@@ -22,7 +22,7 @@ Token Meter reads local agent traces, calculates clearly labeled usage estimates
 | `token_meter/runtimes/` | Registered runtime discovery, parsing, revisions, and safe projections |
 | `token_meter/platforms/` | Host paths, process/update policy, and trash behavior |
 | `token_meter/domain/` | Runtime-neutral usage, timing, tools, insights, and aggregates |
-| `token_meter/domain/compare.py` | Content-free session comparison, insights, and same-title matching |
+| `token_meter/domain/compare.py` | Content-free session comparison, cheapest/fastest picks, and same-title matching |
 | `token_meter/projections.py` | Explicit allowlisted public compatibility projections |
 | `page.html` | Entire browser dashboard: markup, styles, routing, and JavaScript |
 | `specs/DESIGN.md` | Visual system: tokens, provider identity, components, readability floor, contract |
 
@@ -786,13 +786,9 @@
 .compareSection h3{margin:0 0 10px;font-size:var(--fs-14)}
 .compareSectionHead{display:flex;align-items:center;justify-content:space-between;gap:12px;margin-bottom:10px}
 .compareSectionHead h3{margin:0}
-.compareInsights{list-style:none;margin:0;padding:0;display:grid;gap:6px}
-.compareInsights li{display:grid;grid-template-columns:10px minmax(0,1fr);gap:8px;align-items:baseline;font-size:var(--fs-13);line-height:1.5;color:var(--dim)}
-.compareInsights li:before{content:"";width:8px;height:8px;border-radius:var(--radius-round);background:var(--accent);transform:translateY(-1px)}
-.compareInsights li.warn:before{background:var(--warn)}
-.compareInsights li.good:before{background:var(--good)}
-.compareInsights li.caveat{color:var(--faint);font-size:var(--fs-12)}
-.compareInsights li.caveat:before{background:var(--faint)}
+.compareWinners{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:12px}
+.compareWinners strong{display:flex;align-items:center;gap:8px}
+.compareWinners small{overflow:hidden;text-overflow:ellipsis;white-space:nowrap}
 .compareTableWrap{overflow-x:auto}
 .compareTable{width:100%;border-collapse:collapse;font-size:var(--fs-13)}
 .compareTable th,.compareTable td{padding:6px 10px;text-align:right;border-bottom:1px solid var(--line);white-space:nowrap}
@@ -3760,6 +3756,7 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
  return `<div class=h>${cursor?'Local cost estimate':'Cost calculation'}</div>${rows}<div class=total><span>Total</span><span>${money(s.total_cost||0)}${s.cost_approx?' est':''}</span></div><div class=note>${esc(note)}</div>`;
 }
 let LATEST=null, CURRENT=null, pinned=null, sessionRequest=0, selectedPollId='', RUNTIME_CATALOG={};
+const LIVE_XSESSION={state:{snapshot:null,revision:''},session:{snapshot:null,revision:''}};
 let AGENT_ACCESS=null, pendingAgentAccess=null, agentAccessLoading=false;
 const CURRENT_SESSION_ORDER_KEY='tm_current_session_order_v1';
 const CURRENT_SESSION_ORDER_MIGRATION_KEY='tm_current_session_newest_first_v1';
@@ -3858,7 +3855,7 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
 function sessionScopeCopy(scope){
  return {
   all:['Local session history.','Search, filter, and compare local session history.'],
-  compare:['Selected sessions side by side.','Compare up to four selected sessions and review content-free insights.'],
+  compare:['Selected sessions side by side.','Compare up to four selected sessions and see which is cheapest and fastest.'],
   subagents:['Investigate child-agent spend, completion, and role patterns.','Filter child agents, inspect incomplete work, and open their parent sessions.'],
  }[scope]||['Live local traces · last 30 minutes.','Select a card to inspect it. Drag a card or press ⌥ and an arrow key to reorder.'];
 }
@@ -4358,7 +4355,7 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
 const NAV_COMMANDS=[
  {id:'sessions',label:'Sessions',description:'Choose a recently active local session',section:'Navigate',action:'sessions'},
  {id:'all-sessions',label:'All sessions',description:'Search, filter, and open recorded session history',aliases:'logs',section:'Navigate',route:'sessions-all',glyph:'All'},
- {id:'compare-sessions',label:'Compare sessions',description:'Put selected sessions side by side with insights',aliases:'diff same prompt runs',section:'Navigate',route:'sessions-compare',glyph:'A|B'},
+ {id:'compare-sessions',label:'Compare sessions',description:'Put selected sessions side by side',aliases:'diff same prompt runs',section:'Navigate',route:'sessions-compare',glyph:'A|B'},
  {id:'spend',label:'Spend',description:'Review spend across days and runtimes',section:'Navigate',route:'spend'},
  {id:'models',label:'Models',description:'Compare model spend, speed, and cache efficiency',section:'Navigate',route:'models'},
  {id:'subagents',label:'Subagents',description:'Track custom-agent spend and inspect child runs',section:'Navigate',route:'subagents'},
@@ -9997,12 +9994,18 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
