Bound autonomous learning snapshots

.github/workflows/autonomous-cycle.yml

@@ -35,7 +35,7 @@ jobs:
       - run: node scripts/run-autonomous-cycle.mjs
         env:
           ARENA_URL: https://polymarket-site-eta.vercel.app
-          EXPECTED_BUILD: "114"
+          EXPECTED_BUILD: "115"
           RUNTIME_BRANCH: runtime-state
           RUNTIME_STATE_PATH: runtime/state.json
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

README.md

@@ -20,7 +20,7 @@ market snapshots. During an outage, cycles continue locally; cached entries are
 allowed for 90 minutes, older snapshots become mark-only, and all cached data
 expires after 24 hours.
 
-Build 114 targets five-minute slots with a serialized, self-chained GitHub Actions runtime.
+Build 115 targets five-minute slots with a serialized, self-chained GitHub Actions runtime.
 It continues from the previous agent snapshot and runs the next due paper cycle
 even when no browser is open. After each successful cycle, one repo-scoped
 workflow dispatch waits for the next five-minute boundary; one concurrency group
@@ -48,7 +48,7 @@ Polymarket event key. Related Ethereum or Bitcoin contracts cannot create
 several simultaneous copies of one move, and outcomes from the same underlying
 three-hour shock window count as one learner event.
 
-Each Build 114 cycle also scans the 1,000 most-active Polymarket events for
+Each Build 115 cycle also scans the 1,000 most-active Polymarket events for
 complete negative-risk bundles and logically nested threshold or deadline
 pairs. Gamma's market-specific fee flag replaces the old blanket 0.5-cent fee
 reserve for markets declared fee-free. The closest 60 structures are then
@@ -64,7 +64,7 @@ The Suggestions view stores scan, depth, fee, actionable, and closest
 executable-margin counts so an empty lane is evidence rather than an ambiguous
 failure.
 
-Build 114 retains the directional learner's exact-fee policy, which replaced the blanket half-cent cost with
+Build 115 retains the directional learner's exact-fee policy, which replaced the blanket half-cent cost with
 the market's Gamma fee schedule at both the entry and future checkpoint, plus a
 separate half-cent round-trip slippage allowance. Fee-free markets pay only the
 slippage allowance; an unavailable fee schedule gets a conservative four-cent
@@ -80,7 +80,7 @@ directional evidence under fee policy 2 and keeps those lanes observation-only
 until current independent 24-hour and 72-hour cohorts pass the existing
 promotion gate.
 
-Build 114 retains Build 100's retirement of the old 3-6 day resolution-window
+Build 115 retains Build 100's retirement of the old 3-6 day resolution-window
 capital permission. That audit clustered confidence by event but still averaged
 several correlated contracts inside each event, while production could choose
 only one. The corrected replay chooses the highest-volume eligible contract per event and
@@ -200,7 +200,7 @@ evidence proves an edge.
 
 Run `npm run evaluate:sports-favorites` for the retired pregame favorite audit.
 Its 12-hour, 60%-75% cohort had positive point estimates but did not establish a
-reliable confidence bound, so Build 114 no longer allocates capital to that rule.
+reliable confidence bound, so Build 115 no longer allocates capital to that rule.
 
 Strategy 62 instead uses the exact-fee settlement calibration's three-day Sports
 NO cohort as a bounded paper exploration lane. The corrected 5,000-market run

