verifiable-thinking-mcp 0.6.0 → 0.6.1

package/dist/index.js

@@ -343,7 +343,7 @@ Verb: "${E}" (base: ${I.toFixed(2)} → boosted: ${C.toFixed(2)})
 Domain: ${W} (weight: ${V.toFixed(2)})
 Intensity: ${w.join(", ")||"none"} (modifier: ${L.toFixed(2)}x)
 
-Calculation: ${C.toFixed(2)} × ${V.toFixed(2)} × ${L.toFixed(2)} = ${d.toFixed(3)}`;return{score:d,tier:zA,confidence:{level:r,inGrayZone:O,boundaryDistance:m},explanation:{verb_type:E,verb_score:C,domain_detected:W,domain_weight:V,intensity_signals:w,intensity_modifier:L},reasoning:$.trim()}}var fcA={min_significant_drop:0.15,min_significant_recovery:0.15,unresolved_threshold:0.3,min_steps:3,overconfident_threshold:0.85,overconfident_max_variance:0.05,cliff_drop_threshold:0.3,declining_final_threshold:0.5};function v_(A,Q={}){let B={...fcA,...Q};if(A.length<B.min_steps)return{drift_score:0,unresolved:!1,min_confidence:A[0]?.verification?.confidence??0.5,min_step:A[0]?.step_number??1,max_drop:0,recovery:0,has_revision_after_drop:!1,pattern:"insufficient",explanation:`Insufficient steps for drift analysis (${A.length} < ${B.min_steps})`,suggestion:null};let E=A.map((zA)=>zA.verification?.confidence??0.5),I=A.map((zA)=>zA.step_number),C=E[0],D=0,U=E[0],Y=0,F=0,g=E[0];for(let zA=1;zA<E.length;zA++){let KA=E[zA];if(KA<C)C=KA,D=zA;if(KA>U)U=KA,Y=zA;if(KA>g)g=KA;else{let kA=g-KA;if(kA>F)F=kA}}let N=E[E.length-1],J=N-C,G=A.slice(D+1).some((zA)=>zA.revises_step!==void 0),M=hcA(E,D,Y,F,J,B),H=Math.max(1,E.length-1-D),X;if(M==="v_shaped"){let zA=F,KA=F*J/H;X=Math.max(zA,KA)}else X=F*0.5;let W=Math.min(1,Math.max(0,X)),V=M==="v_shaped",L=M==="stable_overconfident",w=M==="cliff",z=F>=B.min_significant_drop,Z=J>=B.min_significant_recovery,K=V&&z&&Z&&!G&&W>=B.unresolved_threshold,R=E.length>=2?E[E.length-2]-E[E.length-1]:0,j=w&&R>=B.cliff_drop_threshold,q=M==="declining"&&N<B.declining_final_threshold,y=K||L||j||q,t=L||j||q?Math.max(W,0.4):W,u=ucA(M,F,J,D,I,G,C),d=y?mcA(I[D],F,M,C):null;return{drift_score:t,unresolved:y,min_confidence:C,min_step:I[D],max_drop:F,recovery:J,has_revision_after_drop:G,pattern:M,explanation:u,suggestion:d}}function hcA(A,Q,B,E,I,C){let D=A.length,U=Math.max(...A)-Math.min(...A);if(Q>0&&Q<D-1&&E>=C.min_significant_drop&&I>=C.min_significant_recovery)return"v_shaped";if(Q>=D-1&&E>=C.min_significant_drop&&D>=2){let M=A[D-2]-A[D-1],H=0;for(let W=1;W<D-1;W++)H+=Math.abs(A[W]-A[W-1]);let X=D>2?H/(D-2):0;if(M>=C.min_significant_drop&&M>=X*2)return"cliff"}if(Math.min(...A)>=C.overconfident_threshold&&U<=C.overconfident_max_variance)return"stable_overconfident";if(U<0.1)return"stable";let g=0;for(let M=1;M<D;M++)if(A[M]<A[M-1])g++;if(g>=(D-1)*0.7)return"declining";let N=0;for(let M=1;M<D;M++)if(A[M]>A[M-1])N++;if(N>=(D-1)*0.7)return"improving";let J=0,G=0;for(let M=1;M<D;M++){let H=Math.sign(A[M]-A[M-1]);if(H!==0&&H!==G)J++,G=H}if(J>=3)return"oscillating";return"stable"}function ucA(A,Q,B,E,I,C,D){let U=(Q*100).toFixed(0),Y=(B*100).toFixed(0),F=I[E];switch(A){case"stable":return"Confidence remained stable throughout reasoning chain.";case"stable_overconfident":return`⚠️ Stable high confidence (≥${((D??0.85)*100).toFixed(0)}%) throughout chain. On complex/trap questions, consistent high confidence without doubt often correlates with incorrect answers.`;case"declining":if(D!==void 0&&D<0.5)return`⚠️ Confidence declined steadily to ${(D*100).toFixed(0)}% (${U}% total drop). Ending with low confidence suggests unresolved uncertainty.`;return`Confidence declined steadily (${U}% total drop). This may indicate increasing uncertainty or problem difficulty.`;case"improving":return`Confidence improved throughout reasoning (${Y}% increase). Good progressive understanding.`;case"v_shaped":if(C)return`V-shaped confidence pattern detected: ${U}% drop at step ${F}, then ${Y}% recovery. Revision step present - uncertainty was addressed.`;else return`⚠️ V-shaped confidence pattern: ${U}% drop at step ${F}, then ${Y}% recovery WITHOUT revision. The reasoning may have "pushed through" uncertainty without addressing it.`;case"oscillating":return"Confidence oscillated throughout reasoning. Multiple uncertainty points encountered.";case"cliff":return`Confidence dropped sharply at the end (${U}% drop). Possible error or contradiction detected late in reasoning.`;case"insufficient":return"Not enough steps for meaningful drift analysis.";default:return"Confidence pattern analyzed."}}function mcA(A,Q,B,E){if(B==="stable_overconfident")return`High confidence (${((E??0.85)*100).toFixed(0)}%+) throughout suggests possible overconfidence. Consider: Is this a trick question? Have you verified assumptions? Adding a self-check step could help catch errors.`;if(B==="cliff")return`Sharp confidence drop (${(Q*100).toFixed(0)}%) at the final step suggests an error or contradiction was detected late. Consider: What caused this doubt? Should you revise earlier steps before concluding?`;if(B==="declining")return`Confidence declined to ${((E??0.5)*100).toFixed(0)}% by the end. The reasoning chain ended with significant doubt. Consider: What's causing the uncertainty? Is the approach valid? Should you try a different method?`;if(Q>=0.3)return`Consider revising from step ${A} where confidence dropped significantly. The recovery without explicit revision suggests the uncertainty was not properly addressed.`;else return`Review step ${A} where confidence was lowest. Adding explicit reasoning about why confidence recovered could strengthen the chain.`}var jYA=/(?:let\s+)?([a-zA-Z_][a-zA-Z0-9_]*)\s*[:=]=?\s*(-?\d+(?:\.\d+)?)