   return `<article class="card compareCard" style="--compare-color:${compareColor(i)}"><div class=compareCardTop><span class=compareLetter>${s.letter}</span><span class="badge app ${appBadgeClass(s)}">${runtimeMarkSvg(s)}${esc(s.label||s.provider)}</span>${baseline&&sessions.length>1?'<span class=compareBaselineTag>Baseline</span>':''}</div><div class=compareCardTitle title="${esc(s.title)}">${esc(s.title)}</div><div class=compareCardMeta>${esc(s.primary_model||(s.models||[]).join(', ')||'model unknown')} · ${esc(s.start||'')}</div><div class="compareCardCost mono">${s.cost==null?costValueHtml('',false):esc(money(s.cost)+(s.cost_approx||s.usage_basis!=='reported'?' est':''))}</div><div class=compareCardMeta>${f(s.turns)} execs · ${s.duration_s==null?'time unavailable':esc(waitFmt(s.duration_s))+' active'} · ${s.tokens==null?'tokens unavailable':esc(compactNumber(s.tokens))+' tok'}</div><div class=compareCardActions><button type=button data-compare-open-session="${esc(s.open_id||s.id)}">Open</button>${sessions.length>1?`<button type=button data-compare-baseline="${esc(s.key)}"${baseline?' disabled':''}>${baseline?'Baseline':'Set as baseline'}</button>`:''}<button type=button data-compare-remove="${esc(s.key)}">Remove</button></div></article>`;
  }).join('')}</div>`;
  if(sessions.length<2){
-  root.innerHTML=`${notice}${cards}<section class="card compareEmpty emptyState"><h3>Add one more session</h3><p>Pick another session in All sessions, or add a matching run below, to see side-by-side metrics and insights.</p><button class=tbtn type=button data-compare-pick>Choose sessions</button></section>${compareMatchesHtml(compareData.matches,sessions)}`;
+  root.innerHTML=`${notice}${cards}<section class="card compareEmpty emptyState"><h3>Add one more session</h3><p>Pick another session in All sessions, or add a matching run below, to see side-by-side metrics.</p><button class=tbtn type=button data-compare-pick>Choose sessions</button></section>${compareMatchesHtml(compareData.matches,sessions)}`;
   return;
  }
- const insights=`<section class="card compareSection"><h3>Insights</h3><ul class=compareInsights>${(compareData.insights||[]).map(item=>`<li class="${esc(item.kind==='caveat'?'caveat':item.tone||'info')}">${esc(item.text)}</li>`).join('')}</ul></section>`;
+ const winner=(key,label,fmt)=>{
+  const index=sessions.findIndex(s=>s.key===best[key]),s=sessions[index],known=sessions.filter(row=>row[key]!=null);
+  const value=s?`<span class=compareLetter style="--compare-color:${compareColor(index)}">${s.letter}</span>${esc(fmt(s))}`:known.length>=2?'Tied':'--';
+  const note=s?esc(s.title):known.length>=2?`All at ${esc(fmt(known[0]))}`:'Not reported for enough sessions';
+  return `<div class=metric><span>${label}</span><strong>${value}</strong><small${s?` title="${esc(s.title)}"`:''}>${note}</small></div>`;
+ };
+ const winners=`<section class="card compareSection"><div class=compareWinners>${winner('cost','Cheapest',s=>money(s.cost)+(s.cost_approx||s.usage_basis!=='reported'?' est':''))}${winner('duration_s','Fastest',s=>waitFmt(s.duration_s)+' active')}</div></section>`;
  const table=`<section class="card compareSection"><div class=compareSectionHead><h3>Side by side</h3><span class=allSessionsHint>Deltas are against baseline ${esc((sessions.find(s=>s.key===compareBaseline)||sessions[0]).letter)} · unavailable evidence shows --</span></div>${compareTableHtml(sessions,best)}</section>`;
- root.innerHTML=notice+cards+insights+table+compareChartHtml(sessions)+compareBreakdownHtml(sessions)+compareMatchesHtml(compareData.matches,sessions);
+ root.innerHTML=notice+cards+winners+table+compareChartHtml(sessions)+compareBreakdownHtml(sessions)+compareMatchesHtml(compareData.matches,sessions);
 }
 $('session-compare').onclick=event=>{
  const target=event.target.closest('button');
@@ -11373,14 +11376,31 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
  if(show)applyHashRoute();
  return refreshSelectedSession({show,notifyOnError:true});
 }
+function adoptXsession(holder,state){
+ if(state?.xsession_unchanged){
+  if(holder.snapshot&&state.xsession_revision===holder.revision){state.xsession=holder.snapshot;return state;}
+  holder.snapshot=null;holder.revision='';
+  return null;
+ }
+ holder.snapshot=state?.xsession||null;
+ holder.revision=holder.snapshot?String(state?.xsession_revision||''):'';
+ return state;
+}
+async function fetchLiveJson(url,holder,label){
+ for(let attempt=0;attempt<2;attempt++){
+  const response=await fetch(url+(url.includes('?')?'&':'?')+'xsession='+encodeURIComponent(holder.revision),{cache:'no-store'});
+  if(!response.ok)throw new Error(`${label} request failed (${response.status}).`);
+  const state=adoptXsession(holder,await response.json());
+  if(state)return state;
+ }
+ throw new Error(`${label} request failed.`);
+}
 async function refreshSelectedSession({show=true,notifyOnError=false}={}){
  const id=pinned,request=sessionRequest;
  if(!id||selectedPollId===id)return null;
  selectedPollId=id;
  try{
-  const response=await fetch('/session?id='+encodeURIComponent(id)+'&live=1',{cache:'no-store'});
-  if(!response.ok)throw new Error(`Session request failed (${response.status}).`);
-  const state=await response.json();
+  const state=await fetchLiveJson('/session?id='+encodeURIComponent(id)+'&live=1',LIVE_XSESSION.session,'Session');
   if(request!==sessionRequest||pinned!==id)return null;
   if(!state||!stateSessionId(state))throw new Error('Session was not found.');
   CURRENT=state;
@@ -11461,9 +11481,7 @@ <h3 id=session-model-identity-title>Assign model to this session</h3>
  if(statePollBusy)return;
  statePollBusy=true;
  try{
-  const response=await fetch('/state',{cache:'no-store'});
-  if(!response.ok)throw new Error(`State request failed (${response.status}).`);
-  applyLiveState(await response.json());
+  applyLiveState(await fetchLiveJson('/state',LIVE_XSESSION.state,'State'));
  }catch(error){
   if($('view-capabilities').classList.contains('on')){
    if(capabilityLastGoodAt)setCapabilityAvailability('stale');
 