index.html

@@ -341,7 +341,7 @@ footer{margin-top:34px;padding-top:22px;border-top:1px solid var(--border);color
   <nav class="topnav">
     <div class="brand">
       <div class="logo">🏆</div>
-      <div><div class="brand-name">Polymarket Arena</div><div class="brand-sub">10 agents · 5 core + 5 aggressive</div><div class="build-badge">Adaptive strategy 62 · self-chained runtime · build 114</div></div>
+      <div><div class="brand-name">Polymarket Arena</div><div class="brand-sub">10 agents · 5 core + 5 aggressive</div><div class="build-badge">Adaptive strategy 62 · bounded learning state · build 115</div></div>
     </div>
     <div class="tabs" id="tabs">
       <button class="tab" data-tab="overview">Overview</button>
@@ -363,7 +363,7 @@ footer{margin-top:34px;padding-top:22px;border-top:1px solid var(--border);color
   <div class="personal-banner" id="personalBanner">
     <b>Personal research mode.</b> This copy is for your own analysis, paper tracking, and manual trade research only. It does not pool money, onboard investors, custody funds, bypass eligibility rules, or place orders without your manual approval.
   </div>
-  <div class="live-build-banner"><b>Build 114 active:</b> The public paper agents target five-minute slots with one serialized, self-chained GitHub Actions successor after every successful cycle and cron as recovery. Each fresh cloud browser reconstructs a bounded offline cache from the shared suggestions and open positions, allowing 90-minute cached entries and 24-hour mark-only continuity without persisting private account data; live-depth bundles still require a live connection. Each live cycle scans the 500 most-active eligible markets and 1,000 most-active events, carries exact Gamma fee schedules through cloud and offline caches, and checks complete bundles against live depth. Strategy 62 adds a bounded sports-contest NO exploration lane: one highest-priced NO contract per real contest near the three-day checkpoint, exact entry fee plus slippage, 0.5% initial size, 3% total cap, and automatic forward promotion or suspension. Intact guaranteed bundles are conservatively held at cost in portfolio accounting, while modeled profit remains unbooked until settlement. This point-positive historical pattern is not a proven edge. Phones and computers display the same read-only autonomous state. Profits are not guaranteed.</div>
+  <div class="live-build-banner"><b>Build 115 active:</b> The public paper agents target five-minute slots with one serialized, self-chained GitHub Actions successor after every successful cycle and cron as recovery. Public forward-learning records retain every field required for future grading, fee-aware returns, cohort calibration, and agent eligibility while dropping duplicated diagnostics, keeping the autonomous state below GitHub's transport ceiling as evidence accumulates. Each fresh cloud browser reconstructs a bounded offline cache from the shared suggestions and open positions, allowing 90-minute cached entries and 24-hour mark-only continuity without persisting private account data; live-depth bundles still require a live connection. Each live cycle scans the 500 most-active eligible markets and 1,000 most-active events, carries exact Gamma fee schedules through cloud and offline caches, and checks complete bundles against live depth. Strategy 62 adds a bounded sports-contest NO exploration lane: one highest-priced NO contract per real contest near the three-day checkpoint, exact entry fee plus slippage, 0.5% initial size, 3% total cap, and automatic forward promotion or suspension. Intact guaranteed bundles are conservatively held at cost in portfolio accounting, while modeled profit remains unbooked until settlement. This point-positive historical pattern is not a proven edge. Phones and computers display the same read-only autonomous state. Profits are not guaranteed.</div>
 
   <!-- ============ OVERVIEW ============ -->
   <section class="tabpanel" data-tab="overview">
@@ -745,7 +745,7 @@ footer{margin-top:34px;padding-top:22px;border-top:1px solid var(--border);color
   </section>
 
