Name Thesis as the fifth primitive (iariw-2026)

slideshows/iariw-2026/SPEAKER-NOTES.md

@@ -40,7 +40,7 @@ modeling groups' publication records.
 | 15:09 | 2 · Chronicle | 32–35 (skip 35) | 3 |
 | 15:12 | 3 · Microcosm | 36–43 (skip 42) | 6 |
 | 15:18 | 4 · PolicyEngine | 44–49 (skip 48) | 5 |
-| 15:23 | 5 · Together | 50–54 | 5 |
+| 15:23 | 5 · Thesis / together | 50–54 | 5 |
 | 15:28 | Buffer | — | 2 |
 
 Cut-first if long: Microplex (37), WhoUsesIt (40), and skim Evo3/Evo5. Static screenshot
@@ -97,15 +97,15 @@ Five primitives, each a project, each with a feedback loop:
 1 rules → Axiom (loop: oracles, zero unexplained mismatches) · 2 official statistics →
 Chronicle (loop: completeness + lineage) · 3 the world at micro level → Microcosm (loop:
 published calibration error, surveys held out) · 4 a model → PolicyEngine (loop: cross-engine
-agreement + users) · 5 judgment — expertise, discretion, uncertainty (loop: forecasts that
-resolve — the most important). Landing: the same primitives norm AIs AND human researchers.
+agreement + users) · 5 judgment → Thesis (loop: forecasts that resolve — the most important).
+Landing: the same primitives norm AIs AND human researchers.
 
 ### 10 · Six years under one roof (1.5m)
 The history: PolicyEngine since 2021 built the first four together — encoded US/UK rules (by
 hand, increasingly AI), microdata imputed + calibrated to official statistics, the simulation
 engine, the web app (still there, no longer the center). The realization behind the split:
 each layer deserves its own name, users, and feedback loop — Axiom, Chronicle, Microcosm,
-with PolicyEngine as the model that composes them.
+PolicyEngine as the model that composes them, and Thesis scoring the forecasts.
 
 ### 11 · Five primitives, five demos (30s)
 The roadmap. "Let's walk them."
@@ -223,9 +223,11 @@ time.
 
 ### 52 · The loop that matters most (2m)
 Every primitive has its gauge; the one that ranks them all is whether forecasts resolve
-against reality. The measurable question: when the stack covers a new country — Belgium took
-weeks because Microcosm calibrates to official statistics — do calibrated agents predict
-outcomes there better? We intend to score that. And the deeper point for 16:00: policy takes
+against reality — which is what Thesis exists to do: open forecasts of public outcomes,
+every prediction published with its reasoning and graded when the official number lands.
+The measurable question: when the stack covers a new country — Belgium took weeks because
+Microcosm calibrates to official statistics — does Thesis predict outcomes there better?
+We intend to score that. And the deeper point for 16:00: policy takes
 a different shape when baseline conditions change quickly; if you assign real probability to
 that, how does this community arm policymakers to respond?
 

slideshows/iariw-2026/slides/ClosingLoopSlide.tsx

@@ -1,6 +1,7 @@
 import Slide from '@/components/core/Slide';
 import SlideHeader from '@/components/layout/SlideHeader';
 import SlideTitle from '@/components/layout/SlideTitle';
+import Image from '@/components/core/BasePathImage';
 
 export default function ClosingLoopSlide() {
   return (
@@ -9,33 +10,57 @@ export default function ClosingLoopSlide() {
         <SlideTitle>The loop that matters most</SlideTitle>
       </SlideHeader>
 
