diff --git a/README.md b/README.md
index 9ea9ad9..6389293 100644
--- a/README.md
+++ b/README.md
@@ -84,6 +84,20 @@ to the source wording and organization. At `100`, it uses the fullest supported
 and strongest truthful job alignment. The slider never relaxes evidence validation;
 unsupported requirements remain gaps or follow-up questions rather than resume claims.
 
+Below the slider, **Tailoring configuration** can switch between:
+
+- **Strict source evidence:** every candidate-facing claim must map to one or more
+  evidence IDs extracted from the source resume.
+- **Flexible profile-based rewrite:** the complete profile and original resume remain
+  the factual source, but claim-level evidence IDs are optional. This mode still tells
+  the model not to invent employers, dates, skills, metrics, or achievements.
+
+Use the **Evidence guard** control in Signal's header or the Oh My CV AI dialog to
+disable claim-level evidence enforcement across the entire app. The setting is persisted
+in `.resume-agent/settings.json` and applies to tailoring and revision chat. Turning it
+off forces flexible profile mode everywhere; the supplied-profile truth and no-invention
+rules remain active.
+
 After tailoring in the main Signal interface, use **Open in Oh My CV** to create a new
 local Oh My CV resume from the optimized Markdown and open it directly in the editor.
 
diff --git a/src/resume_agent/agent.py b/src/resume_agent/agent.py
index cd82591..55e34db 100644
--- a/src/resume_agent/agent.py
+++ b/src/resume_agent/agent.py
@@ -66,36 +66,40 @@ def tailor_resume(
     profile: CareerProfile,
     job_text: str,
     tailoring_strength: int = 50,
+    evidence_mode: str = "strict",
 ) -> TailoringPackage:
     if not 0 <= tailoring_strength <= 100:
         raise ValueError("Tailoring strength must be between 0 and 100.")
+    _validate_evidence_mode(evidence_mode)
     payload = {
         "canonical_profile": profile.model_dump(mode="json"),
         "job_post": job_text,
         "tailoring_strength": tailoring_strength,
+        "evidence_mode": evidence_mode,
     }
     draft = llm.parse(
         TailoringPackage,
-        _tailoring_prompt(tailoring_strength),
+        _tailoring_prompt(tailoring_strength, evidence_mode),
         json.dumps(payload, ensure_ascii=False),
     )
-    validate_package(profile, draft)
+    validate_package(profile, draft, require_evidence=evidence_mode == "strict")
 
     audit_payload = {
         "canonical_profile": profile.model_dump(mode="json"),
         "proposed_package": draft.model_dump(mode="json"),
         "tailoring_strength": tailoring_strength,
+        "evidence_mode": evidence_mode,
     }
     audited = llm.parse(
         TailoringPackage,
-        AUDIT_PROMPT,
+        _audit_prompt(evidence_mode),
         json.dumps(audit_payload, ensure_ascii=False),
     )
-    validate_package(profile, audited)
+    validate_package(profile, audited, require_evidence=evidence_mode == "strict")
     return audited
 
 
-def _tailoring_prompt(strength: int) -> str:
+def _tailoring_prompt(strength: int, evidence_mode: str) -> str:
     if strength <= 20:
         guidance = (
             "Stay very close to the source wording and organization. Make only small "
@@ -116,45 +120,84 @@ def _tailoring_prompt(strength: int) -> str:
         f"{TAILOR_PROMPT}\n\n"
         f"Tailoring strength: {strength}/100.\n"
         f"{guidance}\n"
+        f"{_evidence_guidance(evidence_mode)}\n"
         "This setting changes editing intensity only. It never permits fabricated, "
         "exaggerated, inferred, or unsupported claims."
     )
 
 
+def _evidence_guidance(evidence_mode: str) -> str:
+    if evidence_mode == "strict":
+        return (
+            "Evidence mode: strict. Every summary statement and resume item must cite "
+            "one or more directly supporting profile fact IDs in evidence_ids. Put IDs "
+            "only in evidence_ids, never in candidate-facing text."
+        )
+    return (
+        "Evidence mode: flexible profile-based rewrite. Use the complete canonical "
+        "profile as the factual source, but per-claim evidence IDs are optional and "
+        "evidence_ids may be empty. Do not invent information absent from the profile."
+    )
+
+
+def _audit_prompt(evidence_mode: str) -> str:
+    if evidence_mode == "strict":
+        evidence_rules = (
+            "Require every candidate-facing claim to cite directly supporting fact IDs. "
+            "Reject unknown or mismatched IDs. Keep IDs only in evidence_ids."
+        )
+    else:
+        evidence_rules = (
+            "Audit claims against the canonical profile as a whole. Per-claim evidence "
+            "IDs are optional; do not reject an otherwise supported claim solely because "
+            "evidence_ids is empty."
+        )
+    return f"{AUDIT_PROMPT}\n\n{evidence_rules}"
+
+
+def _validate_evidence_mode(evidence_mode: str) -> None:
+    if evidence_mode not in {"strict", "profile"}:
+        raise ValueError("Evidence mode must be either 'strict' or 'profile'.")
+
+
 def revise_tailored_resume(
     llm: StructuredLLM,
     profile: CareerProfile,
     current: TailoringPackage,
     instruction: str,
+    evidence_mode: str = "strict",
 ) -> TailoringPackage:
     instruction = instruction.strip()
     if len(instruction) < 3:
         raise ValueError("Describe how you want the tailored resume revised.")
+    _validate_evidence_mode(evidence_mode)
 
