diff --git a/CHANTIERS_RESTANTS_EXECUTION_PLAN.md b/CHANTIERS_RESTANTS_EXECUTION_PLAN.md
new file mode 100644
index 0000000..8b0a089
--- /dev/null
+++ b/CHANTIERS_RESTANTS_EXECUTION_PLAN.md
@@ -0,0 +1,123 @@
+# Plan d'execution des chantiers restants (zero regression)
+
+Date: 2026-03-09  
+Scope: ETHICA, Tracking, Factory SaaS, Multi-install WEVADS/ADX
+
+## 1) Objectif de gouvernance
+
+Mettre en place un dispositif **durable** qui evite les regressions et permet de traiter les chantiers restants en sequence controlee:
+
+- Gate GO/NO-GO automatise (script `nonreg-framework.sh`)
+- Preflight multi-install sans risque (script `multiinstall-safe-preflight.sh`)
+- Validation de qualite par lot (batch) avant extension de perimetre
+
+---
+
+## 2) Priorites P0 (immediat)
+
+### P0.1 - Stabilite multi-install (sans toucher PMTA/SSH/global tuning)
+
+Definition of Done:
+- 100% des serveurs d'un batch passent le preflight
+- 0 serveur lance avec lock dpkg actif
+- 0 echec "undefined / nothing in process" sur batch valide
+
+Procedure:
+1. Construire `servers.csv` (id, ip, username, password)
+2. Executer `./multiinstall-safe-preflight.sh servers.csv`
+3. Lancer uniquement les serveurs `ready=YES`
+4. Ne jamais depasser la taille de batch validee (ex: 3-5)
+
+### P0.2 - Gate anti-regression avant toute release
+
+Definition of Done:
+- 0 FAIL sur rapport anti-regression
+- WEVIA greeting < 3s
+- WEVIA deep < 60s
+- Pages critiques en HTTP 200
+- Scan confidentialite propre
+
+Procedure:
+1. `chmod +x nonreg-framework.sh`
+2. `./nonreg-framework.sh`
+3. Si FAIL > 0 => NO-GO
+
+---
+
+## 3) Priorites P1 (fiabilisation metier)
+
+### P1.1 - ETHICA
+
+Objectifs:
+- fiabiliser les sources (fallback MarocMedecin, Tabibi listing-based)
+- conserver cadence crons sans grossissement de logs
+
+Checklist:
+- [ ] Source alternative active si Cloudflare bloque
+- [ ] Tabibi passe en mode listing (plus de dependance ID-only)
+- [ ] logrotate actif sur logs scrapers
+- [ ] KPI min: croissance `medecins_real` + taux de validation stable
+
+### P1.2 - Tracking S3 + endpoints
+
+Objectifs:
+- assurer coherence tracking URL partout
+- eliminer drift entre S3 redirect / configs locales / BDD
+
+Checklist:
+- [ ] redirect.html S3 aligne sur tracking actuel
+- [ ] configs app (wevads/fmg) alignees
+- [ ] domaine tracking resolu et accessible (200/301/302)
+
+### P1.3 - Factory SaaS
+
+Objectifs:
+- smoke tests API avant publication
+- distinction claire: app LIVE vs landing-only
+
+Checklist:
+- [ ] endpoints smoke verifies (DeliverScore/MedReach/GPU/Content)
+- [ ] mapping modeles GPU aligne UI/backend
+- [ ] status public documente par produit (LIVE/BETA/LANDING)
+
+---
+
+## 4) SLO / Six Sigma operable (pragmatique)
+
+Metriques cibles:
+- Disponibilite checks critiques: >= 99.5% (sur 7 jours glissants)
+- Regressions bloquantes post-release: 0
+- Echecs batch multi-install: < 5%
+- MTTR incident critique: < 30 min
+
+Cadence:
+- Daily: run anti-regression
+- Avant deploy: gate obligatoire
+- Apres incident: post-mortem + ajout test de non-regression associe
+
+---
+
+## 5) Commandes utiles
+
+```bash
+# 1) Gate anti-regression
+chmod +x nonreg-framework.sh
+./nonreg-framework.sh
+
+# 2) Gate anti-regression avec API key (GPU check actif)
+API_KEY="xxx" GPU_MODEL="qwen2.5:3b" ./nonreg-framework.sh
+
+# 3) Preflight multi-install
+chmod +x multiinstall-safe-preflight.sh
+./multiinstall-safe-preflight.sh servers.csv
+```
+
+---
+
+## 6) Regles anti-incident (obligatoires)
+
+1. Ne pas modifier PMTA/JAR/SSH global pour accelerer un batch.
