Complete the privacy notice, and stop user text reaching logs

app/api/chat/route.js

@@ -33,7 +33,7 @@ import { stripSupersededFigures } from '../../../lib/countProvenance.mjs'
 import { isAggregateClassPartner } from '../../../lib/classPartners.mjs'
 import { planNextAttempt } from '../../../lib/callBudget.mjs'
 import { detectJailbreakRule } from '../../../lib/jailbreak.mjs'
-import { safeToolArgs } from '../../../lib/safeToolArgs.mjs'
+import { safeToolArgs, safeText } from '../../../lib/safeToolArgs.mjs'
 import { createRunSignal, throwIfAborted, isRunAbortedWith, NOBODY_WAITING } from '../../../lib/runSignal.mjs'
 import { parseScrnaseqClusters, parseClusterExpression, extractRequestedGenes, buildExpressionMatrix, renderExpressionMarkdown } from '../../../lib/scrnaseq.mjs'
 import { pickSeedIndividuals, parseSimilarityHits, groupSimilarByClass } from '../../../lib/similarNeurons.mjs'
@@ -1203,7 +1203,7 @@ async function callMcpToolWithRetry(client, name, args, { retries = VFB_MCP_MAX_
       // `spent` and `next` are the two facts that were missing when this was
       // diagnosed from logs alone: a stall looks exactly like a server refusing
       // three times unless the line says which budget ran out.
-      console.error(`[VFBchat] MCP CALL FAILED | tool=${name} | attempt=${attempt + 1}/${retries + 1} | transient=${transient} | spent=${elapsedMs}ms/${budgetMs}ms | next=${next.reason} | args=${safeToolArgs(args)} | error=${error?.message || error}`)
+      console.error(`[VFBchat] MCP CALL FAILED | tool=${name} | attempt=${attempt + 1}/${retries + 1} | transient=${transient} | spent=${elapsedMs}ms/${budgetMs}ms | next=${next.reason} | args=${safeToolArgs(args)} | error=${safeText(error?.message || error)}`)
       if (!next.retry) throw error
       attemptTimeoutMs = next.timeoutMs
       if (next.waitMs > 0) await new Promise(resolve => setTimeout(resolve, next.waitMs))
@@ -1277,7 +1277,7 @@ async function callMcpToolTextWithForceRefresh(client, name, args, { budget } =
     if (!isFailedRunQueryPayload(retryText)) return retryText
     return annotateFailedRunQuery(retryText)
   } catch (error) {
-    console.error(`[VFBchat] force_refresh retry for ${name} failed: ${error?.message || error}`)
+    console.error(`[VFBchat] force_refresh retry for ${name} failed: ${safeText(error?.message || error)}`)
     return annotateFailedRunQuery(text)
   }
 }

app/page.js

@@ -1031,7 +1031,7 @@ export default function Home() {
 - Raw security logs are retained for up to 30 days
 - We do not store full chat content for routine analytics
 - If you report a problem, you can optionally attach the visible chat for investigation for up to 30 days
-- Do not share confidential or sensitive information
+- Do not share personal, confidential or sensitive information
 - Use this tool to enhance your understanding of neuroscience concepts
 - See the [Privacy Notice](/privacy) for more information
 

app/privacy/page.js

@@ -19,6 +19,36 @@ export default function PrivacyPage() {
           VFB Chat is a public-facing AI-assisted interface to Virtual Fly Brain data.
         </p>
 