-      <div className="mt-10 max-w-5xl space-y-8">
-        <p className="text-2xl text-gray-800 leading-relaxed">
-          Every primitive has its gauge &mdash; conformance suites, provenance,
-          calibration error, cross-engine agreement. The one that ranks them
-          all: <span className="font-medium text-pe-teal">do the forecasts
-          resolve against reality?</span>
-        </p>
-
-        <div className="content-card p-7">
-          <div className="slide-tag mb-3">The measurable question</div>
-          <p className="text-xl text-gray-700 leading-relaxed">
-            When this stack covers a new country, do calibrated agents predict
-            outcomes there better? We built the Belgian prototype in weeks
-            because Microcosm calibrates to official statistics rather than
-            rebuilding a survey pipeline &mdash; the same recipe now applies
-            country by country. Whether that improves resolved forecasts is a
-            question we can score, and intend to.
+      <div className="mt-8 grid grid-cols-[1.35fr_0.65fr] gap-8 items-start">
+        <div className="space-y-6">
+          <p className="text-2xl text-gray-800 leading-relaxed">
+            Every primitive has its gauge &mdash; conformance suites,
+            provenance, calibration error, cross-engine agreement. The one
+            that ranks them all:{' '}
+            <span className="font-medium text-pe-teal">
+              do the forecasts resolve against reality?
+            </span>
           </p>
+
+          <div className="content-card p-7">
+            <div className="slide-tag mb-3">The measurable question</div>
+            <p className="text-xl text-gray-700 leading-relaxed">
+              When this stack covers a new country, does Thesis predict
+              outcomes there better? We built the Belgian prototype in weeks
+              because Microcosm calibrates to official statistics rather than
+              rebuilding a survey pipeline &mdash; the same recipe now applies
+              country by country. Whether that improves resolved forecasts is
+              a question we can score, and intend to.
+            </p>
+          </div>
+
+          <div className="accent-block">
+            <p className="text-xl text-gray-800 leading-relaxed">
+              And policy takes a different shape when baseline conditions
+              change quickly. If you assign any real probability to rapid
+              AI-driven change, the question for this community is how policy
+              analysis arms policymakers to respond &mdash; a thread for the
+              roundtable at 16:00.
+            </p>
+          </div>
         </div>
 
-        <div className="accent-block">
-          <p className="text-xl text-gray-800 leading-relaxed">
-            And policy takes a different shape when baseline conditions change
-            quickly. If you assign any real probability to rapid AI-driven
-            change, the question for this community is how policy analysis
-            arms policymakers to respond &mdash; a thread for the roundtable
-            at 16:00.
+        <div className="content-card p-7 flex flex-col items-center text-center gap-4">
+          <Image
+            src="/logos/thesis-institute.png"
+            alt="The Thesis Institute"
+            width={532}
+            height={112}
+            style={{ width: '260px', height: 'auto' }}
+          />
+          <p className="text-lg text-gray-700 leading-relaxed">
+            Open forecasts of public outcomes, scored against reality.
+          </p>
+          <p className="text-base text-gray-500 leading-snug">
+            Every prediction published with its chain of reasoning, graded
+            when the official number lands. The track record is the product.
+          </p>
+          <p className="font-mono text-sm text-pe-teal mt-auto">
+            thesisinstitute.org
           </p>
         </div>
       </div>

slideshows/iariw-2026/slides/FivePrimitivesSlide.tsx

@@ -29,8 +29,8 @@ const primitives = [
   },
   {
     n: '5',
-    need: 'Judgment to put it together',
-    project: 'Expertise, discretion, uncertainty',
+    need: 'Judgment to put it together — expertise, discretion, uncertainty',
+    project: 'Thesis',
     loop: 'Loop: forecasts that resolve against reality — the most important one.',
   },
 ];

slideshows/iariw-2026/slides/HistorySlide.tsx

@@ -65,7 +65,8 @@ export default function HistorySlide() {
           <p className="text-lg leading-relaxed text-white/85">
             So the rules layer became the Axiom Foundation, the fact store
             became Chronicle, the population layer became Microcosm &mdash;
-            and PolicyEngine is the model that composes them.
+            PolicyEngine is the model that composes them, and Thesis scores
+            the forecasts against reality.
           </p>
         </div>
       </div>

slideshows/iariw-2026/slides/OutlineSlide.tsx

@@ -25,8 +25,8 @@ const sections = [
   },
   {
     number: '5',
-    title: 'Together',
-    detail: 'Judgment, uncertainty, and the loop that matters most.',
+    title: 'Thesis',
+    detail: 'Judgment and uncertainty — open forecasts, scored against reality.',
   },
 ];
 

slideshows/iariw-2026/slides/SectionDivider.tsx

@@ -67,7 +67,7 @@ export const EngineDividerSlide = () => (
 export const TogetherDividerSlide = () => (
   <SectionDivider
     number="5"
-    title="Together"
-    subtitle="Judgment, uncertainty, and the loop that matters most."
+    title="Thesis"
+    subtitle="Putting it together — open forecasts of public outcomes, scored against reality."
   />
 );