@@ -20,7 +20,7 @@ Token Meter reads local agent traces, calculates clearly labeled usage estimates
 | `token_meter/runtimes/` | Registered runtime discovery, parsing, revisions, and safe projections |
 | `token_meter/platforms/` | Host paths, process/update policy, and trash behavior |
 | `token_meter/domain/` | Runtime-neutral usage, timing, tools, insights, and aggregates |
-| `token_meter/domain/compare.py` | Content-free session comparison, insights, and same-title matching |
+| `token_meter/domain/compare.py` | Content-free session comparison, cheapest/fastest picks, and same-title matching |
 | `token_meter/projections.py` | Explicit allowlisted public compatibility projections |
 | `page.html` | Entire browser dashboard: markup, styles, routing, and JavaScript |
 | `specs/DESIGN.md` | Visual system: tokens, provider identity, components, readability floor, contract |
 
@@ -245,6 +245,13 @@ anonymous child-agent cohorts and `/session` exposes only the selected
 allowlisted group; neither native nor MCP projections receive these browser
 fields.
 
+Cross-session snapshots are replaced wholesale and never mutated in place.
+`/state` and `/session` encode each snapshot's browser projection once and tag
+it with `xsession_revision`. A client that sends `?xsession=<revision>` for the
+snapshot it holds gets `xsession_unchanged: true` without the block, which
+otherwise dominates every one-second poll. Mutating a published snapshot in
+place would serve stale cached JSON; rebuild a new snapshot instead.
+
 `/session/compare?ids=` projects one to four selected traces through
 `token_meter/domain/compare.py`: an allowlisted, content-free record per trace
 (no paths, prompts, or per-execution text), best-per-metric markers, rule-based
 
@@ -148,74 +148,30 @@ def test_letters_best_and_same_prompt(self):
         self.assertEqual(result["best"]["cost"], "a")
         self.assertEqual(result["best"]["duration_s"], "b")
         self.assertEqual(result["sessions"][0]["key"], "a")
-        texts = " ".join(item["text"] for item in result["insights"])
-        self.assertIn("matching titles", texts)
-        self.assertIn("3.0×", texts)
-        self.assertIn("Identical setup", texts)
+        self.assertNotIn("insights", result)
 