/g,T_={always:["never","sometimes","rarely"],never:["always","sometimes","often"],all:["none","some","few"],none:["all","some","many"],every:["no","some","few"],must:["cannot","might","may not"],cannot:["must","can","might"],true:["false"],false:["true"],impossible:["possible","certain","likely"],certain:["impossible","uncertain","unlikely"]},k_={positive:["negative","zero"],negative:["positive","zero"],increasing:["decreasing","constant"],decreasing:["increasing","constant"],greater:["less","equal"],less:["greater","equal"],above:["below","at"],below:["above","at"]};function lcA(A,Q){let B=new Map,E;jYA.lastIndex=0;while((E=jYA.exec(A))!==null){let I=E[1]?.toLowerCase(),C=E[2];if(!I||!C)continue;let D=Math.max(0,E.index-20),U=Math.min(A.length,E.index+E[0].length+20),Y=A.slice(D,U);B.set(I,{step:Q,value:C,context:Y})}return B}function ccA(A,Q){let B=new Map,E=A.toLowerCase();for(let[I,C]of Object.entries(T_))if(new RegExp(`\\b${I}\\b`,"gi").test(E)){let U=E.indexOf(I),Y=Math.max(0,U-30),F=Math.min(A.length,U+I.length+30),g=A.slice(Y,F),N=`logical:${I}`;B.set(N,{step:Q,value:I,context:g})}return B}function pcA(A,Q){let B=new Map,E=A.toLowerCase();for(let[I,C]of Object.entries(k_))if(new RegExp(`\\b${I}\\b`,"gi").test(E)){let U=E.indexOf(I),Y=Math.max(0,U-30),F=Math.min(A.length,U+I.length+30),g=A.slice(Y,F),N=g.match(/\b(the\s+)?([a-zA-Z_][a-zA-Z0-9_]*)\b/i),J=N?N[2]||"value":"value",G=`sign:${I}`;B.set(G,{step:Q,value:I,context:`${J}: ${g}`})}return B}function icA(A,Q,B){if(A.match(/^[a-z_][a-z0-9_]*$/i)){if(Q.value!==B.value)return{type:"value_reassignment",description:`Variable '${A}' was ${Q.value} in step ${Q.step}, now ${B.value}`,subject:A,original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.9}}if(A.startsWith("logical:")){if(T_[Q.value]?.includes(B.value))return{type:"logical_conflict",description:`Logical conflict: '${Q.value}' in step ${Q.step} vs '${B.value}' in step ${B.step}`,subject:A.split(":")[2]||"claim",original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.85}}if(A.startsWith("sign:")){if(k_[Q.value]?.includes(B.value)){let I=A.split(":")[1]||"value";return{type:"sign_flip",description:`Sign flip for '${I}': ${Q.value} → ${B.value}`,subject:I,original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.8}}}return null}function dcA(A){if(A.length<2)return{has_contradictions:!1,contradictions:[],steps_analyzed:A.length};let Q=[],B=new Map,E=new Map,I=new Map;for(let{step:C,thought:D}of A){let U=lcA(D,C),Y=ccA(D,C),F=pcA(D,C);for(let[g,N]of U){let J=B.get(g);if(J&&J.step!==C){let G=icA(g,J,N);if(G)Q.push(G)}B.set(g,N)}for(let[g,N]of Y){let J=T_[N.value];if(J)for(let G of J){let M=E.get(G);if(M&&M.step!==C)Q.push({type:"logical_conflict",description:`Logical conflict: '${M.value}' in step ${M.step} vs '${N.value}' in step ${C}`,subject:"claim",original_step:M.step,original_value:M.value,conflicting_step:C,conflicting_value:N.value,confidence:0.85})}E.set(N.value,N)}for(let[g,N]of F){let J=k_[N.value];if(J)for(let G of J){let M=I.get(G);if(M&&M.step!==C){let X=N.context.match(/^([^:]+):/)?.[1]??"value";Q.push({type:"sign_flip",description:`Sign flip for '${X}': ${M.value} → ${N.value}`,subject:X,original_step:M.step,original_value:M.value,conflicting_step:C,conflicting_value:N.value,confidence:0.8})}}I.set(N.value,N)}}return{has_contradictions:Q.length>0,contradictions:Q,steps_analyzed:A.length}}function qYA(A,Q){return dcA([...Q,A]).contradictions.filter((E)=>E.conflicting_step===A.step)}function C8(A){if(/\d+\s*[+\-*/^=]\s*\d+|equation|solve|derivative|integral|sum\s+of/i.test(A))return"math";if(/function|class|return|const|let|var|def\s|import\s|async|await|=>|->|fn\s|impl\s/i.test(A))return"code";if(/if\s+.+\s+then|therefore|implies|hence|thus|conclude|premise|valid|invalid/i.test(A))return"logic";return"general"}var ncA=[/\b(?:therefore|thus|hence|consequently)\b.*\b(?:true|correct|valid|proven|confirmed)\b/i,/\b(?:this\s+)?(?:confirms?|proves?|shows?|demonstrates?)\s+(?:that\s+)?(?:the\s+)?hypothesis\b/i,/\b(?:QED|Q\.E\.D\.|quod\s+erat\s+demonstrandum)\b/i,/\b(?:we\s+have\s+shown|we\s+conclude|this\s+establishes)\b/i,/\bhypothesis\s+(?:is\s+)?(?:true|correct|valid|confirmed)\b/i,/\bas\s+(?:we\s+)?(?:hypothesized|expected|predicted)\b/i,/\bsuccess(?:fully)?\s+(?:verified|confirmed|proven)\b/i],ocA=[/\b(?:contradiction|contradicts?|inconsistent)\b/i,/\b(?:impossible|cannot\s+be|can't\s+be)\b/i,/\b(?:disprove[ds]?|refute[ds]?|falsif(?:y|ied))\b/i,/\bhypothesis\s+(?:is\s+)?(?:false|incorrect|invalid|wrong|fails?)\b/i,/\b(?:this\s+)?(?:fails?|violates?|breaks?)\s+(?:the\s+)?(?:assumption|hypothesis)\b/i,/\b(?:counterexample|counter-example)\b/i,/\bnot\s+(?:true|valid|correct|possible)\b/i,/\b(?:rejected?|abandon|discard)\s+(?:the\s+)?hypothesis\b/i],rcA=[/\b(?:inconclusive|undetermined|unclear|uncertain)\b/i,/\b(?:need|require)s?\s+(?:more|further|additional)\s+(?:evidence|proof|analysis)\b/i,/\b(?:cannot\s+(?:yet\s+)?(?:determine|conclude|decide))\b/i,/\b(?:insufficient\s+(?:evidence|data|information))\b/i];function __(A,Q){for(let B of Q){let E=A.match(B);if(E)return E[0]}return null}function SYA(A,Q){let B=A.toLowerCase(),I=Q.toLowerCase().split(/\s+/).filter((D)=>D.length>3);return I.filter((D)=>B.includes(D)).length>=I.length*0.5}function PYA(A,Q,B,E){let I={resolved:!1,outcome:null,confidence:0,resolved_at_step:null,evidence:null,hypothesis:Q,success_criteria:B,suggestion:"Continue testing the hypothesis."},C=__(A,ocA);if(C)return{...I,resolved:!0,outcome:"refuted",confidence:0.9,resolved_at_step:E,evidence:C,suggestion:"Hypothesis refuted. Consider abandoning this branch or revising the hypothesis."};let D=__(A,rcA);if(D)return{...I,resolved:!0,outcome:"inconclusive",confidence:0.7,resolved_at_step:E,evidence:D,suggestion:"Hypothesis inconclusive. Gather more evidence or reformulate."};let U=__(A,ncA);if(U){let Y=0.85;if(B&&SYA(A,B))Y=0.95;return{...I,resolved:!0,outcome:"confirmed",confidence:Y,resolved_at_step:E,evidence:U,suggestion:"Hypothesis confirmed. Consider merging findings back to main branch."