+2. Toujours preflight avant lancement.
+3. Pas de nouveau batch tant que le precedent n'est pas stable.
+4. Toute correction manuelle en prod => ajouter un test dans le framework.
+
diff --git a/RAPPORT_CHANTIERS_FINAL.md b/RAPPORT_CHANTIERS_FINAL.md
new file mode 100644
index 0000000..2c01964
--- /dev/null
+++ b/RAPPORT_CHANTIERS_FINAL.md
@@ -0,0 +1,60 @@
+# Rapport final - chantiers restants (execution Codex)
+
+Date: 2026-03-09  
+Branche: `cursor/ethica-saas-chantiers-a789`
+
+## 1) Ce qui a ete livre
+
+1. **Framework anti-regression**: `nonreg-framework.sh`
+   - checks pages critiques (site + produits)
+   - checks WEVIA (greeting/deep avec seuils)
+   - checks APIs SaaS (DeliverScore/MedReach/GPU)
+   - checks tracking (IP + domaine)
+   - scan confidentialite (mode strict optionnel)
+   - sortie rapport markdown horodate dans `reports/`
+
+2. **Preflight multi-install safe**: `multiinstall-safe-preflight.sh`
+   - aucun changement PMTA/SSH/global config
+   - verifie reachability, auth, disque, RAM, lock dpkg, sante apt
+   - sort un CSV de readiness batch avant lancement multi-install
+
+3. **Plan d'execution chantiers**: `CHANTIERS_RESTANTS_EXECUTION_PLAN.md`
+   - priorites P0/P1
+   - gates GO/NO-GO
+   - regles anti-incident
+   - metriques d'operation (SLO)
+
+4. **Template d'entree multi-install**: `servers.example.csv`
+
+---
+
+## 2) Resultat du run live
+
+Run effectue: `./nonreg-framework.sh`
+
+- PASS: 24
+- WARN: 4
+- FAIL: 0
+- Verdict: **GO**
+
+Warnings detectes:
+- termes sensibles encore presents sur:
+  - `/products/wevads-ia.html`
+  - `/products/workspace.html`
+- DeliverScore en rate-limit (429) sans cle API
+- test GPU saute (API key non fournie)
+
+> Pour mode blocant total, lancer:
+>
+> `STRICT_CONFIDENTIALITY=1 API_KEY="xxx" ./nonreg-framework.sh`
+
+---
+
+## 3) Cadre no-regression active
+
+Le dispositif est maintenant en place:
+
+- **Avant chaque release**: run anti-regression obligatoire
+- **Avant chaque batch multi-install**: run preflight obligatoire
+- **Si FAIL > 0**: NO-GO automatique
+
diff --git a/README.md b/README.md
index 4b54aba..1aae73a 100644
--- a/README.md
+++ b/README.md
@@ -4,4 +4,10 @@
 - **RAM**: 62GB DDR4
 - **Disk**: 1.7TB NVMe
 - **Ollama**: localhost:11434
-- **Models**: deepseek-r1:8b, deepseek-r1:32b, llama3.1:8b
+- **Legacy local models**: deepseek-r1:8b, deepseek-r1:32b, llama3.1:8b
+
+## Ops scripts in this repo
+
+- `nonreg-framework.sh`: anti-regression gate (HTTP/API/WEVIA/tracking/confidentiality checks)
+- `multiinstall-safe-preflight.sh`: safe server preflight before multi-install batches
+- `CHANTIERS_RESTANTS_EXECUTION_PLAN.md`: execution plan and GO/NO-GO criteria
diff --git a/multiinstall-safe-preflight.sh b/multiinstall-safe-preflight.sh
new file mode 100755
index 0000000..e4457a6
--- /dev/null
+++ b/multiinstall-safe-preflight.sh
@@ -0,0 +1,138 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# -------------------------------------------------------------------
+# Multi-install SAFE preflight
+# Goal: reduce failed batches without touching PMTA/SSH/global config.