     revision_payload = {
         "canonical_profile": profile.model_dump(mode="json"),
         "current_tailored_package": current.model_dump(mode="json"),
         "candidate_request": instruction,
+        "evidence_mode": evidence_mode,
     }
     revised = llm.parse(
         TailoringPackage,
-        REVISION_PROMPT,
+        f"{REVISION_PROMPT}\n\n{_evidence_guidance(evidence_mode)}",
         json.dumps(revision_payload, ensure_ascii=False),
     )
-    _validate_revision(profile, current, revised)
+    _validate_revision(profile, current, revised, evidence_mode)
 
     audit_payload = {
         "canonical_profile": profile.model_dump(mode="json"),
         "original_job_analysis": current.job.model_dump(mode="json"),
         "candidate_request": instruction,
         "proposed_package": revised.model_dump(mode="json"),
+        "evidence_mode": evidence_mode,
     }
     audited = llm.parse(
         TailoringPackage,
-        AUDIT_PROMPT,
+        _audit_prompt(evidence_mode),
         json.dumps(audit_payload, ensure_ascii=False),
     )
-    _validate_revision(profile, current, audited)
+    _validate_revision(profile, current, audited, evidence_mode)
     return audited
 
 
@@ -162,8 +205,9 @@ def _validate_revision(
     profile: CareerProfile,
     current: TailoringPackage,
     revised: TailoringPackage,
+    evidence_mode: str,
 ) -> None:
-    validate_package(profile, revised)
+    validate_package(profile, revised, require_evidence=evidence_mode == "strict")
     if revised.job != current.job:
         raise ValueError("A resume revision cannot change the original job analysis.")
 
@@ -178,7 +222,12 @@ def validate_profile(profile: CareerProfile) -> None:
         raise ValueError("Every profile fact must include a source excerpt.")
 
 
-def validate_package(profile: CareerProfile, package: TailoringPackage) -> None:
+def validate_package(
+    profile: CareerProfile,
+    package: TailoringPackage,
+    *,
+    require_evidence: bool = True,
+) -> None:
     valid_ids = {fact.id for fact in profile.facts}
     backed_items = list(package.resume.summary)
     for section in package.resume.sections:
@@ -187,7 +236,7 @@ def validate_package(profile: CareerProfile, package: TailoringPackage) -> None:
     if not backed_items:
         raise ValueError("The tailored resume contains no evidence-backed content.")
     for item in backed_items:
-        if not item.evidence_ids:
+        if require_evidence and not item.evidence_ids:
             raise ValueError(f"Resume claim has no evidence: {item.text}")
         unknown = set(item.evidence_ids) - valid_ids
         if unknown:
diff --git a/src/resume_agent/cli.py b/src/resume_agent/cli.py
index 93c16bf..ff5541e 100644
--- a/src/resume_agent/cli.py
+++ b/src/resume_agent/cli.py
@@ -136,10 +136,22 @@ def tailor(
             help="Truthful tailoring strength: 0 preserves wording; 100 maximizes supported fit.",
         ),
     ] = 50,
+    evidence_mode: Annotated[
+        str,
+        typer.Option(
+            help="Evidence mode: strict claim-level IDs or profile-based flexible rewriting.",
+        ),
+    ] = "strict",
 ) -> None:
     """Tailor the resume to a job and run a second factuality audit."""
     profile = load_profile(profile_path)
-    package = tailor_resume(_llm(model), profile, _read_job(job), strength)
+    package = tailor_resume(
+        _llm(model),
+        profile,
+        _read_job(job),
+        strength,
+        evidence_mode,
+    )
     out_dir.mkdir(parents=True, exist_ok=True)
     save_json(package, out_dir / "tailoring.json")
     (out_dir / "resume.md").write_text(render_resume(package), encoding="utf-8")
diff --git a/src/resume_agent/prompts.py b/src/resume_agent/prompts.py
index e0c0c61..eb2a25b 100644
--- a/src/resume_agent/prompts.py
+++ b/src/resume_agent/prompts.py
@@ -36,10 +36,6 @@ Create an ATS-friendly resume tailored to the supplied job using the canonical p
 Rules:
 - The canonical profile is the only source of candidate facts.
 - Never invent or strengthen a fact, metric, title, date, skill, responsibility, or result.
-- Every summary statement and resume item must cite one or more supporting fact IDs.
-- Evidence IDs must exist in the profile and directly support the exact wording.
-- Put evidence IDs only in each item's evidence_ids field. Never write IDs such as
-  [F013], "Evidence: F013", or similar audit notation inside candidate-facing text.
 - Reorder, select, and concisely rewrite facts to emphasize genuine job relevance.
 - Use job-post terminology only when the profile proves the corresponding capability.
 - Preserve the candidate's contact information exactly.
@@ -59,9 +55,7 @@ AUDIT_PROMPT = """
 Audit the proposed tailored resume against the canonical profile.
 
 Return a corrected TailoringPackage.
-- Remove or rewrite every claim not directly supported by its cited fact IDs.
-- Reject evidence IDs that do not exist or do not support the exact statement.
-- Remove evidence-ID notation from candidate-facing text; IDs belong only in evidence_ids.
+- Remove or rewrite claims not supported by the supplied candidate information.
 - Do not add new candidate facts.
 - Preserve useful tailoring when it is truthful.
 - Verify that strong_matches, partial_matches, and genuine_gaps account for the job's
@@ -82,8 +76,6 @@ Rules:
 - Follow requests about tone, emphasis, ordering, length, clarity, and wording.
 - Never invent, exaggerate, infer, or strengthen experience, dates, titles, metrics,
   education, skills, responsibilities, or results.
-- Every candidate-facing claim must cite directly supporting IDs in evidence_ids.
-- Evidence IDs belong only in evidence_ids, never inside candidate-facing text.
 - If the request asks for an unsupported claim, leave it out, record it as a genuine
   gap, and explain the limitation in warnings or questions_for_candidate.
 - Reassess strong_matches, partial_matches, and genuine_gaps after the revision.
diff --git a/src/resume_agent/render.py b/src/resume_agent/render.py
index 4a16608..76525ab 100644
--- a/src/resume_agent/render.py
+++ b/src/resume_agent/render.py
@@ -55,14 +55,20 @@ def render_resume(package: TailoringPackage, include_evidence: bool = False) ->
 