+        <section style={{ marginTop: '28px' }}>
+          <h2 style={{ color: '#fff' }}>Who is responsible for your data</h2>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            The <strong>University of Edinburgh</strong> is the data controller for the personal data
+            described in this notice. The University is registered with the Information
+            Commissioner&rsquo;s Office under registration number Z6426984. VFB Chat is operated by the
+            Virtual Fly Brain project in the School of Informatics.
+          </p>
+        </section>
+
+        <section style={{ marginTop: '28px' }}>
+          <h2 style={{ color: '#fff' }}>Our lawful basis</h2>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            We process this data in the performance of a <strong>task carried out in the public
+            interest</strong> (UK GDPR Article 6(1)(e)). The University&rsquo;s constitution is
+            statutory &mdash; the Universities (Scotland) Acts 1858 to 1966 &mdash; and the Further and
+            Higher Education (Scotland) Act 2005 provides for the undertaking of research among
+            fundable bodies, of which the University is one. Operating Virtual Fly Brain as an open
+            research resource, and making its data reachable, falls within those functions. We process
+            only the minimum technical data needed to run the service securely and to show how much it
+            is used; the IP address held for abuse prevention is processed on the same basis, because
+            it exists solely to keep that service available.
+          </p>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            Where you choose to attach a conversation to a problem report, we rely on your{' '}
+            <strong>consent</strong> for that transcript. You do not have to attach one, and you can
+            ask us to delete it at any time.
+          </p>
+        </section>
+
         <section style={{ marginTop: '28px' }}>
           <h2 style={{ color: '#fff' }}>What We Collect</h2>
           <ul style={{ lineHeight: 1.7 }}>
@@ -67,6 +97,12 @@ export default function PrivacyPage() {
             models. Individual prompts are not routinely read; ELM examines them only in exceptional circumstances,
             for safeguarding or legal compliance.
           </p>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            To find published papers, the service may also search PubMed (US National Library of
+            Medicine) and bioRxiv. What is sent to them is a search term built from the Virtual Fly
+            Brain terms your question resolved to &mdash; not your question, and nothing that
+            identifies you.
+          </p>
           <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
             Please still avoid typing personal information into your question. We ask this because it is good practice,
             not because we can see it.
@@ -101,9 +137,36 @@ export default function PrivacyPage() {
         <section style={{ marginTop: '28px' }}>
           <h2 style={{ color: '#fff' }}>Your Rights and Contact</h2>
           <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
-            Every answer carries a response ID, shown beneath it. We hold no account and no name for you,
-            so that ID is the only way we can find the records relating to a particular exchange. Quote it
-            when you ask what we hold about one, or ask for it to be deleted.
+            Under data protection law you have the right to:
+          </p>
+          <ul style={{ lineHeight: 1.7 }}>
+            <li><strong>Access</strong> &mdash; ask for a copy of the personal data we hold about you</li>
+            <li><strong>Rectification</strong> &mdash; ask us to correct data that is inaccurate</li>
+            <li><strong>Erasure</strong> &mdash; ask us to delete data we hold about you</li>
+            <li><strong>Restriction</strong> &mdash; ask us to limit how we use it</li>
+            <li><strong>Objection</strong> &mdash; object to our processing it</li>
+            <li><strong>Portability</strong> &mdash; where it applies, receive it in a structured, machine-readable form</li>
+            <li><strong>Withdraw consent</strong> &mdash; where we rely on consent, which for this service means a conversation you attached to a problem report</li>
+          </ul>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            <strong>How to exercise them.</strong> Every answer carries a response ID, shown beneath it.
+            We hold no account and no name for you, so that ID is the only way we can find the records
+            relating to a particular exchange &mdash; quote it when you ask what we hold, or ask for it
+            to be deleted. Because we hold nothing that identifies you, there will be cases where we
+            cannot locate any data for a request; where that happens we will say so.
+          </p>
+          <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
+            <strong>If you are unhappy with how we have handled your data</strong>, please contact the
+            University&rsquo;s Data Protection Officer using the details below. You also have the right
+            to complain to the UK supervisory authority, the{' '}
+            <a
+              href="https://ico.org.uk/make-a-complaint/"
+              target="_blank"
+              rel="noopener noreferrer"
+              style={{ color: '#66d9ff', textDecoration: 'underline' }}
+            >
+              Information Commissioner&rsquo;s Office
+            </a>, at any time.
           </p>
           <p style={{ color: '#b8b8b8', lineHeight: 1.6 }}>
             For the main Virtual Fly Brain website privacy notice and broader policy information, see{' '}

docker-compose.yml

@@ -20,6 +20,7 @@ services:
       - SEARCH_ALLOWLIST=${SEARCH_ALLOWLIST:-virtualflybrain.org,*.virtualflybrain.org,flybase.org,neurofly.org,*.neurofly.org,vfb-connect.readthedocs.io}
       - OUTBOUND_ALLOWLIST=${OUTBOUND_ALLOWLIST:-virtualflybrain.org,*.virtualflybrain.org,flybase.org,neurofly.org,*.neurofly.org,vfb-connect.readthedocs.io,doi.org,pubmed.ncbi.nlm.nih.gov,biorxiv.org,medrxiv.org}
       - REVIEWED_DOCS_INDEX_FILE=${REVIEWED_DOCS_INDEX_FILE:-/app/config/reviewed-docs-index.json}
+      - NCBI_EMAIL=${NCBI_EMAIL:-vfb@virtualflybrain.org}
       - GA_MEASUREMENT_ID=${GA_MEASUREMENT_ID}
       - GA_API_SECRET=${GA_API_SECRET}
     volumes:

lib/governance.js

@@ -289,8 +289,15 @@ export function recordAnalyticsEvent(event) {
     total_images: 0,
     total_citations: 0,
     tool_usage: {},
-    blocked_requested_domains: {},
-    blocked_response_domains: {},
+    // Counts only. The hostnames themselves are NOT kept here.
+    //
+    // A blocked "requested" domain is a hostname the user typed into a question,
+    // extracted by regex before any outbound call is made. This store is the
+    // 26-month tier, and a fragment of what someone wrote does not belong in it —
+    // the 30-day security and blocked-search logs already hold the names, which
+    // is where an abuse investigation would look anyway.
+    blocked_requested_domain_count: 0,
+    blocked_response_domain_count: 0,
     vfb_term_ids: {}
   }
 