}}if(B&&SYA(A,B))return{...I,resolved:!1,outcome:null,confidence:0.6,evidence:"Success criteria keywords detected",suggestion:"Success criteria may be satisfied. Verify and explicitly confirm or refute."};return I}var $YA=wA.object({operation:wA.enum(["step","navigate","branch","revise","complete","augment","override","hint","mistakes","spot_check","challenge"]).describe("Operation to perform"),confidence_threshold:wA.number().min(0).max(1).default(0.8).describe("Chain confidence threshold to suggest completion"),token_budget:wA.number().int().min(100).default(3000).describe("Max tokens before auto-compressing new steps"),warn_at_tokens:wA.number().int().min(100).optional().describe("Warn when cumulative session tokens exceed this threshold (soft limit, cost control)"),hard_limit_tokens:wA.number().int().min(100).optional().describe("Hard stop when cumulative session tokens exceed this threshold. Returns budget_exhausted status and blocks further operations."),thought:wA.string().optional().describe("Current reasoning/analysis (step/branch/revise)"),purpose:wA.enum(["analysis","action","reflection","decision","summary","validation","exploration","hypothesis","correction","planning"]).optional().describe("Step category"),outcome:wA.string().optional().describe("Result or conclusion from this step"),confidence:wA.number().min(0).max(1).optional().describe("Confidence in this step (0-1). Contributes to chain average."),context:wA.string().optional().describe("Prior context or findings"),verify:wA.boolean().optional().describe("Run domain verification. Auto-enabled for chains >3 steps. Set to false to disable."),domain:wA.enum(["math","logic","code","general"]).optional(),local_compute:wA.boolean().default(!1).describe("Try local compute for math"),augment_compute:wA.boolean().default(!0).describe("Auto-inject computed values into thought (default: true)"),compress:wA.boolean().default(!0).describe("Compress thought before storing"),compression_query:wA.string().optional().describe("Query for context-aware compression"),max_step_tokens:wA.number().int().min(10).optional().describe("Max tokens for this step. Rejects if exceeded (default: no limit)"),force_large:wA.boolean().default(!1).describe("Allow step even if it exceeds max_step_tokens"),preconditions:wA.array(wA.string()).optional().describe("Assumptions that MUST be true for this step (e.g., 'x > 0', 'file exists')"),view:wA.enum(["history","branches","step","path"]).optional().describe("What to view: history (all steps), branches (list), step (specific), path (lineage)"),step_id:wA.number().int().min(1).optional().describe("Step number to view"),branch_id:wA.string().optional().describe("Filter history by branch"),limit:wA.number().int().min(1).max(50).default(10).describe("Max steps to return"),from_step:wA.number().int().min(1).optional().describe("Step to branch from (default: current)"),branch_name:wA.string().optional().describe("Human-readable branch name"),hypothesis:wA.string().optional().describe("Falsifiable hypothesis this branch will test (e.g., 'Assume X is prime')"),success_criteria:wA.string().optional().describe("What observation proves/disproves this hypothesis"),target_step:wA.number().int().min(1).optional().describe("Step number to revise"),reason:wA.string().optional().describe("Why revising this step / Why overriding verification"),summary:wA.string().optional().describe("Final summary/conclusion"),final_answer:wA.string().optional().describe("The answer/result"),question:wA.string().optional().describe("Original question. On step: enables trap priming and stores for auto spot-check. On complete: enables spot-check."),text:wA.string().optional().describe("Text containing math expressions to compute and inject (augment/mistakes)"),system_context:wA.string().optional().describe("System prompt context for domain filtering"),store_as_step:wA.boolean().default(!1).describe("Store augmented result as a reasoning step"),acknowledge:wA.boolean().optional().describe("Confirm you understand verification failed but want to proceed"),failed_step:wA.number().int().min(1).optional().describe("Step number that failed verification"),expression:wA.string().optional().describe("Math expression to simplify. Omit to continue from previous hint in session."),reveal_count:wA.number().int().min(1).optional().describe("Number of steps to reveal. Omit to auto-increment when continuing."),cumulative:wA.boolean().default(!0).describe("Show all steps up to reveal_count (true) or just the nth step (false)"),reset:wA.boolean().default(!1).describe("Reset hint state and start from beginning"),answer:wA.string().optional().describe("The proposed answer to check for trap patterns"),challenge_type:wA.enum(["assumption_inversion","edge_case","premise_check","steelman_counter","all"]).optional().describe("Type of challenge to generate (default: all)"),target_claim:wA.string().optional().describe("Specific claim to challenge (optional - if omitted, extracts claims from steps)")});function acA(A){let Q=A.match(/\d+(?:\.\d+)?/g);if(!Q)return[];return Q.map((B)=>parseFloat(B))}function tcA(A){let Q=A.match(/(\d+(?:\.\d+)?)/);if(!Q||Q[1]===void 0)return null;return parseFloat(Q[1])}function scA(A){let Q=new Map;for(let B of A)Q.set(B,(Q.get(B)||0)+1);return Q}function ecA(A){let Q=A.toLowerCase(),B=[],E=0;if(/(?:total|sum|together|cost).*?(?:more|less)\s*than/i.test(Q))E+=0.8,B.push("additive_system");if(/doubles?|triples?|exponential/i.test(Q)&&/half|quarter|fraction|percent/i.test(Q))E+=0.8,B.push("nonlinear_growth");if(/(\d+)\s*(?:machines?|workers?|people|printers?).*?(?:minutes?|hours?|days?|seconds?)/i.test(Q))E+=0.6,B.push("rate_pattern");if(/average\s*speed|speed.