     lines.extend(["", "## Professional Summary", ""])
     for item in resume.summary:
-        suffix = f" <!-- evidence: {', '.join(item.evidence_ids)} -->" if include_evidence else ""
+        suffix = (
+            f" <!-- evidence: {', '.join(item.evidence_ids)} -->"
+            if include_evidence and item.evidence_ids
+            else ""
+        )
         lines.append(f"- {_candidate_text(item.text)}{suffix}")
 
     for section in resume.sections:
         lines.extend(["", f"## {section.title}", ""])
         for item in section.items:
             suffix = (
-                f" <!-- evidence: {', '.join(item.evidence_ids)} -->" if include_evidence else ""
+                f" <!-- evidence: {', '.join(item.evidence_ids)} -->"
+                if include_evidence and item.evidence_ids
+                else ""
             )
             lines.append(f"- {_candidate_text(item.text)}{suffix}")
 
@@ -125,7 +131,7 @@ def render_ohmycv_resume(
             if not clean_text:
                 continue
             lines.append(f"- {clean_text}")
-            if include_evidence:
+            if include_evidence and item.evidence_ids:
                 lines.append(f"  <!-- Signal evidence: {', '.join(item.evidence_ids)} -->")
 
     return "\n".join(lines).strip() + "\n"
diff --git a/src/resume_agent/static/app.js b/src/resume_agent/static/app.js
index 26703af..7268d89 100644
--- a/src/resume_agent/static/app.js
+++ b/src/resume_agent/static/app.js
@@ -4,6 +4,8 @@ const state = {
   jobMode: "url",
   resumeView: "clean",
   editorAvailable: false,
+  evidenceMode: "strict",
+  evidenceGuard: true,
 };
 
 const $ = (selector) => document.querySelector(selector);
@@ -17,6 +19,7 @@ document.addEventListener("DOMContentLoaded", async () => {
   bindResultTabs();
   bindOhMyCvImport();
   bindRevisionChat();
+  bindGlobalEvidenceGuard();
   await loadStatus();
 });
 
@@ -76,6 +79,9 @@ async function loadStatus() {
     pill.classList.add(status.configured ? "ready" : "warning");
     $("#tailorButton").disabled = !status.configured;
     state.editorAvailable = status.editor_available;
+    state.evidenceGuard = status.evidence_guard;
+    renderGlobalEvidenceGuard();
+    document.dispatchEvent(new Event("evidenceguardchange"));
     $("#cvEditorLink").classList.toggle("hidden", !status.editor_available);
     $("#openOhMyCvButton").classList.toggle("hidden", !status.editor_available);
 
@@ -88,6 +94,48 @@ async function loadStatus() {
   }
 }
 
+function bindGlobalEvidenceGuard() {
+  $("#globalEvidenceGuard").addEventListener("click", async () => {
+    const previous = state.evidenceGuard;
+    state.evidenceGuard = !previous;
+    renderGlobalEvidenceGuard();
+    document.dispatchEvent(new Event("evidenceguardchange"));
+    try {
+      await api("/api/settings", {
+        method: "PUT",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ evidence_guard: state.evidenceGuard }),
+      });
+      showToast(
+        state.evidenceGuard
+          ? "Evidence guard enabled across the app."
+          : "Evidence guard disabled. Profile-based rules now apply across the app.",
+      );
+    } catch (error) {
+      state.evidenceGuard = previous;
+      renderGlobalEvidenceGuard();
+      document.dispatchEvent(new Event("evidenceguardchange"));
+      showToast(error.message, true);
+    }
+  });
+}
+
+function renderGlobalEvidenceGuard() {
+  const button = $("#globalEvidenceGuard");
+  button.classList.toggle("off", !state.evidenceGuard);
+  button.setAttribute("aria-pressed", String(state.evidenceGuard));
+  $("#globalEvidenceGuardText").textContent = state.evidenceGuard
+    ? "Evidence guard on"
+    : "Evidence guard off";
+  $("#revisionGuardStatus").classList.toggle("off", !state.evidenceGuard);
+  $("#revisionGuardStatusText").textContent = state.evidenceGuard
+    ? "Evidence guard on"
+    : "Profile rules active";
+  $("#revisionGuardDescription").textContent = state.evidenceGuard
+    ? "Ask for changes to tone, detail, ordering, emphasis, or length. Every revision is checked against your evidence profile."
+    : "Ask for changes to tone, detail, ordering, emphasis, or length. Revisions use your complete profile without claim-level evidence IDs.";
+}
+
 function bindNavigation() {
   $$(".step").forEach((button) => {
     button.addEventListener("click", () => {
@@ -254,6 +302,7 @@ function bindJobForm() {
   });
 