   <footer>
-    Build 114 · Adaptive strategy 62 · Serialized self-chained five-minute slots with cron recovery · Restart-safe offline runtime · Exact-fee directional learning · Conservative intact-bundle accounting · Contest-deduplicated sports exploration · Live depth, fee, and size-verified bundle audit · Corrected one-decision-per-event settlement audit · Autonomous runtime 2 · Offline runtime 3 · Maker research 3 · Agent learning 3 · Paper trading only · Live prices from Polymarket's public Gamma and CLOB APIs · Not financial advice ·
+    Build 115 · Adaptive strategy 62 · Serialized self-chained five-minute slots with cron recovery · Bounded forward-learning transport · Restart-safe offline runtime · Exact-fee directional learning · Conservative intact-bundle accounting · Contest-deduplicated sports exploration · Live depth, fee, and size-verified bundle audit · Corrected one-decision-per-event settlement audit · Autonomous runtime 2 · Offline runtime 3 · Maker research 3 · Agent learning 3 · Paper trading only · Live prices from Polymarket's public Gamma and CLOB APIs · Not financial advice ·
     <a class="market-link" href="https://github.com/theodore-song/polymarket-analyst" target="_blank" rel="noopener">Source on GitHub</a>
   </footer>
 </div>
@@ -774,7 +774,7 @@ const POLITICS_TREND_MIN_HOLD_HOURS = 72;
 const EXIT_CONFIRM_HOURS = 6;
 const AGENTS_KEY = "pma_agents_v2";
 const SUG_KEY = "pma_suggestions_v5";
-const BUILD_VERSION = 114;
+const BUILD_VERSION = 115;
 const AGENT_LEARNING_VERSION = 3;
 const SUGGESTION_ENGINE_VERSION = 62;
 const MAKER_STRATEGY_VERSION = 3;
@@ -943,6 +943,7 @@ const SIGNAL_EVAL_TOLERANCE_HOURS = 6;
 const SIGNAL_EVAL_HOURS = SIGNAL_EVAL_HORIZONS[0];
 const SIGNAL_LEDGER_PENDING_LIMIT = 600;
 const SIGNAL_LEDGER_OUTCOME_LIMIT = 1000;
+const PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES = 875000;
 const SIGNAL_LEDGER_RETRY_HOURS = SIGNAL_EVAL_HORIZONS[SIGNAL_EVAL_HORIZONS.length-1]+SIGNAL_EVAL_TOLERANCE_HOURS;
 const SIGNAL_LEDGER_DUE_FETCH_LIMIT = 80;
 const currentCycleHour = () => {const p=partsInLocalTime();return `${p.year}-${p.month}-${p.day}T${p.hour}:${p.minute}|s${SUGGESTION_ENGINE_VERSION}`;};
@@ -1967,11 +1968,29 @@ function compactSuggestionsForSync(payload,limits=SYNC_LIMITS){
 }
 function compactPublicSuggestionsForSync(payload){
   const compact=compactSuggestionsForSync(payload,PUBLIC_RUNTIME_LIMITS);
-  compact.suggestions=(compact.suggestions||[]).map(suggestion=>Object.assign({},suggestion,{
+  compact.suggestions=(compact.suggestions||[]).map(suggestion=>{
-    drivers:Array.isArray(suggestion.drivers)?suggestion.drivers.slice(0,2).map(driver=>String(driver).slice(0,80)):suggestion.drivers,
+    const out=Object.assign({},suggestion,{
-  }));
+      drivers:Array.isArray(suggestion.drivers)?suggestion.drivers.slice(0,1).map(driver=>String(driver).slice(0,80)):suggestion.drivers,
+      rationale:typeof suggestion.rationale==="string"?suggestion.rationale.slice(0,240):suggestion.rationale,
+    });
+    if(!out.trade_ready){delete out.clob_yes;delete out.clob_no;}
+    return out;
+  });
   return compact;
 }
+const PUBLIC_SIGNAL_FIELDS=Object.freeze(["key","market_id","event_key","observed_at","graded_horizons","side","entry_price",
+  "signal_type","quality","category","days_to_resolution","trade_ready_at_observation","fees_enabled","fee_schedule",
+  "eligible_agent_ids","signal_policy_version","strategy_version","evaluated_at","target_horizon_hours","horizon_hours",
+  "future_price","gross_return","estimated_cost_per_share","estimated_fee_per_share","estimated_slippage_per_share",
+  "estimated_cost_return","return"]);