-    def test_setup_difference_and_cost_driver(self):
+    def test_tied_metric_has_no_best(self):
         result = compare_sessions(self.entries(
-            ("a", {}, {"cost": 1.0}),
-            ("b", {"primary_model": "claude-sonnet-5-5", "models": ["sonnet-5-5"]},
-             {"cost": 2.0, "parts": {"input": 0.1, "cache_write": 1.3, "cache_read": 0.3, "output": 0.3}}),
-        ))
-        texts = " ".join(item["text"] for item in result["insights"])
-        self.assertIn("model", texts)
-        self.assertIn("claude-sonnet-5-5", texts)
-        self.assertIn("cache writes", texts)
-
-    def test_cost_driver_larger_than_net_gap_is_described_as_offset(self):
-        result = compare_sessions(self.entries(
-            ("a", {}, {"cost": 1.0, "parts": {"input": 0.1, "cache_write": 0.6, "cache_read": 0.1, "output": 0.2}}),
-            ("b", {}, {"cost": 1.5, "parts": {"input": 0.1, "cache_write": 0.1, "cache_read": 1.1, "output": 0.2}}),
-        ))
-        driver = [i["text"] for i in result["insights"] if i["kind"] == "cost_driver"]
-        self.assertEqual(len(driver), 1)
-        self.assertIn("The biggest cost difference is cache reads", driver[0])
-        self.assertIn("net gap of $0.5.", driver[0])
-
-    def test_context_pressure_is_one_grouped_warning(self):
-        result = compare_sessions(self.entries(
-            ("a", {}, {"peak": 170000}), ("b", {}, {"peak": 190000}), ("c", {}, {"peak": 50000})))
-        context = [i["text"] for i in result["insights"] if i["kind"] == "context" and i["tone"] == "warn"]
-        self.assertEqual(context, ["A (85%) and B (95%) came within 20% of the context window at peak."])
+            ("a", {}, {"cost": 2.0, "duration": 300}), ("b", {}, {"cost": 2.0, "duration": 100})))
+        self.assertNotIn("cost", result["best"])
+        self.assertEqual(result["best"]["duration_s"], "b")
 
     def test_unavailable_cost_is_excluded_not_compared_as_zero(self):
         entries = self.entries(("a", {}, {"cost": 1.0}), ("b", {}, {"cost": 2.0}))
         entries[1]["cost"] = None
         result = compare_sessions(entries)
         self.assertNotIn("cost", result["best"])
-        texts = " ".join(item["text"] for item in result["insights"])
-        self.assertIn("Cost is unavailable for B", texts)
-        self.assertNotIn("×", " ".join(i["text"] for i in result["insights"] if i["kind"] == "cost"))
-
-    def test_variance_needs_three_same_prompt_runs(self):
-        two = compare_sessions(self.entries(("a", {}, {"cost": 1.0}), ("b", {}, {"cost": 2.0})))
-        self.assertFalse(any(i["kind"] == "variance" for i in two["insights"]))
-        three = compare_sessions(self.entries(
-            ("a", {}, {"cost": 1.0}), ("b", {}, {"cost": 2.0}), ("c", {}, {"cost": 1.5})))
-        self.assertTrue(any(i["kind"] == "variance" for i in three["insights"]))
-
-    def test_different_titles_and_tool_errors(self):
+
+    def test_different_titles(self):
         result = compare_sessions(self.entries(
-            ("a", {"title": "One thing"}, {"errors": 0}),
-            ("b", {"title": "Another thing"}, {"errors": 4}),
-        ))
+            ("a", {"title": "One thing"}, {}), ("b", {"title": "Another thing"}, {})))
         self.assertFalse(result["same_title"])
-        texts = " ".join(item["text"] for item in result["insights"])
-        self.assertIn("B hit 4 tool errors", texts)
-        self.assertNotIn("matching titles", texts)
 
-    def test_single_session_has_no_insights(self):
+    def test_single_session_has_no_best(self):
         result = compare_sessions(self.entries(("a", {}, {})))
-        self.assertEqual(result["insights"], [])
         self.assertEqual(result["best"], {})
         self.assertEqual(result["sessions"][0]["letter"], "A")
 
-    def test_quality_caveat_is_always_last(self):
-        result = compare_sessions(self.entries(("a", {}, {}), ("b", {}, {})))
-        self.assertEqual(result["insights"][-1]["kind"], "caveat")
-
 
 class TraceKeyTests(unittest.TestCase):
     def test_keys_are_opaque_and_distinct_for_forks_and_repeated_file_names(self):