🚀 QWED v5.0.0: Enforcement Boundary Hardening

[Rahul Dass (rahuldass19)]

Enforcement Boundary Hardening

Release Date: April 4, 2026

We're excited to announce QWED v5.0.0, a major hardening release focused on one thing: making the verification boundary fail-closed, deterministic about what it proves, and substantially harder to bypass under adversarial conditions.

This release consolidates the post-v4.0.1 hardening cycle, including the full PR 0–5 enforcement series, broad security cleanup, runtime hardening, and trust-boundary alignment.

---

🛡️ What's New in v5.0.0

Enforcement Boundary Hardening

• Removed unsafe in-process execution fallbacks on protected verification paths
• Enforced secure Docker-only execution for stats and consensus flows
• Made blocked and secure-runtime failure states explicit instead of silently degrading behavior
• Hardened Redis-backed rate limiting to fail closed on backend failure

Agent Runtime Hardening

• ActionContext is now mandatory for agent verification
• Added replay detection and repetitive loop blocking
• Added in-flight step reservations to reduce concurrent replay races
• Budget denials no longer incorrectly consume conversation state
• Agent token verification now uses constant-time hmac.compare_digest

Determinism & Trust Boundary Alignment

• Natural-language math no longer presents LLM-mediated interpretation as formally verified truth
• Added explicit trust_boundary metadata to API responses
• Symbolic identity numerical-sampling fallback now returns UNKNOWN instead of overstating equivalence
• Heuristic outcomes are now clearly labeled instead of being presented as proof

Critical Security Fixes

• Removed the raw logic-verifier eval() fallback
• Fixed consensus fact self-attestation behavior
• Prevented internal exception leakage in protected API flows
• Locked down /metrics and /metrics/prometheus to privileged authenticated access
• Enforced startup environment integrity before database initialization

Tooling & Supply Chain

• Added enforcement-oriented repository governance and review rules
• Pinned third-party GitHub Actions revisions
• Merged additional security cleanup and supply-chain hardening work across the release cycle

---

⚠️ Breaking Changes

This is a major release because it includes externally visible breaking changes:

• security_checks is no longer a client-controlled opt-in boundary for agent verification
• verify_action() integrations must now provide ActionContext with conversation_id and step_number
• /metrics and /metrics/prometheus now require authenticated privileged access
• Downstream consumers must explicitly handle INCONCLUSIVE, UNKNOWN, and BLOCKED
• Natural-language math may now return INCONCLUSIVE where earlier integrations may have assumed a stronger verification claim

---

📦 Version Updates

Package	Version
qwed (PyPI)	5.0.0
qwed_sdk (Python)	5.0.0
@qwed-ai/sdk (NPM)	5.0.0

Install / Update:

pip install qwed==5.0.0
npm install @qwed-ai/sdk@5.0.0
docker pull qwedai/qwed-verification:5.0.0

---

🧪 Validation

Release readiness was validated through merged regression suites and a final sanity sweep.

Sanity Sweep Results:

• ✅ 162 passed
• ⏭️ 11 skipped
• ❌ 0 failures

This release gate confirmed:

• No observed major fail-open or unsafe-execution regressions
• Fail-closed behavior on protected runtime paths
• Active replay and loop protection for agent flows
• Deterministic trust-boundary semantics for heuristic and translated verification results

---

🧠 Why This Release Matters

QWED is a verification boundary — not just a response layer.

v5.0.0 makes that boundary significantly stronger:

• ✅ Less silent degradation
• ✅ Less heuristic overclaiming
• ✅ Less room for bypass under failure
• ✅ More honest, machine-readable proof semantics

Core principle: Generated output is not trusted unless the boundary can actually verify it.

---

🔗 Links

• 📦 GitHub Release
• 📋 Changelog
• 🗂️ Releases Page
• 📖 Documentation

---

Thanks to everyone who contributed through code review, CI, release validation, and documentation! 🙌