From 53cc47224745478f4c67776cd34e1d2ce16896af Mon Sep 17 00:00:00 2001 From: TheFlow Date: Fri, 13 Mar 2026 15:27:10 +1300 Subject: [PATCH] feat: add Guardian Agents to village-case-study.html - Add Guardian Agents verification as step 6 in governance flow (8 steps, was 7) - Add Guardian Agents featured card in features section - Update deployment facts: 5 governed features (was 4) - Add verification evidence to "What This Demonstrates" - Add Guardian Agents Philosophy paper link to CTA section - Update EN locale to match Co-Authored-By: Claude Opus 4.6 --- public/locales/en/village-case-study.json | 23 +++++++--- public/village-case-study.html | 51 +++++++++++++++++++---- 2 files changed, 62 insertions(+), 12 deletions(-) diff --git a/public/locales/en/village-case-study.json b/public/locales/en/village-case-study.json index f91e1892..1473dc5c 100644 --- a/public/locales/en/village-case-study.json +++ b/public/locales/en/village-case-study.json @@ -22,7 +22,7 @@ "ai_label": "AI Model:", "ai_value": "Sovereign Llama (QLoRA fine-tuned)", "features_label": "AI Features:", - "features_value": "4 governed features live", + "features_value": "5 governed features live", "infra_label": "Infrastructure:", "infra_value": "NZ + EU (no US dependency)" } @@ -53,7 +53,7 @@ }, "flow": { "heading": "How Governance Works in Practice", - "intro": "When a member uses any AI feature, the request passes through six governance checks before a response is generated. Each check is independent and can block or modify the request.", + "intro": "When a member uses any AI feature, the request passes through seven governance checks before a response reaches them. Each check is independent and can block or modify the request.", "steps": { "1": { "title": "Member request received", @@ -76,12 +76,16 @@ "desc": "The sovereign Llama model generates a response using RAG context filtered by the member's permissions. All processing stays on-infrastructure." }, "6": { + "title": "Guardian Agents verification", + "desc": "Every factual claim in the response is checked against source material using embedding cosine similarity — mathematical measurement, not generative checking. Each claim receives a confidence badge visible to the member." + }, + "7": { "title": "Scope verification", "desc": "Is the response appropriate to what was asked? Detects scope creep and blocks responses that exceed the original request." }, - "7": { + "8": { "title": "Delivery with attribution", - "desc": "Response delivered to the member with source attribution. Every step is logged for audit." + "desc": "Response delivered to the member with confidence badges, source attribution, and full audit trail. Every step is logged." } } }, @@ -106,6 +110,13 @@ "title": "AI Memory Transparency", "benefit": "Members can see, edit, and delete what the AI \"remembers\" about them. Full audit dashboard shows every AI interaction.", "governance": "Governance: Multi-stakeholder consent required. Persistence decisions classified and auditable. Members control their own data." + }, + "guardian": { + "title": "Guardian Agents", + "badge": "NEW", + "benefit": "Every AI response is verified against source material using embedding cosine similarity — mathematical measurement, not another generative model checking the first. Each factual claim gets a confidence badge (high, medium, low, unverified) visible to the member.", + "governance": "Governance: Verification operates in a fundamentally different epistemic domain from the generation layer — avoiding common-mode failure. Moderator corrections feed back into verification thresholds. All tenant-scoped.", + "link": "Philosophical foundations →" } }, "limitations": { @@ -143,7 +154,8 @@ "• Polycentric governance can operate in production without prohibitive overhead", "• Multi-tenant isolation with per-community governance is achievable", "• Governance violations are detectable and auditable", - "• The framework learns from failures (documented incident responses)" + "• The framework learns from failures (documented incident responses)", + "• Verification without common-mode failure is achievable using mathematical similarity rather than generative checking" ] }, "does_not_support": { @@ -162,6 +174,7 @@ "visit_village": "Visit the Village →", "village_ai": "Sovereign Language Model →", "research_paper": "Research Paper →", + "guardian_paper": "Guardian Agents Philosophy →", "research_details": "Research Details →" } } diff --git a/public/village-case-study.html b/public/village-case-study.html index a5a288e0..ce61badc 100644 --- a/public/village-case-study.html +++ b/public/village-case-study.html @@ -127,7 +127,7 @@
  • AI Features: - 4 governed features live + 5 governed features live
  • Infrastructure: @@ -207,7 +207,7 @@

    How Governance Works in Practice

    - When a member uses any AI feature, the request passes through six governance checks before a response is generated. Each check is independent and can block or modify the request. + When a member uses any AI feature, the request passes through seven governance checks before a response reaches them. Each check is independent and can block or modify the request.

    @@ -250,15 +250,22 @@
  • 6
    -

    Scope verification

    -

    Is the response appropriate to what was asked? Detects scope creep and blocks responses that exceed the original request.

    +

    Guardian Agents verification

    +

    Every factual claim in the response is checked against source material using embedding cosine similarity — mathematical measurement, not generative checking. Each claim receives a confidence badge visible to the member.

  • -
    7
    +
    7
    -

    Delivery with attribution

    -

    Response delivered to the member with source attribution. Every step is logged for audit.

    +

    Scope verification

    +

    Is the response appropriate to what was asked? Detects scope creep and blocks responses that exceed the original request.

    +
    +
    • +
  • +
    8
    +
    +

    Delivery with attribution

    +

    Response delivered to the member with confidence badges, source attribution, and full audit trail. Every step is logged.

    • @@ -342,6 +349,30 @@

    + + +
    +
    +
    + + + +
    +
    +
    +

    Guardian Agents

    + NEW +
    +

    + Every AI response is verified against source material using embedding cosine similarity — mathematical measurement, not another generative model checking the first. Each factual claim gets a confidence badge (high, medium, low, unverified) visible to the member. +

    +

    + Governance: Verification operates in a fundamentally different epistemic domain from the generation layer — avoiding common-mode failure. Moderator corrections feed back into verification thresholds. All tenant-scoped. +

    + Philosophical foundations → +
    +
    +
    @@ -416,6 +447,7 @@
  • • Multi-tenant isolation with per-community governance is achievable
  • • Governance violations are detectable and auditable
  • • The framework learns from failures (documented incident responses)
    • +
  • • Verification without common-mode failure is achievable using mathematical similarity rather than generative checking
    • @@ -455,6 +487,11 @@ data-i18n="cta.research_paper"> Research Paper → + + Guardian Agents Philosophy → +