fix(dashboard): headline = share of ALL paid spend, not just measured rows

The 'share of total bill saved' headline divided savings by
baseline_input_weighted, which only accumulated on rows where the
count_tokens probe succeeded AND the request had upstream usage. Every
passthrough turn, every probe-failed turn, and every flap-polluted
untransformed turn was excluded from the denominator. That's a
cherry-pick: it answers 'did pixelpipe help on the rows where it ran'
instead of 'did pixelpipe move my real bill'.

Track allActualInputWeighted / allOutputWeighted ungated on probe
success (any request with a usage block contributes), expose
saved_pct_of_all_spend on /proxy-stats, and rewrite the headline card
to use it. Numerator is still measured-rows-only because that's the
only place we have a counterfactual baseline; you can't fabricate
savings for rows where the probe failed. But the denominator now
matches the user's actual bill.

The legacy saved_pct_of_total_bill stays on the wire for back-compat
(marked deprecated in types.ts). The headline card no longer reads it.

Card flips color: positive % stays green, negative % goes red, because
flap-driven net losses are real and should be visible.

src/dashboard.ts

@@ -159,6 +159,23 @@ interface Totals {
    *  doesn't touch output). Without this the headline ignores half the
    *  bill on output-heavy sessions. */
   outputWeighted: number;
+  /** Sum of weighted actual input tokens across ALL requests that had a
+   *  usage block, regardless of whether the probe succeeded — i.e. every
+   *  request the user actually paid for, measured or not, transformed or
+   *  passthrough. Forms the honest denominator for "share of total spend
+   *  saved": you can only credit savings against the slice where pixelpipe
+   *  ran AND we measured, but you have to divide by the full bill (passthrough
+   *  + probe-failed + unmeasured + measured) to answer "did pixelpipe move
+   *  my real bill" instead of "did pixelpipe help on rows where it ran". */
+  allActualInputWeighted: number;
+  /** Sum of output_tokens × OUTPUT_TOKEN_RATE across the same all-rows set
+   *  as allActualInputWeighted. Output is on both sides of the savings ratio
+   *  at its actual 5× rate; numerator stays measured-rows-only because saved
+   *  cancels output (proxy doesn't touch it). */
+  allOutputWeighted: number;
+  /** Count of requests that contributed to allActualInputWeighted (had a
+   *  usage block). Lets the UI annotate "N of M paid requests". */
+  allUsageRequests: number;
   /** Sum of ground-truth output character counts from the SSE/JSON scanner
    *  (see `OutputMeasurement` in proxy.ts). These three accumulators are
    *  independent of Anthropic's `usage.output_tokens` — they let the operator
@@ -226,6 +243,9 @@ export class DashboardState {
     actualInputWeighted: 0,
     baselineInputWeighted: 0,
     outputWeighted: 0,
+    allActualInputWeighted: 0,
+    allOutputWeighted: 0,
+    allUsageRequests: 0,
     textCharsMeasured: 0,
     thinkingCharsMeasured: 0,
     toolUseCharsMeasured: 0,
@@ -377,6 +397,15 @@ export class DashboardState {
       this.totals.actualInputWeighted += actualInputEff;
       this.totals.outputWeighted += outputEquiv;
     }
+    // All-rows spend, ungated on the probe. The "share of total bill saved"
+    // headline divides measured-rows savings into this, so passthrough rows,
+    // probe-failed rows, and uncompressed rows all dilute the headline the
+    // same way they dilute the user's real bill.
+    if (haveUsage) {
+      this.totals.allActualInputWeighted += actualInputEff;
+      this.totals.allOutputWeighted += outputEquiv;
+      this.totals.allUsageRequests += 1;
+    }
 
     // Measurement totals are independent of usage/baseline gating — they
     // accumulate whenever the scanner produced numbers. The scanner sets
@@ -505,6 +534,18 @@ export class DashboardState {
     const baselineTotal = baseline + output;
     const actualTotal = actual + output;
     const pctTotal = baselineTotal > 0 ? (saved / baselineTotal) * 100 : 0;
+
+    // Share-of-all-spend: honest denominator. The numerator can only credit
+    // savings against rows where we have a probe baseline (otherwise it's
+    // estimation), but the denominator MUST include every request the user
+    // actually paid for — including passthrough rows, probe-failed rows,
+    // and untransformed turns the gate said no to. Otherwise the headline
+    // answers "did pixelpipe help on the rows where it ran" instead of
+    // "did pixelpipe move my real bill". The first is a cherry-pick.
+    const allActual = this.totals.allActualInputWeighted;
+    const allOutput = this.totals.allOutputWeighted;
+    const allSpend = allActual + allOutput;
+    const pctAllSpend = allSpend > 0 ? (saved / allSpend) * 100 : 0;
     const uptimeSec = Date.now() / 1000 - this.totals.startedAt;
     const payload = {
       requests: this.totals.requests,
@@ -517,6 +558,15 @@ export class DashboardState {
       saved_pct: round1(pctInput),
       saved_pct_input_only: round1(pctInput),
       saved_pct_of_total_bill: round1(pctTotal),
+      // Honest "share of total bill saved" — measured-rows numerator over
+      // ALL paid requests in the denominator (compressed + passthrough +
+      // probe-failed). This is the number users actually want when they
+      // ask "is pixelpipe helping". Negative when flap-pollution from
+      // passthrough turns exceeds the collapse win on measured turns.
+      saved_pct_of_all_spend: round1(pctAllSpend),
+      all_actual_input_weighted: Math.round(allActual),
+      all_output_weighted: Math.round(allOutput),
+      all_usage_requests: this.totals.allUsageRequests,
       saved_usd: round4((saved * ASSUMED_INPUT_USD_PER_MTOK) / 1e6),
       output_weighted: Math.round(output),
       baseline_token_equivalent: Math.round(baselineTotal),