*average/i.test(Q)&&/(?:round\s*trip|return|back|there and back)/i.test(Q))E+=0.9,B.push("harmonic_mean");if(/(?:row|consecutive|straight|times)/i.test(Q)&&/(?:probability|chance|likely|odds)/i.test(Q))E+=0.7,B.push("independence");if(/(?:minimum|least|fewest)/i.test(Q)&&/(?:guarantee|ensure|certain|must)/i.test(Q))E+=0.7,B.push("pigeonhole");if(/(?:test|positive|negative)/i.test(Q)&&/(?:1\s*in\s*\d+|rare|uncommon|\d+%\s*(?:of|have))/i.test(Q))E+=0.75,B.push("base_rate");if(/\d+!/i.test(Q)&&/(?:zero|factor|digit)/i.test(Q))E+=0.7,B.push("factorial_counting");if(/clock/i.test(Q)&&/(?:overlap|coincide|same position)/i.test(Q))E+=0.8,B.push("clock_overlap");if(/(?:given|if|knowing|after)/i.test(Q)&&/(?:probability|chance|what.*odds)/i.test(Q))E+=0.6,B.push("conditional_probability");if(/(?:more likely|which.*probable|what.*probability)/i.test(Q)&&/(?:and|both|also)/i.test(Q)&&/(?:bank teller|feminist|active|personality|description)/i.test(Q))E+=0.8,B.push("conjunction_fallacy");if(/(?:door|box|curtain|envelope)/i.test(Q)&&/(?:switch|stay|change|keep)/i.test(Q)&&/(?:reveal|open|show)/i.test(Q))E+=0.85,B.push("monty_hall");if(/(?:estimate|guess|how (?:many|much|long))/i.test(Q)&&/(?:spin|wheel|number|digit|wrote|shown)/i.test(Q))E+=0.6,B.push("anchoring");if(/(?:already|spent|invested|paid|cost)/i.test(Q)&&/(?:should|continue|keep|stop|quit|abandon|walk away)/i.test(Q))E+=0.75,B.push("sunk_cost");if((/(?:save|saved|survive|lives?)/i.test(Q)||/(?:die|death|lost|killed)/i.test(Q))&&/(?:program|option|choice|treatment|plan) [ab]/i.test(Q))E+=0.7,B.push("framing_effect");return{required:E>=0.6,score:Math.min(1,E),categories:B}}function D8(A,Q){let B=acA(A),E=tcA(Q),I=A.toLowerCase(),C=FpA(I,Q)||gpA(I,E,Q)||NpA(I,Q)||JpA(I,Q);if(C)return C;if(E===null||B.length===0)return vYA();return ApA(I,B,E)||QpA(I,B,E)||BpA(I,B,E)||IpA(I,B,E)||EpA(I,E)||CpA(I,B,E)||DpA(I,B,E)||UpA(I,B,E)||YpA(I,E)||vYA()}function vYA(){return{passed:!0,warning:null,hint:null,trapType:null,confidence:0.5}}function aB(A,Q,B,E){return{passed:!1,warning:Q,hint:B,trapType:A,confidence:E}}function ApA(A,Q,B){if(!/(?:total|sum|together|cost).*?(?:more|less)\s*than/i.test(A))return null;if(Q.length<2)return null;let E=[...Q].sort((g,N)=>N-g),I=E[0],C=E[1];if(I===void 0||C===void 0)return null;let D=I-C,U=(I-C)/2,Y=Math.abs(B-D)<0.01||Math.abs(B-D*100)<0.5||Math.abs(B/100-D)<0.01,F=Math.abs(B-U)<0.01||Math.abs(B-U*100)<0.5||Math.abs(B/100-U)<0.01;if(Y&&!F)return aB("additive_system",`Potential trap: ${B} might be (${I} - ${C}) without solving the system`,`This is a system: x + y = ${I}, x - y = ${C}. Solve: y = (${I} - ${C}) / 2 = ${U}`,0.85);return null}function QpA(A,Q,B){if(!/doubles?|triples?/i.test(A))return null;if(!/half|quarter/i.test(A))return null;let E=Q.filter((C)=>C>5&&Number.isInteger(C));if(E.length===0)return null;let I=Math.max(...E);if(Math.abs(B-I/2)<0.5)return aB("nonlinear_growth",`Potential trap: ${B} is ${I}/2, but exponential growth doesn't work linearly`,`If something doubles each period and is full at time ${I}, it was half-full at time ${I-1}`,0.9);return null}function BpA(A,Q,B){if(!/machines?|workers?|people|printers?/i.test(A))return null;if(!/minutes?|hours?|seconds?/i.test(A))return null;let E=scA(Q),I=null;for(let[D,U]of E)if(U>=3){I=D;break}if(I===null)return null;let C=Q.filter((D)=>D!==I&&(E.get(D)??0)>=2);for(let D of C)if(Math.abs(B-D)<0.1)return aB("rate_pattern",`Potential trap: ${B} assumes time scales with quantity`,`If ${I} machines make ${I} widgets in ${I} min, each machine makes 1 widget in ${I} min. More machines = same time, more output.`,0.85);return null}function IpA(A,Q,B){if(!/average\s*speed|speed.*average/i.test(A))return null;if(!/(?:round\s*trip|return|back)/i.test(A))return null;let E=Q.filter((Y)=>Y>0&&Y<200);if(E.length<2)return null;let I=E[0],C=E[1];if(I===void 0||C===void 0||I===C)return null;let D=(I+C)/2,U=2*I*C/(I+C);if(Math.abs(B-D)<0.5&&Math.abs(B-U)>1)return aB("harmonic_mean",`Potential trap: ${B} is the arithmetic mean (${I}+${C})/2`,`For round trips over fixed distance, use harmonic mean: 2×${I}×${C}/(${I}+${C}) = ${U.toFixed(1)}`,0.9);return null}function EpA(A,Q){if(!/coin|dice?|flip|roll/i.test(A))return null;if(!/(?:row|consecutive|straight|times)/i.test(A))return null;if(!/(?:probability|chance|likely)/i.test(A))return null;if(!(Math.abs(Q-50)<2||Math.abs(Q-0.5)<0.02)&&Q>0&&Q<100)return aB("independence","Potential gambler's fallacy: previous outcomes don't affect independent events","Each flip/roll is independent. Past results don't change future probability.",0.8);return null}function CpA(A,Q,B){if(!/(?:minimum|least|fewest)/i.test(A))return null;if(!/(?:guarantee|ensure|certain)/i.test(A))return null;if(!/(?:match|pair|same)/i.test(A))return null;if(B===2)return aB("pigeonhole","Potential trap: 2 items could all be different","Pigeonhole principle: with N categories, you need N+1 items to guarantee a match.",0.85);let E=Math.max(...Q.filter((I)=>I<100));if(B>E&&E>2)return aB("pigeonhole","Potential trap: you don't need majority, just one more than categories",`With ${E} categories (if that's the count), you need at most ${E+1} to guarantee a match.`,0.7);return null}function DpA(A,Q,B){if(!/(?:test|positive|negative)/i.test(A))return null;if(!/(?:probability|chance)/i.test(A))return null;let E=A.match(/1\s*(?:in|out of)\s*(\d+)/i);if(!E||!E[1])return null;let I=parseInt(E[1],10),C=1/I,U=Q.filter((M)=>M>=90&&M<=100)[0];if(U===void 0)return null;let Y=U/100,F=Y,g=1-Y,N=F*C+g*(1-C),J=F*C/N,G=B>1?B:B*100;if(G>80&&J<0.2)return aB("base_rate",`Potential base rate neglect: ${G.toFixed(0)}% ignores the low prevalence (1 in ${I})`,`Apply Bayes: P(disease|positive) ≈ ${(J*100).toFixed(0)}%, not ${(Y*100).toFixed(0)}%`,0.85);return null}function UpA(A,Q,B){if(!/trailing.*zero|zero.*trailing/i.test(A))return null;if(!/\d+!/i.test(A))return null;let E=A.match(/(\d+)!/);if(!E||!E[1])return null;let I=parseInt(E[1],10),C=0,D=5;while(D<=I)C+=Math.floor(I/D),D*=5;let U=Math.floor(I/5),Y=Math.floor(I/10);if(Math.abs(B-U)<0.5&&U!