docs: add AI safety portfolio map

Author: safal207

AI_SAFETY_PORTFOLIO.md

@@ -0,0 +1,104 @@
+# AI Safety Portfolio Map
+
+Status: reviewer-facing map for grant, fellowship, and collaboration reviewers.
+
+This file explains how the related repositories fit together. The goal is to make the portfolio look like one research program, not a scattered set of unrelated prototypes.
+
+## One-line research program
+
+Deterministic evidence, intent, and causal accountability layers for high-risk AI-agent actions before they create real-world effects.
+
+## ProofPath in the portfolio
+
+### ProofPath
+
+**Role:** verifiable intent and action-boundary audit.
+
+ProofPath focuses on whether a critical action is causally authorized and auditable at the execution boundary.
+
+```text
+valid credential != valid action != valid scope != valid reversibility != valid approval
+```
+
+ProofPath is strongest when the reviewer cares about API boundaries, payment guards, CI evidence gates, or local approval rails for coding agents.
+
+## Related layers
+
+### PythiaLabs
+
+**Role:** pre-execution evidence gates.
+
+PythiaLabs evaluates whether a proposed high-risk AI-agent action has enough evidence, authorization, context, and recovery viability to proceed.
+
+```text
+AI agent proposes action -> evidence gate -> ALLOW / BLOCK / ESCALATE
+```
+
+PythiaLabs is the cleanest grant-facing entry point for the broader portfolio.
+
+### CML — Causal Memory Layer
+
+**Role:** causal permission and responsibility lineage.
+
+CML records not only what happened, but why an action was allowed, blocked, or escalated. It is the causal accountability layer behind oversight decisions.
+
+### LTP — Liminal Thread Protocol
+
+**Role:** trace, replay, and admissibility path.
+
+LTP structures multi-step agent traces so that decisions can be replayed, compared, and audited across sessions.
+
+### LiminalQAengineer
+
+**Role:** QA reliability substrate.
+
+LiminalQAengineer applies causal and bi-temporal reasoning to CI/test workflows. It is not the main AI safety object, but it supports the engineering reliability background behind the portfolio.
+
+## Recommended reviewer paths
+
+### For AI safety / security reviewers
+
+1. Start with `docs/START_HERE_V0_1.md`.
+2. Review the CI evidence gate path.
+3. Review the Personal Agent Guard path.
+4. Review Agent Payment Guard if payment authorization is relevant.
+5. Read PythiaLabs for the broader pre-execution evidence-gate framing.
+
+### For grant reviewers
+
+Use ProofPath as evidence that the portfolio is not only conceptual. It contains concrete action-boundary demos, audit logs, CI checks, and local guard patterns.
+
+## What this portfolio is not
+
+This portfolio does not claim:
+
+- full AI alignment;
+- complete agent safety;
+- production security certification;
+- regulatory compliance;
+- replacement of human review;
+- universal prevention of unsafe actions.
+
+The current contribution is narrower:
+
+```text
+make high-risk AI-agent actions inspectable before execution.
+```
+
+## Bottom line
+
+The portfolio should be read as a layered safety stack:
+
+```text
+PythiaLabs -> evidence gate
+ProofPath -> intent and audit boundary
+CML -> causal accountability
+LTP -> trace and replay protocol
+LiminalQAengineer -> reliability substrate
+```
+
+The shared thesis:
+
+```text
+AI-agent actions should be reviewable, replayable, and evidence-backed before execution.
+```