+#
+# Input file format (CSV-like, no header):
+#   server_id,ip,username,password
+# Example:
+#   180,101.46.69.207,root,Yacine.123
+# -------------------------------------------------------------------
+
+INPUT_FILE="${1:-}"
+CONNECT_TIMEOUT="${CONNECT_TIMEOUT:-5}"
+SSH_BIN="${SSH_BIN:-ssh}"
+SSHPASS_BIN="${SSHPASS_BIN:-sshpass}"
+OUT_DIR="${OUT_DIR:-./reports}"
+RUN_ID="$(date +%Y%m%d_%H%M%S)"
+OUT_CSV="${OUT_DIR}/multiinstall_preflight_${RUN_ID}.csv"
+
+if [[ -z "${INPUT_FILE}" || ! -f "${INPUT_FILE}" ]]; then
+  echo "Usage: $0 <servers.csv>"
+  echo "Missing input file: ${INPUT_FILE:-<empty>}"
+  exit 1
+fi
+
+mkdir -p "${OUT_DIR}"
+echo "server_id,ip,ssh_tcp,ssh_auth,disk_ok,ram_ok,dpkg_lock,apt_health,ready,notes" > "${OUT_CSV}"
+
+check_tcp_22() {
+  local ip="$1"
+  timeout "${CONNECT_TIMEOUT}" bash -c "exec 3<>/dev/tcp/${ip}/22" >/dev/null 2>&1
+}
+
+run_ssh_password() {
+  local user="$1" ip="$2" pass="$3" cmd="$4"
+  "${SSHPASS_BIN}" -p "${pass}" "${SSH_BIN}" \
+    -o StrictHostKeyChecking=no \
+    -o UserKnownHostsFile=/dev/null \
+    -o ConnectTimeout="${CONNECT_TIMEOUT}" \
+    "${user}@${ip}" "${cmd}"
+}
+
+run_ssh_key() {
+  local user="$1" ip="$2" cmd="$3"
+  "${SSH_BIN}" \
+    -o StrictHostKeyChecking=no \
+    -o UserKnownHostsFile=/dev/null \
+    -o ConnectTimeout="${CONNECT_TIMEOUT}" \
+    "${user}@${ip}" "${cmd}"
+}
+
+HAVE_SSHPASS=0
+if command -v "${SSHPASS_BIN}" >/dev/null 2>&1; then
+  HAVE_SSHPASS=1
+fi
+
+while IFS=',' read -r server_id ip username password; do
+  [[ -z "${server_id}" ]] && continue
+  [[ "${server_id}" =~ ^# ]] && continue
+
+  ssh_tcp="FAIL"
+  ssh_auth="FAIL"
+  disk_ok="FAIL"
+  ram_ok="FAIL"
+  dpkg_lock="UNKNOWN"
+  apt_health="UNKNOWN"
+  ready="NO"
+  notes=""
+
+  if check_tcp_22 "${ip}"; then
+    ssh_tcp="PASS"
+  else
+    notes="port22_unreachable"
+    echo "${server_id},${ip},${ssh_tcp},${ssh_auth},${disk_ok},${ram_ok},${dpkg_lock},${apt_health},${ready},${notes}" >> "${OUT_CSV}"
+    continue
+  fi
+
+  if [[ "$HAVE_SSHPASS" == "1" ]]; then
+    SSH_RUN=(run_ssh_password "${username}" "${ip}" "${password}")
+  else
+    SSH_RUN=(run_ssh_key "${username}" "${ip}")
+    notes="${notes:+$notes|}sshpass_missing_using_key_auth"
+  fi
+
+  if "${SSH_RUN[@]}" "echo ok" >/dev/null 2>&1; then
+    ssh_auth="PASS"
+  else
+    notes="ssh_auth_failed"
+    echo "${server_id},${ip},${ssh_tcp},${ssh_auth},${disk_ok},${ram_ok},${dpkg_lock},${apt_health},${ready},${notes}" >> "${OUT_CSV}"
+    continue
+  fi
+
+  # Disk check: >= 8GB free on /
+  if "${SSH_RUN[@]}" \
+      "avail=\$(df -BG / | awk 'NR==2 {gsub(\"G\",\"\",\$4); print \$4}'); [ \"\${avail:-0}\" -ge 8 ]"; then
+    disk_ok="PASS"
+  else
+    notes="${notes:+$notes|}low_disk"
+  fi
+
+  # RAM check: >= 2GB
+  if "${SSH_RUN[@]}" \
+      "mem=\$(awk '/MemTotal/ {print int(\$2/1024/1024)}' /proc/meminfo); [ \"\${mem:-0}\" -ge 2 ]"; then
+    ram_ok="PASS"
+  else
+    notes="${notes:+$notes|}low_ram"
+  fi
+
+  # dpkg/apt lock check
+  if "${SSH_RUN[@]}" \
+      "if fuser /var/lib/dpkg/lock >/dev/null 2>&1 || fuser /var/lib/dpkg/lock-frontend >/dev/null 2>&1; then exit 1; else exit 0; fi"; then
+    dpkg_lock="PASS"
+  else
+    dpkg_lock="FAIL"