==C)return aB("factorial_counting",`Potential trap: ${B} only counts single factors of 5`,`Count ALL factors of 5: ⌊n/5⌋ + ⌊n/25⌋ + ⌊n/125⌋ + ... = ${C}`,0.85);if(Math.abs(B-Y)<0.5&&Y!==C)return aB("factorial_counting","Potential trap: trailing zeros come from factors of 5 (not 10)",`Since 2s are abundant, count factors of 5: ${C}`,0.8);return null}function YpA(A,Q){if(!/clock/i.test(A))return null;if(!/(?:overlap|coincide)/i.test(A))return null;if(!/(?:how many|times)/i.test(A))return null;if(/12\s*hours?/i.test(A)&&Math.abs(Q-12)<0.5)return aB("clock_overlap","Potential trap: hands overlap 11 times in 12 hours, not 12","The 12:00 overlap is shared. Hands overlap every ~65.45 minutes → 11 times per 12 hours.",0.9);if(/24\s*hours?/i.test(A)&&Math.abs(Q-24)<0.5)return aB("clock_overlap","Potential trap: hands overlap 22 times in 24 hours, not 24","11 overlaps per 12-hour period × 2 = 22 total.",0.9);return null}function FpA(A,Q){if(!/(?:more likely|which.*probable|what.*probability)/i.test(A))return null;let B=Q.toLowerCase();if(!/(?:and|both|as well)/i.test(A))return null;if(/\b[bB]\b/.test(Q)||/(?:and|both|as well)/i.test(B))return aB("conjunction_fallacy","Potential conjunction fallacy: P(A and B) ≤ P(A) always","A conjunction cannot be more probable than either of its parts. The more specific option is LESS likely.",0.85);return null}function gpA(A,Q,B){if(!(/monty\s*hall/i.test(A)||/(?:door|box|curtain)/i.test(A)&&/(?:switch|stay|change|keep)/i.test(A)&&/(?:reveal|open|show|goat)/i.test(A)))return null;let I=B.toLowerCase();if(/(?:probability|chance)/i.test(A)&&Q!==null){if(Math.abs(Q-50)<2||Math.abs(Q-0.5)<0.02)return aB("monty_hall","Potential Monty Hall trap: it's NOT 50/50 after a door is revealed","Switching wins 2/3 of the time, staying wins 1/3. The reveal gives you information.",0.9)}if(/(?:should|better|strategy)/i.test(A)){if(/(?:stay|keep|doesn't matter|50.?50|same|either)/i.test(I))return aB("monty_hall","Potential Monty Hall trap: switching is actually the better strategy","Switching wins 2/3 of the time. The host's reveal changes the odds.",0.85)}return null}function NpA(A,Q){if(!/(?:already|spent|invested|paid|cost)/i.test(A))return null;if(!/(?:should|continue|keep|stop|quit|abandon|walk away|finish)/i.test(A))return null;let B=Q.toLowerCase(),E=/(?:already spent|already invested|can't waste|too much invested|come this far|so much into)/i.test(B)||/(?:continue|keep going|finish).*(?:because|since).*(?:spent|invested|paid)/i.test(B),I=/^(?:yes|continue|keep|finish)/i.test(B.trim())&&!/(?:future value|expected return|profitable going forward|worth it regardless)/i.test(B)||/(?:wasted|thrown away|lost|for nothing)/i.test(B);if(E||I)return aB("sunk_cost","Potential sunk cost fallacy: past investment shouldn't influence future decisions","Sunk costs are gone - focus on whether FUTURE benefits justify FUTURE costs. What's already spent is irrelevant.",0.8);return null}function JpA(A,Q){if(!((/(?:save|saved|survive|lives?)/i.test(A)||/(?:die|death|lost|killed)/i.test(A))&&/(?:program|option|choice|treatment|plan) [ab]/i.test(A)))return null;let E=Q.toLowerCase(),I=/(?:save|saved|survive)/i.test(A)&&!/(?:die|death|killed)/i.test(A),C=/(?:die|death|killed)/i.test(A)&&!/(?:save|saved|survive)/i.test(A);if(I||C){if(!/(?:equivalent|same|framing|mathematically|expected value|doesn't matter)/i.test(E)){if(/\b[ab]\b/i.test(E))return aB("framing_effect","Potential framing effect: check if options are mathematically equivalent","The way choices are presented (lives saved vs lives lost) often triggers different intuitive responses to identical expected outcomes. Calculate expected values to decide rationally.",0.7)}}return null}var TYA={minConfidence:0.7,maxCombined:1,excludeTypes:[]};var GpA={additive_system:"⚠️ System of equations detected. Define variables x,y and solve algebraically.",nonlinear_growth:"⚠️ Exponential growth. Work backwards from the end state, not forwards.",rate_pattern:"⚠️ Rate problem. Calculate rate per unit first, then scale.",harmonic_mean:"⚠️ Round trip speed. Use harmonic mean: 2ab/(a+b), not arithmetic.",independence:"⚠️ Independent events. Past outcomes don't affect future probability.",pigeonhole:"⚠️ Guarantee problem. Consider worst case: need categories + 1.",base_rate:"⚠️ Rare condition + test. Apply Bayes' theorem with base rate.",factorial_counting:"⚠️ Factorial zeros. Count factors of 5: ⌊n/5⌋ + ⌊n/25⌋ + ...",clock_overlap:"⚠️ Clock hands overlap 11 times per 12 hours, not 12.",conditional_probability:"⚠️ Conditional probability. Use P(A|B) = P(A∩B)/P(B).",conjunction_fallacy:"⚠️ Conjunction trap. P(A and B) ≤ P(A) always.",monty_hall:"⚠️ Revealed information changes odds. Switching wins 2/3.",anchoring:"⚠️ Ignore irrelevant numbers. Base estimate on actual data only.",sunk_cost:"⚠️ Sunk cost trap. Past spending is irrelevant to future decisions.",framing_effect:"⚠️ Check framing. Calculate expected values for both options."};function kYA(A,Q){let B=typeof Q==="number"?{...TYA,maxCombined:Q}:{...TYA,...Q},E=ecA(A);if(!E.required||E.categories.length===0)return{shouldPrime:!1,trapTypes:[],primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:"no_traps_detected"};if(E.score<B.minConfidence)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`confidence_below_threshold:${E.score.toFixed(2)}<${B.minConfidence}`};let I=E.categories.filter((Y)=>!B.excludeTypes.includes(Y));if(I.length===0)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`all_types_excluded:${E.categories.join(",")}`};let C=I.slice(0,B.maxCombined),D=[];for(let Y of C){let F=GpA[Y];if(F)D.push(F)}if(D.length===0)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`no_prompts_for_types:${C.join(",")}`};let U=null;if(D.length===1)U=D[0]??null;else if(D.length>1)U=D.map((Y,F)=>`${F+1}. ${Y.replace("⚠️ ","")}`).join(`
