[Audit] Re: Charter Comments -- more specific on the scope?

"Heime.Jorgen" <heime.jorgen@proton.me> Mon, 12 October 2026 15:49 UTC

Received: from mail-10698.protonmail.ch (mail-10698.protonmail.ch [79.135.106.98]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519 server-signature ECDSA (prime256v1) server-digest SHA256) (No client certificate requested) by mx.ietf.org (Postfix) with ESMTPS id 2594846 for <audit@ietf.org>; Mon, 12 Oct 2026 15:49:22 +0000 (UTC)
Authentication-Results: mx.ietf.org; dkim=pass header.d=proton.me header.s=protonmail2 header.b=vXyGQJCp; spf=pass (mx.ietf.org: domain of heime.jorgen@proton.me designates 79.135.106.98 as permitted sender) smtp.mailfrom=heime.jorgen@proton.me; dmarc=pass (policy=quarantine) header.from=proton.me
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=proton.me; s=protonmail2; t=1791820155; x=1792079355; bh=BOrzRZkK7L1RC/De8gjgPswnzmAvg4xc42wJT8SF6sk=; h=Date:To:From:Subject:Message-ID:In-Reply-To:References: Feedback-ID:From:To:Cc:Date:Subject:Reply-To:Feedback-ID: Message-ID:BIMI-Selector; b=vXyGQJCpKUQZWKKew5T6Ue4j6Oqx1Ll4iRtNpyVqes4gvHz3885lEUJewUE6hg2ri RUFexmYKHwGYGy8Vk6suBQRPCZP+kwNHG3RwQc6ppFp/1AkH4yujOMfYR3GrJgVfkJ ceXxrA5wyL+WfROzTYuRKm5qnTxQBludXrh7exkGVKl+febEwLsEPPFsMDgdK9bs91 Y5DWB5NwBsLYLYnMyV5ALgmdSSRYEHNCnnAIvEhCTt5Vac89qq+cM1cDkVIabx96g/ CJPwHLQN6dzRS/L0ioVR0i7JN0FaFkkKFVLYwx8z1fuP2fY002Q2XfGOZ+Djo7vxSB ZU7sxqXmUg+GA==
Date: Mon, 12 Oct 2026 15:49:10 +0000
To: audit <audit@ietf.org>
From: "Heime.Jorgen" <heime.jorgen@proton.me>
Message-ID: <umTeji61noCNHB77NTY134jTglHBPqjNaFew1Td-6-HyPR86NHTo3fWfLzMncGQf6NweVzzVjZ0oWBTw1ELIyBZH-4omExTW7Z_UoZKZJ1Q=@proton.me>
In-Reply-To: <Im9tMT3y8YRLhDDm3kFOx20BFMxeMyyqgwa3tDFUGoh7T611pfJJyN8Bf1zBJwIkiqEpVyLNNjlhhfYyE7hmb4yVLVclFse_CHMECO5gA-w=@vaara.io>
References: <Im9tMT3y8YRLhDDm3kFOx20BFMxeMyyqgwa3tDFUGoh7T611pfJJyN8Bf1zBJwIkiqEpVyLNNjlhhfYyE7hmb4yVLVclFse_CHMECO5gA-w=@vaara.io>
Feedback-ID: 179960286:user:proton
X-Pm-Message-ID: 64feba4dc1b77d1529a9d07658f466e984aa3021
MIME-Version: 1.0
Content-Type: multipart/alternative; boundary="b1=_1Z237OPJzLLLOxSGrfxHiUaI87hUdgkK9s1K9XBz3cw"
X-Spamd-Bar: -----------------
Message-ID-Hash: SA7LRGY3I6F5SSS7ZZ6WKWCPDAWOIYBP
X-Message-ID-Hash: SA7LRGY3I6F5SSS7ZZ6WKWCPDAWOIYBP
X-MailFrom: heime.jorgen@proton.me
X-Mailman-Rule-Misses: dmarc-mitigation; no-senders; approved; loop; banned-address; emergency; member-moderation; nonmember-moderation; administrivia; implicit-dest; max-recipients; max-size; news-moderation; no-subject; digests; suspicious-header
X-Mailman-Version: 3.3.10
Precedence: list
Subject: [Audit] Re: Charter Comments -- more specific on the scope?
List-Id: "Agent Use of Delegation and Interaction Traceability (AUDIT)" <audit.ietf.org>
Archived-At: <https://mailarchive.ietf.org/arch/msg/audit/cc1wLGMgsKomrtifgb8_w1ICGLc>
List-Archive: <https://mailarchive.ietf.org/arch/browse/audit>
List-Help: <mailto:audit-request@ietf.org?subject=help>
List-Owner: <mailto:audit-owner@ietf.org>
List-Post: <mailto:audit@ietf.org>
List-Subscribe: <mailto:audit-join@ietf.org>
List-Unsubscribe: <mailto:audit-leave@ietf.org>

Hi Henri, Sumit, Lars, Iman,​

Thanks for sharing the trail_decision_v0 test vectors. Separating an explicit evaluation failure from a missing input (`not_evaluated`) is a clean approach to handling runtime telemetry exceptions.

However, from an agent safety and formal verification standpoint, there is a fundamental distinction between dynamic runtime assertions and kernel-checked invariant proofs:

1. Runtime Exception Handling vs. Mathematical Invariants

Categorizing a refusal as `not_evaluated` simply flags that a dynamic script or rule couldn't execute due to missing telemetry. While helpful for debugging, dynamic JSON checks cannot guarantee safety across unobserved edge cases. In contrast, K-Lean formalizes agent policies into Lean 4 propositions (P ⟹ Q). The Lean kernel mathematically proves that policy invariants hold across the entire state space before any action is taken.

2. Trusted Computing Base (TCB) & Parser Drift

Relying on custom JSON schemas or dynamic engines expands the TCB meaning any bug, parser drift, or unhandled case in the evaluation logic invalidates the verdict. K-Lean reduces the TCB strictly to the minimal Lean 4 microkernel (`kernel-standard`), eliminating software bug risks in the policy engine itself.

3. Post-hoc Audit Trail vs. Pre-effect Execution Gate

A refusal block provides a useful audit trail explaining why a decision was denied. But for autonomous agents, post-hoc logging is insufficient. K-Lean binds kernel-verified manifests directly into 180-octet TTTPS cryptographic receipts to form a strict Pre-effect Admission Gate making unverified or policy-violating actions physically unexecutable.

Explicitly tracking missing inputs is a solid step for dynamic logging, but ensuring true agent safety requires proving that policy invariants hold at the kernel level before any side-effect hits production.

Heime
-------- Original Message --------
On Monday, 10/12/26 at 21:56 Henri Sirkkavaara <hello@vaara.io> wrote:

> Sumit, Lars, Iman,
> Vaara's decision receipt carries Case 9 as fields. A deny has a refusal block whose basis is evaluated, not_evaluated or undetermined, and with not_evaluated it names the unavailable input. The refusal stays in the decision field, so the decision and its grounds are separate, as Iman framed it. Vectors with an evaluated and a not_evaluated deny: https://github.com/vaaraio/vaara/tree/main/tests/vectors/trail_decision_v0
>
> Henri Sirkkavaara
> Vaara - Accountable Autonomy
> Built to see over the noise.
> vaara.ioHelsinki, Finland