+    notes="${notes:+$notes|}dpkg_lock_detected"
+  fi
+
+  # apt health check (read-only)
+  if "${SSH_RUN[@]}" "apt-cache policy >/dev/null 2>&1"; then
+    apt_health="PASS"
+  else
+    apt_health="FAIL"
+    notes="${notes:+$notes|}apt_health_failed"
+  fi
+
+  if [[ "${ssh_tcp}" == "PASS" && "${ssh_auth}" == "PASS" && "${disk_ok}" == "PASS" && "${ram_ok}" == "PASS" && "${dpkg_lock}" == "PASS" && "${apt_health}" == "PASS" ]]; then
+    ready="YES"
+  fi
+
+  echo "${server_id},${ip},${ssh_tcp},${ssh_auth},${disk_ok},${ram_ok},${dpkg_lock},${apt_health},${ready},${notes}" >> "${OUT_CSV}"
+done < "${INPUT_FILE}"
+
+echo "Preflight report generated: ${OUT_CSV}"
+echo "Ready servers:"
+awk -F',' 'NR>1 && $9=="YES" {print " - " $1 " (" $2 ")"}' "${OUT_CSV}"
+
diff --git a/nonreg-framework.sh b/nonreg-framework.sh
new file mode 100755
index 0000000..712ee92
--- /dev/null
+++ b/nonreg-framework.sh
@@ -0,0 +1,285 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# -------------------------------------------------------------------
+# WEVADS / WEVIA anti-regression framework
+# Safe by design: read-only HTTP checks, no infra mutation.
+# -------------------------------------------------------------------
+
+BASE_URL="${BASE_URL:-https://weval-consulting.com}"
+TRACKING_BASE_URL="${TRACKING_BASE_URL:-http://151.80.235.110}"
+TRACKING_DOMAIN_URL="${TRACKING_DOMAIN_URL:-https://culturellemejean.charity}"
+API_KEY="${API_KEY:-}"
+GPU_MODEL="${GPU_MODEL:-qwen2.5:3b}"
+MAX_GREETING_SECONDS="${MAX_GREETING_SECONDS:-3}"
+MAX_DEEP_SECONDS="${MAX_DEEP_SECONDS:-60}"
+STRICT_CONFIDENTIALITY="${STRICT_CONFIDENTIALITY:-0}"
+
+REPORT_DIR="${REPORT_DIR:-./reports}"
+RUN_ID="$(date +%Y%m%d_%H%M%S)"
+REPORT_FILE="${REPORT_DIR}/nonreg_${RUN_ID}.md"
+
+mkdir -p "${REPORT_DIR}"
+
+PASS_COUNT=0
+FAIL_COUNT=0
+WARN_COUNT=0
+
+declare -a FAILURES
+declare -a WARNINGS
+
+log() { printf '%s\n' "$*"; }
+
+record_pass() {
+  PASS_COUNT=$((PASS_COUNT + 1))
+  log "PASS | $1"
+}
+
+record_fail() {
+  FAIL_COUNT=$((FAIL_COUNT + 1))
+  FAILURES+=("$1")
+  log "FAIL | $1"
+}
+
+record_warn() {
+  WARN_COUNT=$((WARN_COUNT + 1))
+  WARNINGS+=("$1")
+  log "WARN | $1"
+}
+
+http_status() {
+  local url="$1"
+  curl -sS -L -o /tmp/nonreg_body_${RUN_ID}.tmp -w "%{http_code} %{time_total}" --max-time 120 "$url"
+}
+
+check_status_200() {
+  local name="$1"
+  local url="$2"
+  local out code t
+  out="$(http_status "$url" || true)"
+  code="$(awk '{print $1}' <<<"$out")"
+  t="$(awk '{print $2}' <<<"$out")"
+  if [[ "$code" == "200" ]]; then
+    record_pass "${name} (${url}) code=${code} t=${t}s"
+  else
+    record_fail "${name} (${url}) expected 200 got ${code:-N/A} t=${t:-N/A}s"
+  fi
+}
+
+check_not_confidential_terms() {
+  local url="$1"
+  local body
+  body="$(curl -sS -L --max-time 60 "$url" || true)"
+  if [[ -z "$body" ]]; then
+    record_fail "Confidentiality scan cannot fetch ${url}"
+    return
+  fi
+
+  if rg -n -i "McKinsey|PwC|Deloitte|OpenAI|Anthropic|Abbott|AbbVie|J&J|CX3|DoubleM|89\\.167\\.40\\.150|88\\.198\\.4\\.195|\\b646\\b|\\b604\\b" <<<"$body" >/dev/null; then
+    if [[ "$STRICT_CONFIDENTIALITY" == "1" ]]; then
+      record_fail "Confidentiality terms detected in ${url}"
+    else