+Calculation: ${C.toFixed(2)} × ${V.toFixed(2)} × ${L.toFixed(2)} = ${d.toFixed(3)}`;return{score:d,tier:zA,confidence:{level:r,inGrayZone:O,boundaryDistance:m},explanation:{verb_type:E,verb_score:C,domain_detected:W,domain_weight:V,intensity_signals:w,intensity_modifier:L},reasoning:$.trim()}}var fcA={min_significant_drop:0.15,min_significant_recovery:0.15,unresolved_threshold:0.3,min_steps:3,overconfident_threshold:0.85,overconfident_max_variance:0.05,cliff_drop_threshold:0.3,declining_final_threshold:0.5};function v_(A,Q={}){let B={...fcA,...Q};if(A.length<B.min_steps)return{drift_score:0,unresolved:!1,min_confidence:A[0]?.verification?.confidence??0.5,min_step:A[0]?.step_number??1,max_drop:0,recovery:0,has_revision_after_drop:!1,pattern:"insufficient",explanation:`Insufficient steps for drift analysis (${A.length} < ${B.min_steps})`,suggestion:null};let E=A.map((zA)=>zA.verification?.confidence??0.5),I=A.map((zA)=>zA.step_number),C=E[0],D=0,U=E[0],Y=0,F=0,g=E[0];for(let zA=1;zA<E.length;zA++){let KA=E[zA];if(KA<C)C=KA,D=zA;if(KA>U)U=KA,Y=zA;if(KA>g)g=KA;else{let kA=g-KA;if(kA>F)F=kA}}let N=E[E.length-1],J=N-C,G=A.slice(D+1).some((zA)=>zA.revises_step!==void 0),M=hcA(E,D,Y,F,J,B),H=Math.max(1,E.length-1-D),X;if(M==="v_shaped"){let zA=F,KA=F*J/H;X=Math.max(zA,KA)}else X=F*0.5;let W=Math.min(1,Math.max(0,X)),V=M==="v_shaped",L=M==="stable_overconfident",w=M==="cliff",z=F>=B.min_significant_drop,Z=J>=B.min_significant_recovery,K=V&&z&&Z&&!G&&W>=B.unresolved_threshold,R=E.length>=2?E[E.length-2]-E[E.length-1]:0,j=w&&R>=B.cliff_drop_threshold,q=M==="declining"&&N<B.declining_final_threshold,y=K||L||j||q,t=L||j||q?Math.max(W,0.4):W,u=ucA(M,F,J,D,I,G,C),d=y?mcA(I[D],F,M,C):null;return{drift_score:t,unresolved:y,min_confidence:C,min_step:I[D],max_drop:F,recovery:J,has_revision_after_drop:G,pattern:M,explanation:u,suggestion:d}}function hcA(A,Q,B,E,I,C){let D=A.length,U=Math.max(...A)-Math.min(...A);if(Q>0&&Q<D-1&&E>=C.min_significant_drop&&I>=C.min_significant_recovery)return"v_shaped";if(Q>=D-1&&E>=C.min_significant_drop&&D>=2){let M=A[D-2]-A[D-1],H=0;for(let W=1;W<D-1;W++)H+=Math.abs(A[W]-A[W-1]);let X=D>2?H/(D-2):0;if(M>=C.min_significant_drop&&M>=X*2)return"cliff"}if(Math.min(...A)>=C.overconfident_threshold&&U<=C.overconfident_max_variance)return"stable_overconfident";if(U<0.1)return"stable";let g=0;for(let M=1;M<D;M++)if(A[M]<A[M-1])g++;if(g>=(D-1)*0.7)return"declining";let N=0;for(let M=1;M<D;M++)if(A[M]>A[M-1])N++;if(N>=(D-1)*0.7)return"improving";let J=0,G=0;for(let M=1;M<D;M++){let H=Math.sign(A[M]-A[M-1]);if(H!==0&&H!==G)J++,G=H}if(J>=3)return"oscillating";return"stable"}function ucA(A,Q,B,E,I,C,D){let U=(Q*100).toFixed(0),Y=(B*100).toFixed(0),F=I[E];switch(A){case"stable":return"Confidence remained stable throughout reasoning chain.";case"stable_overconfident":return`⚠️ Stable high confidence (≥${((D??0.85)*100).toFixed(0)}%) throughout chain. On complex/trap questions, consistent high confidence without doubt often correlates with incorrect answers.`;case"declining":if(D!==void 0&&D<0.5)return`⚠️ Confidence declined steadily to ${(D*100).toFixed(0)}% (${U}% total drop). Ending with low confidence suggests unresolved uncertainty.`;return`Confidence declined steadily (${U}% total drop). This may indicate increasing uncertainty or problem difficulty.`;case"improving":return`Confidence improved throughout reasoning (${Y}% increase). Good progressive understanding.`;case"v_shaped":if(C)return`V-shaped confidence pattern detected: ${U}% drop at step ${F}, then ${Y}% recovery. Revision step present - uncertainty was addressed.`;else return`⚠️ V-shaped confidence pattern: ${U}% drop at step ${F}, then ${Y}% recovery WITHOUT revision. The reasoning may have "pushed through" uncertainty without addressing it.`;case"oscillating":return"Confidence oscillated throughout reasoning. Multiple uncertainty points encountered.";case"cliff":return`Confidence dropped sharply at the end (${U}% drop). Possible error or contradiction detected late in reasoning.`;case"insufficient":return"Not enough steps for meaningful drift analysis.";default:return"Confidence pattern analyzed."}}function mcA(A,Q,B,E){if(B==="stable_overconfident")return`High confidence (${((E??0.85)*100).toFixed(0)}%+) throughout suggests possible overconfidence. Consider: Is this a trick question? Have you verified assumptions? Adding a self-check step could help catch errors.`;if(B==="cliff")return`Sharp confidence drop (${(Q*100).toFixed(0)}%) at the final step suggests an error or contradiction was detected late. Consider: What caused this doubt? Should you revise earlier steps before concluding?`;if(B==="declining")return`Confidence declined to ${((E??0.5)*100).toFixed(0)}% by the end. The reasoning chain ended with significant doubt. Consider: What's causing the uncertainty? Is the approach valid? Should you try a different method?`;if(Q>=0.3)return`Consider revising from step ${A} where confidence dropped significantly. The recovery without explicit revision suggests the uncertainty was not properly addressed.`;else return`Review step ${A} where confidence was lowest. Adding explicit reasoning about why confidence recovered could strengthen the chain.`}var jYA=/(?:let\s+)?([a-zA-Z_][a-zA-Z0-9_]*)\s*[:=]=?\s*(-?\d+(?:\.\d+)?)/g,T_={always:["never","sometimes","rarely"],never:["always","sometimes","often"],all:["none","some","few"],none:["all","some","many"],every:["no","some","few"],must:["cannot","might","may not"],cannot:["must","can","might"],true:["false"],false:["true"],impossible:["possible","certain","likely"],certain:["impossible","uncertain","unlikely"]},k_={positive:["negative","zero"],negative:["positive","zero"],increasing:["decreasing","constant"],decreasing:["increasing","constant"],greater:["less","equal"],less:["greater","equal"],above:["below","at"],below:["above","at"]};function lcA(A,Q){let B=new Map,E;jYA.lastIndex=0;while((E=jYA.exec(A))!