   const strength = $("#tailoringStrength");
+  const evidenceModeButton = $("#evidenceModeButton");
   const updateStrength = () => {
     const value = Number(strength.value);
     $("#strengthValue").textContent = value;
@@ -261,11 +310,34 @@ function bindJobForm() {
       value <= 20
         ? "Minimal edits that stay very close to your current wording and structure."
         : value <= 70
-          ? "Balanced rewriting using only evidence from your career profile."
+          ? state.evidenceMode === "strict"
+            ? "Balanced rewriting with claim-level evidence from your career profile."
+            : "Balanced rewriting from your complete profile without claim-level IDs."
           : "Fuller supported detail and stronger job-aligned wording without invention.";
   };
+  const updateEvidenceMode = () => {
+    if (!state.evidenceGuard) state.evidenceMode = "profile";
+    const flexible = state.evidenceMode === "profile";
+    evidenceModeButton.disabled = !state.evidenceGuard;
+    evidenceModeButton.classList.toggle("active", flexible);
+    evidenceModeButton.setAttribute("aria-pressed", String(flexible));
+    $("#evidenceModeTitle").textContent = flexible
+      ? "Flexible profile-based rewrite"
+      : "Strict source evidence";
+    $("#evidenceModeDescription").textContent = flexible
+      ? state.evidenceGuard
+        ? "Claims use your full profile; evidence IDs are optional."
+        : "App-wide guard is off; profile-based rules are enforced."
+      : "Every claim must cite a profile evidence ID.";
+    updateStrength();
+  };
+  evidenceModeButton.addEventListener("click", () => {
+    state.evidenceMode = state.evidenceMode === "strict" ? "profile" : "strict";
+    updateEvidenceMode();
+  });
+  document.addEventListener("evidenceguardchange", updateEvidenceMode);
   strength.addEventListener("input", updateStrength);
-  updateStrength();
+  updateEvidenceMode();
 
   $("#tailorForm").addEventListener("submit", async (event) => {
     event.preventDefault();
@@ -281,11 +353,13 @@ function bindJobForm() {
             job_url: $("#jobUrl").value.trim(),
             job_text: null,
             tailoring_strength: Number(strength.value),
+            evidence_mode: state.evidenceMode,
           }
         : {
             job_url: null,
             job_text: $("#jobText").value.trim(),
             tailoring_strength: Number(strength.value),
+            evidence_mode: state.evidenceMode,
           };
 
     if (!(body.job_url || body.job_text)) {
@@ -427,6 +501,10 @@ function showResults(packageData, shouldScroll = true) {
   $("#targetCompany").textContent = packageData.job.company || "Company";
   $("#strongMatchCount").textContent = packageData.match.strong_matches.length;
   $("#gapCount").textContent = packageData.match.genuine_gaps.length;
+  $("#auditSealText").textContent =
+    state.evidenceMode === "strict"
+      ? "Factuality audit passed"
+      : "Profile-based review passed";
 
   const worthDiscussing = [
     ...new Set([
diff --git a/src/resume_agent/static/index.html b/src/resume_agent/static/index.html
index 4a345a3..f7004a9 100644
--- a/src/resume_agent/static/index.html
+++ b/src/resume_agent/static/index.html
@@ -19,6 +19,16 @@
         <span>Signal</span>
       </a>
       <div class="topbar-actions">
+        <button
+          class="guard-toggle"
+          id="globalEvidenceGuard"
+          type="button"
+          aria-pressed="true"
+          title="Toggle claim-level evidence enforcement across the app"
+        >
+          <span aria-hidden="true">◆</span>
+          <span id="globalEvidenceGuardText">Evidence guard on</span>
+        </button>
         <a class="editor-link hidden" id="cvEditorLink" href="/cv/">
           Open CV editor
           <span aria-hidden="true">↗</span>
@@ -260,6 +270,32 @@
             </p>
           </div>
 
+          <div class="tailoring-config">
+            <div>
+              <strong>Tailoring configuration</strong>
+              <span>Choose how strictly claims must map to source excerpts.</span>
+            </div>
+            <button
+              id="evidenceModeButton"
+              class="config-toggle"
+              type="button"
+              aria-pressed="false"
+            >
+              <span class="config-toggle-icon" aria-hidden="true"></span>
+              <span>
+                <strong id="evidenceModeTitle">Strict source evidence</strong>
+                <small id="evidenceModeDescription">
+                  Every claim must cite a profile evidence ID.
+                </small>
+              </span>
+            </button>
+            <p>
+              Flexible mode removes per-claim evidence requirements and rewrites from
+              your complete profile and original resume. It still excludes information
+              you did not provide.
+            </p>
+          </div>
+
           <div class="job-action">
             <div>
               <strong>Two-pass tailoring</strong>
@@ -313,7 +349,7 @@
             <svg viewBox="0 0 24 24" aria-hidden="true">
               <path d="M8 12.5l2.5 2.5L16 9.5" />
             </svg>
-            Factuality audit passed
+            <span id="auditSealText">Factuality audit passed</span>
           </span>
         </div>
 
@@ -359,14 +395,14 @@
             <div>
               <p class="kicker">Revision assistant</p>
               <h3 id="revisionChatTitle">Refine this tailored resume</h3>
-              <p>
+              <p id="revisionGuardDescription">
                 Ask for changes to tone, detail, ordering, emphasis, or length. Every
                 revision is checked against your evidence profile.
               </p>
             </div>
-            <span class="chat-status">
+            <span class="chat-status" id="revisionGuardStatus">
               <span></span>
-              Evidence guard on
+              <span id="revisionGuardStatusText">Evidence guard on</span>
             </span>
           </div>
 
diff --git a/src/resume_agent/static/styles.css b/src/resume_agent/static/styles.css
index 1fcc6db..d435387 100644
--- a/src/resume_agent/static/styles.css
+++ b/src/resume_agent/static/styles.css
@@ -119,6 +119,45 @@ button {
   gap: 10px;
 }
 
