v1.2.0

v1.2.0 — the first OMEGA-era release

m1nd = operational intelligence for coding agents. 1.2.0 turns the loop from "retrieve, then hope" into pre-orient → act on calibrated verdicts → capture what you learned. Same principle as the trust layer: an honest no beats a confident guess.

m1nd-core / m1nd-ingest / m1nd-mcp → 1.2.0 (crates.io + @maxkle1nz/m1nd on npm). m1nd-openclaw stays 0.1.0 (versioned independently).

Highlights

• north(task) — pre-orient in one call. Composes binding trust, task context (focus nodes + PageRank anchors), prior cross-session memory (with real age + author), a sufficiency signal, one next_move, and honest_gaps. needs_ingest is a real answer for an empty graph.
• Conformal calibration on prediction. calibrate_predict arms a per-repo gate; verdicts then read act / reverify / abstain, where abstain = uncalibrated or insufficient. Ships dark: until calibrated, verdicts cap at reverify. (First real number on m1nd's own history: τ=0.60, coverage 14.6%, act-band precision 28.3% — the calibrator's job was to surface that the strength model is coarse, and it did.)
• trust_envelope on seek (ships dark) and a closure verdict on why (blocked = the path rests on an unresolved edge). trust_band: insufficient_evidence now means no evidence — the honest cold-start answer, distinct from a risk band.
• Memory grew a provenance spine — claims carry real age + author, supersede older claims, age out, and respect a recency cap, so remembered knowledge states its own freshness.
• Smoothed-Jaccard co-change — ghost_edges / predict normalize coupling instead of counting raw co-commits (calibration-proven +3 points over raw counts).
• Binary version + sha fingerprint — --version prints 1.2.0 (<sha>); M1ND_EXPECTED_VERSION / M1ND_EXPECTED_SHA (+ M1ND_STRICT_VERSION) detect and refuse a drifted binary.
• Agent-native MCP instructions + local-only field reports. The initialize instructions every host receives now are the operating loop. Agents can leave one telemetry signal — learn on a retrieval verdict, or a line in ~/.m1nd/field-reports.jsonl when m1nd itself misbehaves. That file is local-only; m1nd never phones home.

Honest notes

• The calibration gate ships dark — no verb earns act until calibrate_predict certifies its precision-at-coverage against a stated risk budget. predict's strength model is still coarse (tops ~28% act-band precision on m1nd's own history); calibration's job is to surface that, not hide it.
• trust_envelope (seek) and the closure verdict ship behind their signals; the poisoned-oracle threat model (who calibrates the calibrator?) stays open, named in the PRD, not waved away.

Install: npm install -g @maxkle1nz/m1nd · m1nd doctor. Full setup: docs/AGENT-PACKS.md.

---

What's Changed

• docs(pathos): checkpoint 3 — v1.1.0 (focus + full resolution arc, battery 28/28) by @maxkle1nz in #177
• fix(mcp): sweep dead instance leases on server boot (non-blocking) by @maxkle1nz in #178
• perf(mcp): read the OS process table once per GC sweep (sysinfo) by @maxkle1nz in #181
• docs: Next-Gen Agent-Native PRD (deep-research synthesis) by @maxkle1nz in #179
• docs(pathos): point the north at the Next-Gen Agent-Native PRD by @maxkle1nz in #180
• fix(trust): replace cold-start 0.5/Unknown verdict with an action-routable band by @maxkle1nz in #182
• chore(battery): track the canonical capability test harness in-repo by @maxkle1nz in #183
• docs: Subsystem D — Agent-Native Memory roadmap by @maxkle1nz in #184
• feat(why): graph-closure verdict — flag answers that rest on an unresolved edge by @maxkle1nz in #185
• docs(pathos): checkpoint 4 — moves #1-#3 shipped, battery re-armed, memory frontier (Subsystem D) active by @maxkle1nz in #186
• feat(memory): stamp Created + Source-Agent on every memorized claim by @maxkle1nz in #187
• chore(deps): bump @babel/core to 7.29.7 (fixes low-sev Dependabot alert) by @maxkle1nz in #188
• feat(memory): surface authored-age + source-agent on recall (honest memory labeling) by @maxkle1nz in #189
• refactor(mcp): collapse 11 duplicate now_ms helpers into one shared util::now_ms by @maxkle1nz in #190
• docs: m1nd-OMEGA — the v1.2→v2.0 era (vision + critic-corrected roadmap) by @maxkle1nz in #191
• feat(omega): Move 0 — conformal calibration harness gating predict with act|reverify|abstain by @maxkle1nz in #192
• docs(pathos): checkpoint 5 — OMEGA era live, Move 0 calibration shipped (real number), tie analysis by @maxkle1nz in #193
• fix(ingest): respect .gitignore (skip vendored/build dirs) — fixes acc timeout + l00p 100x over-ingest by @maxkle1nz in #194
• feat(omega): Move 1 — trust-gated answer envelope on seek (calibrated weighting, ships dark) by @maxkle1nz in #195
• docs: Ω+1 ambient-loop design (critic-corrected) + PATHOS checkpoint 6 (Move 1 shipped) by @maxkle1nz in #196
• feat(ambient): north-packet pre-orient verb (compose binding+context+memory in one call) by @maxkle1nz in #197
• feat(memory): age-staleness signal (aged_out) orthogonal to evidence-freshness by @maxkle1nz in #198
• fix(activation): activate_temporal reads DomainConfig per-type half-lives (dead table now live) by @maxkle1nz in #199
• feat(memory): supersession-on-rewrite — invalidate-and-keep with per-slug flock by @maxkle1nz in #200
• feat(memory): recency-cap the agent-memory auto-load (default unlimited; missing-Created exempt) by @maxkle1nz in #201
• docs(pathos): checkpoint 7 — memory #3-#6 shipped + pre-flight A/B (helps orient, compounding needs serve/attach) by @maxkle1nz in #202
• docs(omega): reframe Move 2 — drop token-economics lie, honest stop_gate by @maxkle1nz in #203
• chore(battery): cover the OMEGA/memory surface (closure, trust_band, calibration, envelope, north, provenance/supersession/aged_out) by @maxkle1nz in #204
• feat(mcp): binary version+sha fingerprint with honest drift warning by @maxkle1nz in #205
• feat(core): normalized co-change coupling (smoothed Jaccard) — calibration-proven over raw counts by @maxkle1nz in #206
• test(memory): de-flake supersession concurrency test (race the memorize, not session init) by @maxkle1nz in #207
• feat(mcp): agent-native server instructions — the OMEGA doctrine by @maxkle1nz in #208
• release: v1.2.0 — the first OMEGA-era release by @maxkle1nz in #209

Full Changelog: v1.1.0...v1.2.0