==null){let I=E[1]?.toLowerCase(),C=E[2];if(!I||!C)continue;let D=Math.max(0,E.index-20),U=Math.min(A.length,E.index+E[0].length+20),Y=A.slice(D,U);B.set(I,{step:Q,value:C,context:Y})}return B}function ccA(A,Q){let B=new Map,E=A.toLowerCase();for(let[I,C]of Object.entries(T_))if(new RegExp(`\\b${I}\\b`,"gi").test(E)){let U=E.indexOf(I),Y=Math.max(0,U-30),F=Math.min(A.length,U+I.length+30),g=A.slice(Y,F),N=`logical:${I}`;B.set(N,{step:Q,value:I,context:g})}return B}function pcA(A,Q){let B=new Map,E=A.toLowerCase();for(let[I,C]of Object.entries(k_))if(new RegExp(`\\b${I}\\b`,"gi").test(E)){let U=E.indexOf(I),Y=Math.max(0,U-30),F=Math.min(A.length,U+I.length+30),g=A.slice(Y,F),N=g.match(/\b(the\s+)?([a-zA-Z_][a-zA-Z0-9_]*)\b/i),J=N?N[2]||"value":"value",G=`sign:${I}`;B.set(G,{step:Q,value:I,context:`${J}: ${g}`})}return B}function icA(A,Q,B){if(A.match(/^[a-z_][a-z0-9_]*$/i)){if(Q.value!==B.value)return{type:"value_reassignment",description:`Variable '${A}' was ${Q.value} in step ${Q.step}, now ${B.value}`,subject:A,original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.9}}if(A.startsWith("logical:")){if(T_[Q.value]?.includes(B.value))return{type:"logical_conflict",description:`Logical conflict: '${Q.value}' in step ${Q.step} vs '${B.value}' in step ${B.step}`,subject:A.split(":")[2]||"claim",original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.85}}if(A.startsWith("sign:")){if(k_[Q.value]?.includes(B.value)){let I=A.split(":")[1]||"value";return{type:"sign_flip",description:`Sign flip for '${I}': ${Q.value} → ${B.value}`,subject:I,original_step:Q.step,original_value:Q.value,conflicting_step:B.step,conflicting_value:B.value,confidence:0.8}}}return null}function dcA(A){if(A.length<2)return{has_contradictions:!1,contradictions:[],steps_analyzed:A.length};let Q=[],B=new Map,E=new Map,I=new Map;for(let{step:C,thought:D}of A){let U=lcA(D,C),Y=ccA(D,C),F=pcA(D,C);for(let[g,N]of U){let J=B.get(g);if(J&&J.step!==C){let G=icA(g,J,N);if(G)Q.push(G)}B.set(g,N)}for(let[g,N]of Y){let J=T_[N.value];if(J)for(let G of J){let M=E.get(G);if(M&&M.step!==C)Q.push({type:"logical_conflict",description:`Logical conflict: '${M.value}' in step ${M.step} vs '${N.value}' in step ${C}`,subject:"claim",original_step:M.step,original_value:M.value,conflicting_step:C,conflicting_value:N.value,confidence:0.85})}E.set(N.value,N)}for(let[g,N]of F){let J=k_[N.value];if(J)for(let G of J){let M=I.get(G);if(M&&M.step!==C){let X=N.context.match(/^([^:]+):/)?.[1]??"value";Q.push({type:"sign_flip",description:`Sign flip for '${X}': ${M.value} → ${N.value}`,subject:X,original_step:M.step,original_value:M.value,conflicting_step:C,conflicting_value:N.value,confidence:0.8})}}I.set(N.value,N)}}return{has_contradictions:Q.length>0,contradictions:Q,steps_analyzed:A.length}}function qYA(A,Q){return dcA([...Q,A]).contradictions.filter((E)=>E.conflicting_step===A.step)}function C8(A){if(/\d+\s*[+\-*/^=]\s*\d+|equation|solve|derivative|integral|sum\s+of/i.test(A))return"math";if(/function|class|return|const|let|var|def\s|import\s|async|await|=>|->|fn\s|impl\s/i.test(A))return"code";if(/if\s+.+\s+then|therefore|implies|hence|thus|conclude|premise|valid|invalid/i.test(A))return"logic";return"general"}var ncA=[/\b(?:therefore|thus|hence|consequently)\b.*\b(?:true|correct|valid|proven|confirmed)\b/i,/\b(?:this\s+)?(?:confirms?|proves?|shows?|demonstrates?)\s+(?:that\s+)?(?:the\s+)?hypothesis\b/i,/\b(?:QED|Q\.E\.D\.|quod\s+erat\s+demonstrandum)\b/i,/\b(?:we\s+have\s+shown|we\s+conclude|this\s+establishes)\b/i,/\bhypothesis\s+(?:is\s+)?(?:true|correct|valid|confirmed)\b/i,/\bas\s+(?:we\s+)?(?:hypothesized|expected|predicted)\b/i,/\bsuccess(?:fully)?\s+(?:verified|confirmed|proven)\b/i],ocA=[/\b(?:contradiction|contradicts?|inconsistent)\b/i,/\b(?:impossible|cannot\s+be|can't\s+be)\b/i,/\b(?:disprove[ds]?|refute[ds]?|falsif(?:y|ied))\b/i,/\bhypothesis\s+(?:is\s+)?(?:false|incorrect|invalid|wrong|fails?)\b/i,/\b(?:this\s+)?(?:fails?|violates?|breaks?)\s+(?:the\s+)?(?:assumption|hypothesis)\b/i,/\b(?:counterexample|counter-example)\b/i,/\bnot\s+(?:true|valid|correct|possible)\b/i,/\b(?:rejected?|abandon|discard)\s+(?:the\s+)?hypothesis\b/i],rcA=[/\b(?:inconclusive|undetermined|unclear|uncertain)\b/i,/\b(?:need|require)s?\s+(?:more|further|additional)\s+(?:evidence|proof|analysis)\b/i,/\b(?:cannot\s+(?:yet\s+)?(?:determine|conclude|decide))\b/i,/\b(?:insufficient\s+(?:evidence|data|information))\b/i];function __(A,Q){for(let B of Q){let E=A.match(B);if(E)return E[0]}return null}function SYA(A,Q){let B=A.toLowerCase(),I=Q.toLowerCase().split(/\s+/).filter((D)=>D.length>3);return I.filter((D)=>B.includes(D)).length>=I.length*0.5}function PYA(A,Q,B,E){let I={resolved:!1,outcome:null,confidence:0,resolved_at_step:null,evidence:null,hypothesis:Q,success_criteria:B,suggestion:"Continue testing the hypothesis."},C=__(A,ocA);if(C)return{...I,resolved:!0,outcome:"refuted",confidence:0.9,resolved_at_step:E,evidence:C,suggestion:"Hypothesis refuted. Consider abandoning this branch or revising the hypothesis."};let D=__(A,rcA);if(D)return{...I,resolved:!0,outcome:"inconclusive",confidence:0.7,resolved_at_step:E,evidence:D,suggestion:"Hypothesis inconclusive. Gather more evidence or reformulate."};let U=__(A,ncA);if(U){let Y=0.85;if(B&&SYA(A,B))Y=0.95;return{...I,resolved:!0,outcome:"confirmed",confidence:Y,resolved_at_step:E,evidence:U,suggestion:"Hypothesis confirmed. Consider merging findings back to main branch."}}if(B&&SYA(A,B))return{...I,resolved:!1,outcome:null,confidence:0.6,evidence:"Success criteria keywords detected",suggestion:"Success criteria may be satisfied. Verify and explicitly confirm or refute."