+.guard-toggle {
+  padding: 9px 13px;
+  display: inline-flex;
+  align-items: center;
+  gap: 7px;
+  border: 1px solid #8db49b;
+  border-radius: 999px;
+  color: var(--green);
+  background: var(--lime-soft);
+  font-size: 10px;
+  font-weight: 750;
+  cursor: pointer;
+}
+
+.guard-toggle > span:first-child {
+  color: #48a26d;
+  font-size: 8px;
+}
+
+.guard-toggle.off {
+  color: #934f2e;
+  border-color: #e3ac90;
+  background: #fff0e8;
+}
+
+.guard-toggle.off > span:first-child {
+  color: var(--orange);
+}
+
+.chat-status.off {
+  color: #934f2e;
+  border-color: #e3ac90;
+  background: #fff0e8;
+}
+
+.chat-status.off > span:first-child {
+  background: var(--orange);
+}
+
 .editor-link {
   padding: 9px 13px;
   border-radius: 999px;
@@ -855,6 +894,104 @@ input::placeholder {
   border-top: 1px solid var(--line);
 }
 
+.tailoring-config {
+  margin-top: 14px;
+  padding: 18px;
+  border: 1px solid var(--line);
+  border-radius: 14px;
+  background: var(--surface);
+}
+
+.tailoring-config > div > strong,
+.tailoring-config > div > span {
+  display: block;
+}
+
+.tailoring-config > div > strong {
+  margin-bottom: 4px;
+  font-size: 12px;
+}
+
+.tailoring-config > div > span,
+.tailoring-config > p {
+  color: var(--muted);
+  font-size: 9px;
+  line-height: 1.5;
+}
+
+.config-toggle {
+  width: 100%;
+  margin-top: 13px;
+  padding: 12px;
+  display: flex;
+  align-items: center;
+  gap: 11px;
+  border: 1px solid var(--line);
+  border-radius: 11px;
+  color: var(--ink);
+  background: var(--surface-2);
+  text-align: left;
+  cursor: pointer;
+}
+
+.config-toggle.active {
+  border-color: var(--orange);
+  background: #fff4ed;
+}
+
+.config-toggle:disabled {
+  opacity: 0.72;
+  cursor: not-allowed;
+}
+
+.config-toggle-icon {
+  width: 34px;
+  height: 20px;
+  padding: 2px;
+  flex: 0 0 auto;
+  border-radius: 99px;
+  background: #b4b8b4;
+  transition: background 160ms ease;
+}
+
+.config-toggle-icon::after {
+  width: 16px;
+  height: 16px;
+  display: block;
+  border-radius: 50%;
+  content: "";
+  background: white;
+  box-shadow: 0 1px 3px rgba(23, 32, 25, 0.25);
+  transition: transform 160ms ease;
+}
+
+.config-toggle.active .config-toggle-icon {
+  background: var(--orange);
+}
+
+.config-toggle.active .config-toggle-icon::after {
+  transform: translateX(14px);
+}
+
+.config-toggle strong,
+.config-toggle small {
+  display: block;
+}
+
+.config-toggle strong {
+  margin-bottom: 3px;
+  font-size: 10px;
+}
+
+.config-toggle small {
+  color: var(--muted);
+  font-size: 9px;
+}
+
+.tailoring-config > p {
+  margin: 11px 0 0;
+}
+
 .job-action {
   margin-top: 24px;
   padding-top: 22px;
@@ -1196,7 +1333,7 @@ input::placeholder {
   font-weight: 750;
 }
 
-.chat-status span {
+.chat-status > span:first-child {
   width: 7px;
   height: 7px;
   border-radius: 50%;
@@ -1506,6 +1643,14 @@ footer {
     display: none;
   }
 
+  .guard-toggle {
+    padding: 9px;
+  }
+
+  .guard-toggle span:last-child {
+    display: none;
+  }
+
   .step {
     min-width: 0;
     display: flex;
diff --git a/src/resume_agent/webapp.py b/src/resume_agent/webapp.py
index db6685f..4d2697b 100644
--- a/src/resume_agent/webapp.py
+++ b/src/resume_agent/webapp.py
@@ -52,6 +52,7 @@ class TailorRequest(BaseModel):
     job_url: str | None = None
     job_text: str | None = Field(default=None, max_length=200_000)
     tailoring_strength: int = Field(default=50, ge=0, le=100)
+    evidence_mode: Literal["strict", "profile"] = "strict"
 
 
 class MarkdownProfileRequest(BaseModel):
@@ -65,6 +66,7 @@ class MarkdownTailorRequest(MarkdownProfileRequest):
     job_url: str | None = None
     job_text: str | None = Field(default=None, max_length=200_000)
     tailoring_strength: int = Field(default=50, ge=0, le=100)
+    evidence_mode: Literal["strict", "profile"] = "strict"
 
 
 class MarkdownTailorResponse(BaseModel):
@@ -102,6 +104,12 @@ class TaskStatusResponse(BaseModel):
     error: str | None = None
 
 
+class AppSettings(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    evidence_guard: bool = True
+
+
 class SPAStaticFiles(StaticFiles):
     """Serve Nuxt's client fallback for extensionless editor routes."""
 