+      record_warn "Confidentiality terms detected in ${url} (strict mode disabled)"
+    fi
+  else
+    record_pass "Confidentiality scan clean for ${url}"
+  fi
+}
+
+check_wevia_greeting() {
+  local out code t
+  out="$(curl -sS -o /tmp/nonreg_wevia_${RUN_ID}.json -w "%{http_code} %{time_total}" \
+    --max-time 60 \
+    -H "Content-Type: application/json" \
+    -d '{"message":"Bonjour","mode":"fast"}' \
+    "${BASE_URL}/api/weval-ia" || true)"
+  code="$(awk '{print $1}' <<<"$out")"
+  t="$(awk '{print $2}' <<<"$out")"
+  if [[ "$code" != "200" ]]; then
+    record_fail "WEVIA greeting expected 200 got ${code:-N/A}"
+    return
+  fi
+  if awk "BEGIN {exit !($t < $MAX_GREETING_SECONDS)}"; then
+    record_pass "WEVIA greeting latency ${t}s < ${MAX_GREETING_SECONDS}s"
+  else
+    record_fail "WEVIA greeting latency ${t}s >= ${MAX_GREETING_SECONDS}s"
+  fi
+}
+
+check_wevia_deep() {
+  local out code t
+  out="$(curl -sS -o /tmp/nonreg_wevia_full_${RUN_ID}.json -w "%{http_code} %{time_total}" \
+    --max-time 120 \
+    -H "Content-Type: application/json" \
+    -d '{"message":"Fais une analyse concise supply chain internationale.","mode":"deep"}' \
+    "${BASE_URL}/api/weval-ia-full" || true)"
+  code="$(awk '{print $1}' <<<"$out")"
+  t="$(awk '{print $2}' <<<"$out")"
+  if [[ "$code" != "200" ]]; then
+    record_fail "WEVIA deep expected 200 got ${code:-N/A}"
+    return
+  fi
+  if awk "BEGIN {exit !($t < $MAX_DEEP_SECONDS)}"; then
+    record_pass "WEVIA deep latency ${t}s < ${MAX_DEEP_SECONDS}s"
+  else
+    record_fail "WEVIA deep latency ${t}s >= ${MAX_DEEP_SECONDS}s"
+  fi
+}
+
+check_gpu_chat() {
+  if [[ -z "$API_KEY" ]]; then
+    record_warn "GPU chat check skipped (API_KEY not set)"
+    return
+  fi
+  local payload out code
+  payload="$(printf '{"message":"Donne 3 points pour optimiser une campagne email.","model":"%s"}' "$GPU_MODEL")"
+  out="$(curl -sS -o /tmp/nonreg_gpu_${RUN_ID}.json -w "%{http_code}" \
+    --max-time 120 \
+    -H "Content-Type: application/json" \
+    -H "X-API-Key: ${API_KEY}" \
+    -d "$payload" \
+    "${BASE_URL}/api/gpu/chat.php" || true)"
+  code="$out"
+  if [[ "$code" == "200" ]]; then
+    if rg -n -i "Model not available" /tmp/nonreg_gpu_${RUN_ID}.json >/dev/null; then
+      record_fail "GPU chat returned model-not-available despite HTTP 200"
+    else
+      record_pass "GPU chat functional (model=${GPU_MODEL})"
+    fi
+  else
+    record_fail "GPU chat expected 200 got ${code:-N/A}"
+  fi
+}
+
+check_tracking_smoke() {
+  local out1 out2 c1 c2
+  out1="$(curl -sS -o /dev/null -w "%{http_code}" --max-time 30 "${TRACKING_BASE_URL}" || true)"
+  out2="$(curl -sS -o /dev/null -w "%{http_code}" --max-time 30 "${TRACKING_DOMAIN_URL}" || true)"
+  c1="$out1"
+  c2="$out2"
+
+  if [[ "$c1" =~ ^(200|301|302)$ ]]; then
+    record_pass "Tracking base reachable (${TRACKING_BASE_URL}) code=${c1}"
+  else
+    record_fail "Tracking base unreachable (${TRACKING_BASE_URL}) code=${c1:-N/A}"
+  fi
+
+  if [[ "$c2" =~ ^(200|301|302)$ ]]; then
+    record_pass "Tracking domain reachable (${TRACKING_DOMAIN_URL}) code=${c2}"
+  else
+    record_fail "Tracking domain unreachable (${TRACKING_DOMAIN_URL}) code=${c2:-N/A}"