@@ -311,21 +318,14 @@ export function recordAnalyticsEvent(event) {
     incrementMap(bucket.tool_usage, toolName, Number(count) || 0)
   }
 
-  for (const hostname of blockedRequested) {
-    incrementMap(bucket.blocked_requested_domains, hostname, 1)
-  }
-
-  for (const hostname of blockedResponse) {
-    incrementMap(bucket.blocked_response_domains, hostname, 1)
-  }
+  bucket.blocked_requested_domain_count += blockedRequested.length
+  bucket.blocked_response_domain_count += blockedResponse.length
 
   for (const termId of event.vfbTermIds || []) {
     incrementMap(bucket.vfb_term_ids, termId, 1)
   }
 
   bucket.tool_usage = normalizeCountMap(bucket.tool_usage)
-  bucket.blocked_requested_domains = normalizeCountMap(bucket.blocked_requested_domains)
-  bucket.blocked_response_domains = normalizeCountMap(bucket.blocked_response_domains)
   bucket.vfb_term_ids = normalizeCountMap(bucket.vfb_term_ids)
 
   data.buckets[bucketKey] = bucket
@@ -373,6 +373,16 @@ export async function sendStructuredTelemetry(event) {
   const gaConfig = getGoogleAnalyticsConfig()
   if (!gaConfig.enabled) return
 
+  // Everything here is generalised or drawn from VFB's own published vocabulary.
+  // Nothing a user typed reaches this payload.
+  //
+  // Two fields are worth being explicit about, because they look like content
+  // and are not. `topic_category` is a classification into one of six fixed
+  // values — anatomy, connectivity, gene expression, images, publications,
+  // how-to — so no text can travel through it. `vfb_term_ids` is matched out of
+  // the ANSWER, not the question, by /\b(?:VFB|FBbt)_\d{8}\b/: VFB and FlyBase
+  // ontology identifiers from VFB's own public data, and structurally incapable
+  // of carrying anything else.
   const payload = {
     client_id: event.requestId || createRequestId(),
     events: [{

lib/liveHarness.mjs

@@ -15,7 +15,7 @@
 // See outputs/reports/vfbchat-harness-design.md §4, §8, §9.
 
 import { runHarness } from './orchestrator.mjs'
-import { safeToolArgs } from './safeToolArgs.mjs'
+import { safeToolArgs, safeText } from './safeToolArgs.mjs'
 import { throwIfAborted, isRunAbortedWith } from './runSignal.mjs'
 
 // The largest MCP result this process will turn into a JavaScript object graph.
@@ -162,8 +162,14 @@ export function buildLiveDeps(p) {
       // shape-only: search_reviewed_docs is called with the whole user question as
       // its `query` on every turn, so JSON.stringify(args) here put the question
       // verbatim into the cluster log on any doc-search blip.
+      //
+      // The ERROR needs the same treatment, and did not have it. An upstream
+      // message quotes what it was given: get_reviewed_page takes a URL whose
+      // path is unconstrained — only the host must be allow-listed — and the
+      // fetcher throws that URL back. So a redacted `args` sat next to an
+      // unredacted `error` carrying the same value. Seen in a public CI log.
       try {
-        console.error(`[VFBchat] TOOL FAILED | tool=${name} | args=${safeToolArgs(args)} | error=${msg}`)
+        console.error(`[VFBchat] TOOL FAILED | tool=${name} | args=${safeToolArgs(args)} | error=${safeText(msg)}`)
       } catch { /* logging best-effort */ }
       return { error: msg, tool: name }
     }

lib/safeToolArgs.mjs

@@ -27,7 +27,15 @@ const SAFE_KEYS = new Set([
 // No whitespace. An id, a query type and a template name have none; a sentence
 // does. A safe KEY carrying a sentence is still a sentence — "the neuron I was
 // looking at earlier" arriving as `id` is the user's prose, not an identifier.
-const SAFE_VALUE_RE = /^[\w.,:/-]{0,80}$/
+//
+// Absence of whitespace was the ONLY test here, and it is not enough: a URL, a
+// filesystem path, or a 70-character lab-internal string all pass it, and all of
+// them are something the user typed. So the rule is identifier-SHAPED rather
+// than merely space-free — no `/` or `:`, which is what excludes URLs and paths,
+// and 40 characters rather than 80, comfortably longer than any VFB id, query
+// type, template or gene symbol and shorter than a phrase worth logging.
+// Anything outside that is rendered as its length, exactly as before.
+const SAFE_VALUE_RE = /^[\w.,+-]{0,40}$/
 
 /**
  * A free-text value rendered for a container log: its length, not its content.