src/dashboard/components/StatsHeader.svelte

@@ -53,35 +53,37 @@
 
   $: pctMath = s && pa
     ? `<div><span class="k">formula:</span> <span class="v">` +
-      ` share_of_bill = saved / (baseline_input + output × ` +
+      ` share_of_spend = saved / (all_actual_input + all_output × ` +
       (pa.output_multiplier ?? 5) +
       `)</span></div>` +
-      `<div><span class="k">why include output:</span> <span class="v">` +
-      `Anthropic's weekly meter counts input + output × 5, the proxy only moves input</span></div>` +
+      `<div><span class="k">why this denominator:</span> <span class="v">` +
+      `the question is "did pixelpipe move my real bill", not "did pixelpipe help on the slice it ran on". ` +
+      `The denominator is every paid request — compressed, passthrough, probe-failed — so passthrough rows ` +
+      `and flap-polluted turns count against the proxy the same way they count against the user.</span></div>` +
       `<div style="height:6px"></div>` +
-      row('saved', s.saved_input_tokens, '(input savings - proxy doesn\'t touch output)') +
-      row('baseline_input', s.baseline_input_weighted, '(cache-aware counterfactual)') +
+      row('saved', s.saved_input_tokens, '(measured-rows numerator; cache-aware)') +
+      row('all_actual_input', s.all_actual_input_weighted, '(every paid request, weighted)') +
       row(
-        'output',
-        ` × ` + (pa.output_multiplier ?? 5),
-        s.output_weighted + ' (weighted output tokens)',
+        'all_output × ' + (pa.output_multiplier ?? 5),
+        s.all_output_weighted,
+        '(every paid request, output × ' + (pa.output_multiplier ?? 5) + ')',
       ) +
       row(
-        'baseline_total',
-        s.baseline_input_weighted + s.output_weighted,
-        `<span class="op">=</span> baseline_input + output`,
+        'all_spend_total',
+        s.all_actual_input_weighted + s.all_output_weighted,
+        `<span class="op">=</span> all_actual_input + all_output`,
       ) +
       row(
-        'share_of_bill',
-        (s.saved_pct_of_total_bill || 0).toFixed(1) + '%',
-        `<span class="op">=</span> saved / baseline_total × 100`,
+        'share_of_spend',
+        (s.saved_pct_of_all_spend || 0).toFixed(1) + '%',
+        `<span class="op">=</span> saved / all_spend_total × 100`,
       ) +
       row(
-        'input-only %',
-        (s.saved_pct_input_only || 0).toFixed(1) + '%',
-        '(sub-line: saved / baseline_input × 100 - output excluded)',
+        'all_usage_requests',
+        s.all_usage_requests,
+        '(denominator request count - compressed + passthrough + probe-failed)',
       ) +
-      `<span class="src">measured - no estimation</span>`
+      `<span class="src">measured numerator, all-rows denominator - no cherry-pick</span>`
     : '';
 
   $: tokeqMath = s && pa
@@ -171,11 +173,10 @@
     {/if}
   </div>
   <div class="card">
-    <div class="label">share of total bill saved</div>
-    <div class="value pos">{(s?.saved_pct_of_total_bill ?? 0).toFixed(1)}%</div>
+    <div class="label">share of spend saved</div>
+    <div class="value" class:pos={(s?.saved_pct_of_all_spend ?? 0) >= 0} class:neg={(s?.saved_pct_of_all_spend ?? 0) < 0}>{(s?.saved_pct_of_all_spend ?? 0).toFixed(1)}%</div>
     <div class="small">
-      ÷ (input + 5×output) - output billed at 5× input rate · input-only:
-      <span class="pos">{(s?.saved_pct_input_only ?? 0).toFixed(1)}%</span>
+      ÷ all paid requests ({numFmt(s?.all_usage_requests)} req · compressed + passthrough + probe-failed) · negative = pixelpipe added cost
     </div>
     {#if s && pa}
       <details class="math">
@@ -236,6 +237,9 @@
   .value.pos {
     color: #3fb950;
   }
+  .value.neg {
+    color: #f85149;
+  }
   .small {
     font-size: 11px;
     color: #6e7681;

src/dashboard/types.ts

@@ -15,7 +15,15 @@ export interface StatsPayload {
   /** Back-compat duplicate of `saved_pct_input_only`. */
   saved_pct: number;
   saved_pct_input_only: number;
+  /** DEPRECATED — denominator was filtered to measured-rows-only, which
+   *  cherry-picks the wins. Kept on the wire for back-compat. */
   saved_pct_of_total_bill: number;
+  /** Honest "share of total bill saved": measured-rows savings ÷ ALL paid
+   *  requests in the window (compressed + passthrough + probe-failed). */
+  saved_pct_of_all_spend: number;
+  all_actual_input_weighted: number;
+  all_output_weighted: number;
+  all_usage_requests: number;
   saved_usd: number;
   output_weighted: number;
   baseline_token_equivalent: number;