+  fi
+}
+
+check_deliverscore_smoke() {
+  local out code t
+  if [[ -n "$API_KEY" ]]; then
+    out="$(curl -sS -o /tmp/nonreg_deliver_${RUN_ID}.json -w "%{http_code} %{time_total}" \
+      --max-time 120 \
+      "${BASE_URL}/api/deliverscore/scan.php?domain=gmail.com&api_key=${API_KEY}" || true)"
+  else
+    out="$(curl -sS -o /tmp/nonreg_deliver_${RUN_ID}.json -w "%{http_code} %{time_total}" \
+      --max-time 120 \
+      "${BASE_URL}/api/deliverscore/scan.php?domain=gmail.com" || true)"
+  fi
+  code="$(awk '{print $1}' <<<"$out")"
+  t="$(awk '{print $2}' <<<"$out")"
+  if [[ "$code" == "200" ]]; then
+    record_pass "DeliverScore smoke code=${code} t=${t}s"
+  elif [[ "$code" == "429" ]]; then
+    record_warn "DeliverScore rate-limited code=429 t=${t}s"
+  elif [[ "$code" =~ ^(401|403)$ ]]; then
+    record_warn "DeliverScore auth required code=${code} (provide API_KEY for strict test)"
+  else
+    record_fail "DeliverScore smoke unexpected code=${code:-N/A} t=${t:-N/A}s"
+  fi
+}
+
+main() {
+  log "=== NON-REG FRAMEWORK START (${RUN_ID}) ==="
+  log "BASE_URL=${BASE_URL}"
+  log "TRACKING_BASE_URL=${TRACKING_BASE_URL}"
+  log "TRACKING_DOMAIN_URL=${TRACKING_DOMAIN_URL}"
+
+  # Core pages
+  check_status_200 "Home" "${BASE_URL}/"
+  check_status_200 "Products hub" "${BASE_URL}/products/"
+  check_status_200 "WEVIA page" "${BASE_URL}/wevia"
+  check_status_200 "Platform" "${BASE_URL}/platform/"
+
+  # Products (known core URLs)
+  declare -a product_pages=(
+    "academy.html"
+    "arsenal.html"
+    "blueprintai.html"
+    "content-factory.html"
+    "deliverscore.html"
+    "gpu-inference.html"
+    "medreach.html"
+    "proposalai.html"
+    "storeforge.html"
+    "wevads.html"
+    "wevads-ia.html"
+    "wevia-whitelabel.html"
+    "workspace.html"
+  )
+  for page in "${product_pages[@]}"; do
+    check_status_200 "Product ${page}" "${BASE_URL}/products/${page}"
+  done
+
+  # Confidentiality scans on strategic pages
+  check_not_confidential_terms "${BASE_URL}/"
+  check_not_confidential_terms "${BASE_URL}/products/"
+  check_not_confidential_terms "${BASE_URL}/products/wevads-ia.html"
+  check_not_confidential_terms "${BASE_URL}/products/workspace.html"
+
+  # WEVIA performance checks
+  check_wevia_greeting
+  check_wevia_deep
+
+  # SaaS API checks (smoke)
+  check_deliverscore_smoke
+  check_status_200 "MedReach smoke" "${BASE_URL}/api/medreach/search.php?specialty=cardiologue&country=FR&limit=3"
+  check_gpu_chat
+  check_tracking_smoke
+
+  {
+    echo "# Rapport anti-regression ${RUN_ID}"
+    echo
+    echo "- Base URL: ${BASE_URL}"
+    echo "- Tracking base: ${TRACKING_BASE_URL}"
+    echo "- Tracking domain: ${TRACKING_DOMAIN_URL}"
+    echo
+    echo "## Resume"
+    echo
+    echo "- PASS: ${PASS_COUNT}"
+    echo "- WARN: ${WARN_COUNT}"
+    echo "- FAIL: ${FAIL_COUNT}"
+    echo
+    if (( WARN_COUNT > 0 )); then
+      echo "## Warnings"
+      printf -- "- %s\n" "${WARNINGS[@]}"
+      echo
+    fi
+    if (( FAIL_COUNT > 0 )); then
+      echo "## Failures"
+      printf -- "- %s\n" "${FAILURES[@]}"
+      echo
+    fi
+    echo "## Verdict"
+    if (( FAIL_COUNT == 0 )); then
+      echo "GO (no hard regression detected)."