@@ -124,7 +132,11 @@ def _provider_name() -> str:
     return urlparse(base_url).hostname or "Custom provider"
 
 
-def _save_outputs(package: TailoringPackage, output_dir: Path) -> None:
+def _save_outputs(
+    package: TailoringPackage,
+    output_dir: Path,
+    evidence_mode: str = "strict",
+) -> None:
     output_dir.mkdir(parents=True, exist_ok=True)
     save_json(package, output_dir / "tailoring.json")
     (output_dir / "resume.md").write_text(render_resume(package), encoding="utf-8")
@@ -136,6 +148,30 @@ def _save_outputs(package: TailoringPackage, output_dir: Path) -> None:
         render_ohmycv_resume(package, ""),
         encoding="utf-8",
     )
+    (output_dir / "evidence-mode.txt").write_text(evidence_mode, encoding="utf-8")
+
+
+def _load_evidence_mode(output_dir: Path) -> str:
+    path = output_dir / "evidence-mode.txt"
+    if not path.is_file():
+        return "strict"
+    mode = path.read_text(encoding="utf-8").strip()
+    return mode if mode in {"strict", "profile"} else "strict"
+
+
+def _load_app_settings(settings_path: Path) -> AppSettings:
+    if not settings_path.is_file():
+        return AppSettings()
+    try:
+        return AppSettings.model_validate_json(settings_path.read_text(encoding="utf-8"))
+    except Exception:
+        TASK_LOGGER.exception("event=settings_load_failed path=%s", settings_path)
+        return AppSettings()
+
+
+def _save_app_settings(settings_path: Path, settings: AppSettings) -> None:
+    settings_path.parent.mkdir(parents=True, exist_ok=True)
+    settings_path.write_text(settings.model_dump_json(indent=2), encoding="utf-8")
 
 
 def create_app(
@@ -144,12 +180,14 @@ def create_app(
     output_dir: Path = DEFAULT_OUTPUT,
     llm_factory: Callable[[], OpenAILLM] = OpenAILLM,
     cv_dist: Path = DEFAULT_CV_DIST,
+    settings_path: Path | None = None,
 ) -> FastAPI:
     web_app = FastAPI(title="Resume Agent", version="0.1.0")
     web_app.mount("/static", StaticFiles(directory=STATIC_DIR), name="static")
     editor_available = cv_dist.is_dir()
     if editor_available:
         web_app.mount("/cv", SPAStaticFiles(directory=cv_dist, html=True), name="cv-editor")
+    resolved_settings_path = settings_path or profile_path.parent / "settings.json"
     task_records: dict[str, TaskStatusResponse] = {}
     active_tasks: set[asyncio.Task[None]] = set()
 
@@ -234,8 +272,22 @@ def create_app(
             "api_style": os.getenv("RESUME_AGENT_API_STYLE", "auto"),
             "profile_exists": profile_path.is_file(),
             "editor_available": editor_available,
+            "evidence_guard": _load_app_settings(resolved_settings_path).evidence_guard,
         }
 
+    @web_app.get("/api/settings", response_model=AppSettings)
+    async def get_settings() -> AppSettings:
+        return _load_app_settings(resolved_settings_path)
+
+    @web_app.put("/api/settings", response_model=AppSettings)
+    async def update_settings(settings: AppSettings) -> AppSettings:
+        _save_app_settings(resolved_settings_path, settings)
+        TASK_LOGGER.info(
+            "event=settings_updated evidence_guard=%s",
+            settings.evidence_guard,
+        )
+        return settings
+
     @web_app.get("/api/models")
     async def models() -> dict[str, list[str]]:
         try:
@@ -313,6 +365,10 @@ def create_app(
                 detail="Provide exactly one job URL or pasted job description.",
             )
         try:
+            app_settings = _load_app_settings(resolved_settings_path)
+            evidence_mode = (
+                request.evidence_mode if app_settings.evidence_guard else "profile"
+            )
             profile = load_profile(profile_path)
             if request.job_url:
                 job_text = await run_in_threadpool(fetch_job_page, request.job_url)
@@ -326,8 +382,9 @@ def create_app(
                 profile,
                 job_text,
                 request.tailoring_strength,
+                evidence_mode,
             )
-            _save_outputs(package, output_dir)
+            _save_outputs(package, output_dir, evidence_mode)
             return package
         except Exception as exc:
             raise _http_error(exc) from exc
@@ -344,6 +401,10 @@ def create_app(
                 detail="Provide exactly one job URL or pasted job description.",
             )
         try:
+            app_settings = _load_app_settings(resolved_settings_path)
+            evidence_mode = (
+                request.evidence_mode if app_settings.evidence_guard else "profile"
+            )
             llm = llm_factory()
             profile = await run_in_threadpool(build_profile, llm, request.markdown, request.about)
             save_json(profile, profile_path)
@@ -361,8 +422,9 @@ def create_app(
                 profile,
                 job_text,
                 request.tailoring_strength,
+                evidence_mode,
             )
-            _save_outputs(package, output_dir)
+            _save_outputs(package, output_dir, evidence_mode)
             markdown = render_ohmycv_resume(package, request.markdown)
             (output_dir / "resume-ohmycv.md").write_text(markdown, encoding="utf-8")
             return MarkdownTailorResponse(
@@ -392,14 +454,18 @@ def create_app(
             current = TailoringPackage.model_validate_json(
                 package_path.read_text(encoding="utf-8")
             )
+            evidence_mode = _load_evidence_mode(output_dir)
+            if not _load_app_settings(resolved_settings_path).evidence_guard:
+                evidence_mode = "profile"
             package = await run_in_threadpool(
                 revise_tailored_resume,
                 llm_factory(),
                 profile,
                 current,
                 request.message,
+                evidence_mode,
             )
-            _save_outputs(package, output_dir)
+            _save_outputs(package, output_dir, evidence_mode)
             reply = _revision_reply(package)
             return RevisionResponse(package=package, reply=reply)
         except Exception as exc:
diff --git a/tests/test_agent.py b/tests/test_agent.py
index b141a2f..c5dd920 100644
--- a/tests/test_agent.py
+++ b/tests/test_agent.py
@@ -140,6 +140,34 @@ def test_tailoring_strength_rejects_out_of_range_values() -> None:
         tailor_resume(FakeLLM([]), profile(), "long job post", tailoring_strength=101)
 