+function compactPublicSignalObservation(item){
+  const out={};PUBLIC_SIGNAL_FIELDS.forEach(key=>{if(item&&item[key]!==undefined)out[key]=item[key];});return out;
+}
+function compactPublicSignalLedger(ledger){
+  return {pending:(ledger&&ledger.pending||[]).slice(-SIGNAL_LEDGER_PENDING_LIMIT).map(compactPublicSignalObservation),
+    outcomes:(ledger&&ledger.outcomes||[]).slice(-SIGNAL_LEDGER_OUTCOME_LIMIT).map(compactPublicSignalObservation),
+    expired_ungraded:Number(ledger&&ledger.expired_ungraded||0)};
+}
 function compactPaperStoreForSync(store,limits=SYNC_LIMITS){
   if(!store||typeof store!=="object"||!store.accounts)return store;
   const accounts={};
@@ -2074,19 +2093,49 @@ function prepareCycleSuggestions(suggestions,runMode,entriesAllowed){
 function collectSyncItems(){const items={};SYNC_KEYS.forEach(k=>{const v=localStorage.getItem(k);if(v!=null)items[k]=v;});return compactSyncItems(items);}
 function collectPublicRuntimeItems(){
   const st=compactAgentStateForSync(loadState(),PUBLIC_RUNTIME_LIMITS);
+  st.signal_ledger=compactPublicSignalLedger(st.signal_ledger);
   AGENTS.forEach(agent=>{if(st.agents&&st.agents[agent.id])st.agents[agent.id].lastDecision=compactDecisionForPublic(st.agents[agent.id].lastDecision);});
   st.pending_sync=false;
   st.runtime_managed=true;
   const suggestions=compactPublicSuggestionsForSync(loadSuggestions());
   return {[AGENTS_KEY]:JSON.stringify(st),[SUG_KEY]:JSON.stringify(suggestions)};
 }
+function runtimeSnapshotByteLength(snapshot){return new TextEncoder().encode(JSON.stringify(snapshot)).length;}
+function fitPublicRuntimeSnapshot(snapshot){
+  const st=stateFromSyncItems(snapshot.items),ledger=st&&st.signal_ledger;
+  if(!ledger)return snapshot;
+  const pending=Array.isArray(ledger.pending)?ledger.pending:[],outcomes=Array.isArray(ledger.outcomes)?ledger.outcomes:[];
+  snapshot.summary.signal_ledger={pending_retained:pending.length,pending_total:pending.length,
+    outcomes_retained:outcomes.length,outcomes_total:outcomes.length,byte_budget:PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES};
+  if(runtimeSnapshotByteLength(snapshot)<=PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES)return snapshot;
+  ledger.outcomes=[];
+  snapshot.items[AGENTS_KEY]=JSON.stringify(st);
+  if(runtimeSnapshotByteLength(snapshot)>PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES&&pending.length){
+    let low=0,high=pending.length;
+    while(low<high){
+      const mid=Math.ceil((low+high)/2);ledger.pending=pending.slice(0,mid);snapshot.items[AGENTS_KEY]=JSON.stringify(st);
+      if(runtimeSnapshotByteLength(snapshot)<=PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES)low=mid;else high=mid-1;
+    }
+    ledger.pending=pending.slice(0,low);
+  }
+  let low=0,high=outcomes.length;
+  while(low<high){
+    const mid=Math.ceil((low+high)/2);ledger.outcomes=outcomes.slice(-mid);snapshot.items[AGENTS_KEY]=JSON.stringify(st);
+    if(runtimeSnapshotByteLength(snapshot)<=PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES)low=mid;else high=mid-1;
+  }
+  ledger.outcomes=low?outcomes.slice(-low):[];snapshot.items[AGENTS_KEY]=JSON.stringify(st);
+  snapshot.summary.signal_ledger.pending_retained=ledger.pending.length;
+  snapshot.summary.signal_ledger.outcomes_retained=ledger.outcomes.length;
+  return snapshot;
+}
 function buildPublicRuntimeSnapshot(){
   const st=loadState();
-  return {schema_version:1,build_version:BUILD_VERSION,strategy_version:SUGGESTION_ENGINE_VERSION,
+  const snapshot={schema_version:1,build_version:BUILD_VERSION,strategy_version:SUGGESTION_ENGINE_VERSION,
