CMAR v1.8.1 — production hardening (owner-type, fail-closed gates)

Production hardening for real daily use.

• GitHub owner-type detection (user/organization/unknown); user accounts scoped with author: (was org: for all → mislabeled repo-commits as authored). neuron7xLab is a User → commits_authored now author-scoped.
• Token redaction of any token-like text in captured gh stderr.
• Stronger release_check: JSON-parse, GitHub fail-closed, manifest-determinism, version-drift gates.
• CI matrix Python 3.10/3.11/3.12 (live GitHub not required).
• docs/OPERATOR_RUNBOOK.md; stopped tracking generated artifacts.

63 tests, release_check PASS, CI matrix green.

🤖 Generated with Claude Code

CMAR v1.8.0 — expansion validity study + evidence-driven projector

Empirical meta-analysis of the expansion projector (studies/expansion_validity/): 4 real measured trajectories, forecast error vs the 0.15 criterion, falsification test of expansion_verdict.

Verdict: NOT falsified — direction is predictive (52/52 windows; degradation correctly DIVERGING). Magnitude (potential_mass) meets the 0.15 criterion only under local linearity (steady 0.00, degradation 0.008, noisy-OLS 0.131) and FAILS on saturating trajectories (0.41–0.83).

Evidence-driven upgrade:

• velocity by OLS over the full series (beats two-endpoint under noise: 0.131 vs 0.163);
• confidence gated on fit quality (LOW <3pts / MEDIUM R²<0.9 / HIGH R²≥0.9), not point count;
• new model, fit_quality, and nonlinearity_warning (flags unreliable magnitude on non-linear trajectories).

52 tests, release_check PASS, CI green.

Install:

pip install https://github.com/neuron7xLab/cmar/releases/download/v1.8.0/cognitive_mass_autofill_runtime-1.8.0-py3-none-any.whl

🤖 Generated with Claude Code

CMAR v1.7.0 — future-state expansion projector

PROTOCOL-02: CMAR now projects where the artifact is heading, not just where it is.

• cmar expand + src/cmar/expander.py: computes potential_mass, velocity vector, projected_states, deterministic entropy_estimate, and expansion_verdict (CONVERGING/STABLE/DIVERGING). Empty history → conservative baseline (confidence LOW); cmar ledger accumulates ledger_history.jsonl so velocity becomes measured.
• integrate embeds expansion / potential_mass / expansion_verdict.
• Falsifier invariant F11 expansion_not_diverging_on_release (RELEASE_WHILE_DIVERGING).
• release_check future-state gate (not diverging + growth projected).

50 tests, release_check PASS, CI green.

Install:

pip install https://github.com/neuron7xLab/cmar/releases/download/v1.7.0/cognitive_mass_autofill_runtime-1.7.0-py3-none-any.whl

🤖 Generated with Claude Code

CMAR v1.6.0 — cross-stream synthesis + HTTP runtime

Deploy/launch milestone: artifact moved from dev to a runtime-consumable channel.

Integrate — emergent cross_stream_synthesis: joins the repository-quality stream and the GitHub-activity stream into a convergence state (CONVERGENT_MATURE / ACTIVITY_WITHOUT_STRUCTURE / STRUCTURE_WITHOUT_ACTIVITY / IMMATURE_BOTH_STREAMS) with stream_coherence and an activity_theater_suspected finding. Descriptive only — overrides_quality: false.

Deploy — cmar serve: zero-dependency stdlib HTTP runtime (/health, /version, /runtime, /integrate, /github-activity) for external agents. Read-only, fixed root, fail-closed, never returns tokens.

Recalibration — commit_activity_ratio adapted to the heavy-tailed GitHub distribution via soft saturation (origin slope preserved, hard clamp removed).

Install:

pip install https://github.com/neuron7xLab/cmar/releases/download/v1.6.0/cognitive_mass_autofill_runtime-1.6.0-py3-none-any.whl

39 tests, release_check PASS, CI green.

🤖 Generated with Claude Code