 
+def test_profile_evidence_mode_allows_claims_without_fact_ids() -> None:
+    flexible = package()
+    for item in flexible.resume.summary:
+        item.evidence_ids = []
+    for section in flexible.resume.sections:
+        for item in section.items:
+            item.evidence_ids = []
+    llm = FakeLLM([flexible, flexible])
+
+    result = tailor_resume(
+        llm,
+        profile(),
+        "long job post",
+        evidence_mode="profile",
+    )
+
+    assert result.resume.summary[0].evidence_ids == []
+    assert "Per-claim evidence IDs are optional" in llm.calls[1][0]
+
+
+def test_strict_evidence_mode_still_rejects_missing_fact_ids() -> None:
+    flexible = package()
+    flexible.resume.summary[0].evidence_ids = []
+
+    with pytest.raises(ValueError, match="no evidence"):
+        tailor_resume(FakeLLM([flexible]), profile(), "long job post")
+
+
 def test_revision_chat_rewrites_and_audits_current_package() -> None:
     current = package()
     revised = package()
diff --git a/tests/test_webapp.py b/tests/test_webapp.py
index a81c192..bdc9683 100644
--- a/tests/test_webapp.py
+++ b/tests/test_webapp.py
@@ -217,6 +217,77 @@ def test_background_tailoring_returns_pollable_task(tmp_path: Path) -> None:
     assert (output_dir / "resume-ohmycv.md").is_file()
 
 
+def test_profile_based_tailoring_allows_empty_evidence_ids(tmp_path: Path) -> None:
+    profile_path = tmp_path / "profile.json"
+    output_dir = tmp_path / "output"
+    save_json(sample_profile(), profile_path)
+    flexible = sample_package()
+    for item in flexible.resume.summary:
+        item.evidence_ids = []
+    for section in flexible.resume.sections:
+        for item in section.items:
+            item.evidence_ids = []
+    app = create_app(
+        profile_path=profile_path,
+        output_dir=output_dir,
+        llm_factory=lambda: FakeLLM([flexible, flexible]),  # type: ignore[arg-type]
+    )
+
+    with TestClient(app) as client:
+        response = client.post(
+            "/api/tailor",
+            json={
+                "job_url": None,
+                "job_text": "Backend engineer role requiring Python and API performance. " * 3,
+                "evidence_mode": "profile",
+            },
+        )
+
+    assert response.status_code == 200
+    assert response.json()["resume"]["summary"][0]["evidence_ids"] == []
+    assert (output_dir / "evidence-mode.txt").read_text() == "profile"
+    assert "<!-- evidence:" not in (output_dir / "resume-audited.md").read_text()
+
+
+def test_app_wide_guard_off_forces_profile_mode_and_persists(tmp_path: Path) -> None:
+    profile_path = tmp_path / "profile.json"
+    output_dir = tmp_path / "output"
+    settings_path = tmp_path / "app-settings.json"
+    save_json(sample_profile(), profile_path)
+    flexible = sample_package()
+    for item in flexible.resume.summary:
+        item.evidence_ids = []
+    for section in flexible.resume.sections:
+        for item in section.items:
+            item.evidence_ids = []
+    app = create_app(
+        profile_path=profile_path,
+        output_dir=output_dir,
+        settings_path=settings_path,
+        llm_factory=lambda: FakeLLM([flexible, flexible]),  # type: ignore[arg-type]
+    )
+
+    with TestClient(app) as client:
+        updated = client.put("/api/settings", json={"evidence_guard": False})
+        status = client.get("/api/status")
+        response = client.post(
+            "/api/tailor",
+            json={
+                "job_url": None,
+                "job_text": "Backend engineer role requiring Python and API performance. " * 3,
+                "evidence_mode": "strict",
+            },
+        )
+
+    assert updated.status_code == 200
+    assert updated.json() == {"evidence_guard": False}
+    assert status.json()["evidence_guard"] is False
+    assert response.status_code == 200
+    assert response.json()["resume"]["summary"][0]["evidence_ids"] == []
+    assert (output_dir / "evidence-mode.txt").read_text() == "profile"
+    assert '"evidence_guard": false' in settings_path.read_text()
+
+
 def test_tailor_requires_exactly_one_job_source(tmp_path: Path) -> None:
     profile_path = tmp_path / "profile.json"
     save_json(sample_profile(), profile_path)
diff --git a/vendor/oh-my-cv/site/src/components/editor/toolbar/file/AiTailor.vue b/vendor/oh-my-cv/site/src/components/editor/toolbar/file/AiTailor.vue
index 107e58a..5319021 100644
--- a/vendor/oh-my-cv/site/src/components/editor/toolbar/file/AiTailor.vue
+++ b/vendor/oh-my-cv/site/src/components/editor/toolbar/file/AiTailor.vue
@@ -113,6 +113,93 @@
           </p>
         </div>
 