     generated_at:nowIso(),updated_at:nowIso(),last_cycle_hour:st.last_cycle_hour||null,
     source:"github-actions",read_only:true,items:collectPublicRuntimeItems(),
     summary:{agents:AGENTS.map(cfg=>{const p=st.agents[cfg.id],eq=equity(p);return {id:cfg.id,equity:+eq.toFixed(2),
       return_pct:+((eq/STARTING_BALANCE-1)*100).toFixed(3),cash:+Number(p.cash||0).toFixed(2),positions:(p.positions||[]).length};})}};
+  return fitPublicRuntimeSnapshot(snapshot);
 }
 function applyPublicRuntimeSnapshot(snapshot){
   if(!snapshot||typeof snapshot!=="object"||!snapshot.items||typeof snapshot.items!=="object")return false;
@@ -6609,6 +6658,14 @@ function runEngineSelfTest(){
   const ledgerState={signal_ledger:{pending:[{key:"ledger-test",market_id:"ledger-test",observed_at:hoursAgo(25),side:"YES",entry_price:0.40,
     signal_type:"trend",quality:"confirmed",category:"Politics"}],outcomes:[]}};
   updateSignalLedger(ledgerState,[market({id:"ledger-test",yes_price:0.50,no_price:0.50})],[]);
+  const compactedLedgerState={signal_ledger:compactPublicSignalLedger({pending:[{key:"ledger-compacted",market_id:"ledger-compacted",
+    event_key:"compacted-event",observed_at:hoursAgo(6.5),graded_horizons:[],side:"YES",entry_price:0.40,signal_type:"trend",
+    quality:"confirmed",category:"Politics",days_to_resolution:30,trade_ready_at_observation:true,fees_enabled:false,
+    fee_schedule:{rate:0,exponent:1,takerOnly:true},eligible_agent_ids:["value"],signal_policy_version:DIRECTIONAL_SIGNAL_POLICY_VERSION,
+    strategy_version:SUGGESTION_ENGINE_VERSION,conviction:99,signal_confidence:0.99,evidence_score:0.99}],outcomes:[]})};
+  updateSignalLedger(compactedLedgerState,[market({id:"ledger-compacted",yes_price:0.50,no_price:0.50})],[]);
+  const compactedValueCalibration=buildSignalCalibration(compactedLedgerState.signal_ledger,"value");
+  const compactedMomentumCalibration=buildSignalCalibration(compactedLedgerState.signal_ledger,"momentum");
   const feeFreeSignalCost=signalExecutionCost({fees_enabled:false,fee_schedule:{rate:0,exponent:1}},0.50,0.60);
   const exactFeeSignalCost=signalExecutionCost({fees_enabled:true,fee_schedule:{rate:0.07,exponent:1}},0.50,0.60);
   const unknownFeeSignalCost=signalExecutionCost({},0.50,0.60);
@@ -7196,6 +7253,11 @@ function runEngineSelfTest(){
       underObservedCohortsAndEventsSampleFirst:coveragePriority.map(x=>x.market_id).join(",")==="new-sports,new-politics,popular-repeat",
       firstHorizonGradesWithoutLookahead:ledgerState.signal_ledger.pending.length===1&&ledgerState.signal_ledger.pending[0].graded_horizons.includes(24)
         &&ledgerState.signal_ledger.outcomes.length===1&&ledgerState.signal_ledger.outcomes[0].target_horizon_hours===24&&ledgerState.signal_ledger.outcomes[0].return===0.2375,
+      publicCompactionPreservesForwardGrading:compactedLedgerState.signal_ledger.pending.length===1
+        &&compactedLedgerState.signal_ledger.pending[0].graded_horizons.includes(6)
+        &&compactedLedgerState.signal_ledger.outcomes.length===1&&compactedLedgerState.signal_ledger.outcomes[0].target_horizon_hours===6,
+      publicCompactionPreservesAgentCalibration:compactedValueCalibration.samples===1&&compactedValueCalibration.pending===1
+        &&compactedMomentumCalibration.samples===0&&compactedMomentumCalibration.pending===0,
       secondHorizonGradesAndCompletes:secondHorizonState.signal_ledger.pending.length===0&&secondHorizonState.signal_ledger.outcomes.length===1
         &&secondHorizonState.signal_ledger.outcomes[0].target_horizon_hours===72,
       holdsSignalsUntilPolicyHorizon:earlyLedgerState.signal_ledger.pending.length===1&&earlyLedgerState.signal_ledger.outcomes.length===0,

