
290 unit + 6 integration tests, 0 failed. 4 new tests for v1.6.