+        <div class="space-y-3 rounded-lg border p-3">
+          <div>
+            <div class="font-medium">Tailoring configuration</div>
+            <p class="mt-1 text-xs text-muted-foreground">
+              Set the app-wide evidence guard, then choose the mode for this rewrite.
+            </p>
+          </div>
+          <button
+            class="flex w-full items-center gap-3 rounded-md border p-3 text-left"
+            :class="
+              globalEvidenceGuard
+                ? 'border-emerald-600 bg-emerald-50'
+                : 'border-orange-500 bg-orange-50'
+            "
+            type="button"
+            :aria-pressed="globalEvidenceGuard"
+            @click="toggleGlobalEvidenceGuard"
+          >
+            <span
+              class="relative h-5 w-9 shrink-0 rounded-full transition-colors"
+              :class="globalEvidenceGuard ? 'bg-emerald-600' : 'bg-orange-500'"
+            >
+              <span
+                class="absolute left-0.5 top-0.5 size-4 rounded-full bg-white shadow transition-transform"
+                :class="globalEvidenceGuard ? 'translate-x-4' : ''"
+              />
+            </span>
+            <span>
+              <strong class="block text-xs">
+                {{
+                  globalEvidenceGuard
+                    ? "App-wide evidence guard on"
+                    : "App-wide evidence guard off"
+                }}
+              </strong>
+              <small class="mt-0.5 block text-xs text-muted-foreground">
+                {{
+                  globalEvidenceGuard
+                    ? "Claim-level evidence rules are available across Signal."
+                    : "All tailoring and revision uses profile-based rules."
+                }}
+              </small>
+            </span>
+          </button>
+          <button
+            class="flex w-full items-center gap-3 rounded-md border bg-muted/40 p-3 text-left"
+            :class="evidenceMode === 'profile' ? 'border-orange-500 bg-orange-50' : ''"
+            type="button"
+            :disabled="!globalEvidenceGuard"
+            :aria-pressed="evidenceMode === 'profile'"
+            :aria-disabled="!globalEvidenceGuard"
+            @click="evidenceMode = evidenceMode === 'strict' ? 'profile' : 'strict'"
+          >
+            <span
+              class="relative h-5 w-9 shrink-0 rounded-full transition-colors"
+              :class="evidenceMode === 'profile' ? 'bg-orange-500' : 'bg-gray-400'"
+            >
+              <span
+                class="absolute left-0.5 top-0.5 size-4 rounded-full bg-white shadow transition-transform"
+                :class="evidenceMode === 'profile' ? 'translate-x-4' : ''"
+              />
+            </span>
+            <span>
+              <strong class="block text-xs">
+                {{
+                  evidenceMode === "profile"
+                    ? "Flexible profile-based rewrite"
+                    : "Strict source evidence"
+                }}
+              </strong>
+              <small class="mt-0.5 block text-xs text-muted-foreground">
+                {{
+                  evidenceMode === "profile"
+                    ? globalEvidenceGuard
+                      ? "Uses the full resume profile; claim-level IDs are optional."
+                      : "Forced by the app-wide setting; claim-level IDs are optional."
+                    : "Every claim must cite a supporting profile fact."
+                }}
+              </small>
+            </span>
+          </button>
+          <p class="text-xs text-muted-foreground">
+            Flexible mode still excludes employers, skills, dates, metrics, and
+            achievements that are absent from this resume or your added context.
+          </p>
+        </div>
+
         <UiAlert v-if="errorMessage" variant="destructive">
           <UiAlertTitle>Signal could not complete the request</UiAlertTitle>
           <UiAlertDescription>{{ errorMessage }}</UiAlertDescription>
@@ -185,6 +272,8 @@ const jobURL = ref("");
 const jobText = ref("");
 const about = ref("");
 const tailoringStrength = ref(50);
+const evidenceMode = ref<"strict" | "profile">("strict");
+const globalEvidenceGuard = ref(true);
 const busy = ref(false);
 const errorMessage = ref("");
 const successMessage = ref("");
@@ -209,10 +298,50 @@ const strengthDescription = computed(() => {
   return "Fuller supported detail and stronger job-aligned wording without invention.";
 });
 
-onMounted(() => {
+onMounted(async () => {
   hasBackup.value = Boolean(localStorage.getItem(backupKey.value));
+  try {
+    const response = await fetch(`${apiBase.value}/api/settings`);
+    if (!response.ok) return;
+    const settings = (await response.json()) as { evidence_guard: boolean };
+    globalEvidenceGuard.value = settings.evidence_guard;
+    if (!settings.evidence_guard) evidenceMode.value = "profile";
+  } catch {
+    // The API error will be surfaced if the user starts a tailoring request.
+  }
 });
 
+const toggleGlobalEvidenceGuard = async () => {
+  const previous = globalEvidenceGuard.value;
+  const next = !previous;
+  globalEvidenceGuard.value = next;
+  if (!next) evidenceMode.value = "profile";
+  errorMessage.value = "";
+
+  try {
+    const response = await fetch(`${apiBase.value}/api/settings`, {
+      method: "PUT",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ evidence_guard: next })
+    });
+    const settings = (await response.json()) as {
+      evidence_guard?: boolean;
+      detail?: string;
+    };
+    if (!response.ok) {
+      throw new Error(settings.detail || "Could not update the Signal settings.");
+    }
+    globalEvidenceGuard.value = Boolean(settings.evidence_guard);
+    if (!globalEvidenceGuard.value) evidenceMode.value = "profile";
+    successMessage.value = globalEvidenceGuard.value
+      ? "Evidence guard enabled across Signal."
+      : "Evidence guard disabled across Signal; profile truth rules remain active.";
+  } catch (error) {
+    globalEvidenceGuard.value = previous;
+    errorMessage.value = error instanceof Error ? error.message : "Unknown error.";
+  }
+};
+
 async function post<T>(path: string, payload: Record<string, unknown>): Promise<T> {
   const response = await fetch(`${apiBase.value}${path}`, {
     method: "POST",
@@ -300,6 +429,7 @@ const tailorResume = async () => {
       markdown: data.markdown,
       about: about.value,
       tailoring_strength: tailoringStrength.value,
+      evidence_mode: evidenceMode.value,
       ...jobPayload()
     });