scripts/run-autonomous-cycle.mjs

@@ -128,7 +128,7 @@ async function main() {
   const branch = process.env.RUNTIME_BRANCH || "runtime-state";
   const pathname = process.env.RUNTIME_STATE_PATH || "runtime/state.json";
   const arenaUrl = (process.env.ARENA_URL || "https://polymarket-site-eta.vercel.app").replace(/\/$/, "");
-  const expectedBuild = Number(required("EXPECTED_BUILD", "114"));
+  const expectedBuild = Number(required("EXPECTED_BUILD", "115"));
   await ensureRuntimeBranch(repository, branch);
   const prior = await readRuntimeFile(repository, branch, pathname);
   if (prior.snapshot) validateRuntimeSnapshot(prior.snapshot, 0, { allowIncomplete: true });

scripts/test-autonomous-runtime.mjs

@@ -10,13 +10,22 @@ const resolutionAudit = JSON.parse(fs.readFileSync(new URL("../research/resoluti
 const sportsContestAudit = JSON.parse(fs.readFileSync(new URL("../research/sports-contest-no-exploration-audit.json", import.meta.url), "utf8"));
 const build = Number(index.match(/const BUILD_VERSION = (\d+);/)?.[1]);
 
-assert.equal(build, 114);
+assert.equal(build, 115);
 assert.deepEqual([...ALLOWED_RUNTIME_KEYS].sort(), ["pma_agents_v2", "pma_suggestions_v5"]);
 assert.match(index, /function collectPublicRuntimeItems\(\)/);
 assert.match(index, /const PUBLIC_RUNTIME_KEYS=Object\.freeze\(\[AGENTS_KEY,SUG_KEY\]\)/);
 assert.match(index, /suggestions:300/);
 assert.match(index, /function compactPublicSuggestionsForSync\(payload\)/);
-assert.match(index, /drivers\.slice\(0,2\)/);
+assert.match(index, /drivers\.slice\(0,1\)/);
+assert.match(index, /if\(!out\.trade_ready\)\{delete out\.clob_yes;delete out\.clob_no;\}/);
+assert.match(index, /function compactPublicSignalLedger\(ledger\)/);
+assert.match(index, /st\.signal_ledger=compactPublicSignalLedger\(st\.signal_ledger\)/);
+assert.match(index, /const PUBLIC_RUNTIME_SNAPSHOT_BUDGET_BYTES = 875000;/);
+assert.match(index, /function fitPublicRuntimeSnapshot\(snapshot\)/);
+assert.match(index, /ledger\.outcomes=low\?outcomes\.slice\(-low\):\[\]/);
+assert.match(index, /return fitPublicRuntimeSnapshot\(snapshot\)/);
+assert.match(index, /publicCompactionPreservesForwardGrading:/);
+assert.match(index, /publicCompactionPreservesAgentCalibration:/);
 assert.match(index, /function buildRuntimeFallbackMarketCache\(stored,state\)/);
 assert.match(index, /buildRuntimeFallbackMarketCache\(stored,st\)/);
 assert.match(index, /window\.PMA_AUTOMATION=Object\.freeze/);
@@ -31,7 +40,7 @@ assert.match(api, /Buffer\.from\(file\.content/);
 assert.match(api, /searchParams\.set\("runtime", `\$\{Date\.now\(\)\}/);
 assert.match(workflow, /cron: "2,7,12,17,22,27,32,37,42,47,52,57 \* \* \* \*"/);
 assert.match(workflow, /contents: write/);
-assert.match(workflow, /EXPECTED_BUILD: "114"/);
+assert.match(workflow, /EXPECTED_BUILD: "115"/);
 assert.match(workflow, /actions: write/);
 assert.match(workflow, /cancel-in-progress: false/);
 assert.match(workflow, /next=\$\(\( \(now \/ 300 \+ 1\) \* 300 \+ 15 \)\)/);

sw.js

@@ -1,4 +1,4 @@
-const CACHE_NAME = "polymarket-arena-build-114";
+const CACHE_NAME = "polymarket-arena-build-115";
 const APP_SHELL = ["/", "/index.html", "/personal.html", "/cycle-worker.js"];
 
 self.addEventListener("install", event => {