+    else
+      echo "NO-GO (${FAIL_COUNT} hard failures)."
+    fi
+  } > "${REPORT_FILE}"
+
+  log "Report written: ${REPORT_FILE}"
+  log "=== NON-REG FRAMEWORK END (${RUN_ID}) ==="
+
+  if (( FAIL_COUNT > 0 )); then
+    exit 1
+  fi
+}
+
+main "$@"
diff --git a/reports/README.md b/reports/README.md
new file mode 100644
index 0000000..9505f92
--- /dev/null
+++ b/reports/README.md
@@ -0,0 +1,11 @@
+# Reports output
+
+This folder stores generated artifacts from:
+
+- `nonreg-framework.sh`
+- `multiinstall-safe-preflight.sh`
+
+Examples currently present:
+
+- `nonreg_*.md`: anti-regression run summaries
+- `multiinstall_preflight_*.csv`: server readiness preflight outputs
diff --git a/reports/multiinstall_preflight_20260309_221914.csv b/reports/multiinstall_preflight_20260309_221914.csv
new file mode 100644
index 0000000..f6896b7
--- /dev/null
+++ b/reports/multiinstall_preflight_20260309_221914.csv
@@ -0,0 +1 @@
+server_id,ip,ssh_tcp,ssh_auth,disk_ok,ram_ok,dpkg_lock,apt_health,ready,notes
diff --git a/reports/multiinstall_preflight_20260309_221935.csv b/reports/multiinstall_preflight_20260309_221935.csv
new file mode 100644
index 0000000..514a851
--- /dev/null
+++ b/reports/multiinstall_preflight_20260309_221935.csv
@@ -0,0 +1,4 @@
+server_id,ip,ssh_tcp,ssh_auth,disk_ok,ram_ok,dpkg_lock,apt_health,ready,notes
+180,101.46.69.207,PASS,FAIL,FAIL,FAIL,UNKNOWN,UNKNOWN,NO,ssh_auth_failed
+181,101.46.69.121,PASS,FAIL,FAIL,FAIL,UNKNOWN,UNKNOWN,NO,ssh_auth_failed
+182,101.46.65.209,PASS,FAIL,FAIL,FAIL,UNKNOWN,UNKNOWN,NO,ssh_auth_failed
diff --git a/reports/nonreg_20260309_221755.md b/reports/nonreg_20260309_221755.md
new file mode 100644
index 0000000..7b2f820
--- /dev/null
+++ b/reports/nonreg_20260309_221755.md
@@ -0,0 +1,20 @@
+# Rapport anti-regression 20260309_221755
+
+- Base URL: https://weval-consulting.com
+- Tracking base: http://151.80.235.110
+- Tracking domain: https://culturellemejean.charity
+
+## Resume
+
+- PASS: 24
+- WARN: 4
+- FAIL: 0
+
+## Warnings
+- Confidentiality terms detected in https://weval-consulting.com/products/wevads-ia.html (strict mode disabled)
+- Confidentiality terms detected in https://weval-consulting.com/products/workspace.html (strict mode disabled)
+- DeliverScore rate-limited code=429 t=0.540570s
+- GPU chat check skipped (API_KEY not set)
+
+## Verdict
+GO (no hard regression detected).
diff --git a/servers.example.csv b/servers.example.csv
new file mode 100644
index 0000000..0c5b444
--- /dev/null
+++ b/servers.example.csv
@@ -0,0 +1,4 @@
+# server_id,ip,username,password
+180,101.46.69.207,root,CHANGE_ME
+181,101.46.69.121,root,CHANGE_ME
+182,101.46.65.209,root,CHANGE_ME