};return I}var $YA=wA.object({operation:wA.enum(["step","navigate","branch","revise","complete","augment","override","hint","mistakes","spot_check","challenge"]).describe("Operation to perform"),confidence_threshold:wA.number().min(0).max(1).optional().describe("Chain confidence threshold to suggest completion (default: 0.8)"),token_budget:wA.number().int().min(100).optional().describe("Max tokens before auto-compressing new steps (default: 3000)"),warn_at_tokens:wA.number().int().min(100).optional().describe("Warn when cumulative session tokens exceed this threshold (soft limit, cost control)"),hard_limit_tokens:wA.number().int().min(100).optional().describe("Hard stop when cumulative session tokens exceed this threshold. Returns budget_exhausted status and blocks further operations."),thought:wA.string().optional().describe("Current reasoning/analysis (step/branch/revise)"),purpose:wA.enum(["analysis","action","reflection","decision","summary","validation","exploration","hypothesis","correction","planning"]).optional().describe("Step category"),outcome:wA.string().optional().describe("Result or conclusion from this step"),confidence:wA.number().min(0).max(1).optional().describe("Confidence in this step (0-1). Contributes to chain average."),context:wA.string().optional().describe("Prior context or findings"),verify:wA.boolean().optional().describe("Run domain verification. Auto-enabled for chains >3 steps. Set to false to disable."),domain:wA.enum(["math","logic","code","general"]).optional(),local_compute:wA.boolean().optional().describe("Try local compute for math (default: false)"),augment_compute:wA.boolean().optional().describe("Auto-inject computed values into thought (default: true)"),compress:wA.boolean().optional().describe("Compress thought before storing (default: true)"),compression_query:wA.string().optional().describe("Query for context-aware compression"),max_step_tokens:wA.number().int().min(10).optional().describe("Max tokens for this step. Rejects if exceeded (default: no limit)"),force_large:wA.boolean().optional().describe("Allow step even if it exceeds max_step_tokens (default: false)"),preconditions:wA.array(wA.string()).optional().describe("Assumptions that MUST be true for this step (e.g., 'x > 0', 'file exists')"),view:wA.enum(["history","branches","step","path"]).optional().describe("What to view: history (all steps), branches (list), step (specific), path (lineage)"),step_id:wA.number().int().min(1).optional().describe("Step number to view"),branch_id:wA.string().optional().describe("Filter history by branch"),limit:wA.number().int().min(1).max(50).optional().describe("Max steps to return (default: 10)"),from_step:wA.number().int().min(1).optional().describe("Step to branch from (default: current)"),branch_name:wA.string().optional().describe("Human-readable branch name"),hypothesis:wA.string().optional().describe("Falsifiable hypothesis this branch will test (e.g., 'Assume X is prime')"),success_criteria:wA.string().optional().describe("What observation proves/disproves this hypothesis"),target_step:wA.number().int().min(1).optional().describe("Step number to revise"),reason:wA.string().optional().describe("Why revising this step / Why overriding verification"),summary:wA.string().optional().describe("Final summary/conclusion"),final_answer:wA.string().optional().describe("The answer/result"),question:wA.string().optional().describe("Original question. On step: enables trap priming and stores for auto spot-check. On complete: enables spot-check."),text:wA.string().optional().describe("Text containing math expressions to compute and inject (augment/mistakes)"),system_context:wA.string().optional().describe("System prompt context for domain filtering"),store_as_step:wA.boolean().optional().describe("Store augmented result as a reasoning step (default: false)"),acknowledge:wA.boolean().optional().describe("Confirm you understand verification failed but want to proceed"),failed_step:wA.number().int().min(1).optional().describe("Step number that failed verification"),expression:wA.string().optional().describe("Math expression to simplify. Omit to continue from previous hint in session."),reveal_count:wA.number().int().min(1).optional().describe("Number of steps to reveal. Omit to auto-increment when continuing."),cumulative:wA.boolean().optional().describe("Show all steps up to reveal_count (true) or just the nth step (false). Default: true"),reset:wA.boolean().optional().describe("Reset hint state and start from beginning (default: false)"),answer:wA.string().optional().describe("The proposed answer to check for trap patterns"),challenge_type:wA.enum(["assumption_inversion","edge_case","premise_check","steelman_counter","all"]).optional().describe("Type of challenge to generate (default: all)"),target_claim:wA.string().optional().describe("Specific claim to challenge (optional - if omitted, extracts claims from steps)")});function acA(A){let Q=A.match(/\d+(?:\.\d+)?/g);if(!Q)return[];return Q.map((B)=>parseFloat(B))}function tcA(A){let Q=A.match(/(\d+(?:\.\d+)?)/);if(!Q||Q[1]===void 0)return null;return parseFloat(Q[1])}function scA(A){let Q=new Map;for(let B of A)Q.set(B,(Q.get(B)||0)+1);return Q}function ecA(A){let Q=A.toLowerCase(),B=[],E=0;if(/(?:total|sum|together|cost).*?(?:more|less)\s*than/i.test(Q))E+=0.8,B.push("additive_system");if(/doubles?|triples?|exponential/i.test(Q)&&/half|quarter|fraction|percent/i.test(Q))E+=0.8,B.push("nonlinear_growth");if(/(\d+)\s*(?:machines?|workers?|people|printers?).*?(?:minutes?|hours?|days?|seconds?)/i.test(Q))E+=0.6,B.push("rate_pattern");if(/average\s*speed|speed.*average/i.test(Q)&&/(?:round\s*trip|return|back|there and back)/i.test(Q))E+=0.9,B.push("harmonic_mean");if(/(?:row|consecutive|straight|times)/i.test(Q)&&/(?:probability|chance|likely|odds)/i.test(Q))E+=0.7,B.push("independence");if(/(?:minimum|least|fewest)/i.test(Q)&&/(?:guarantee|ensure|certain|must)/i.test(Q))E+=0.7,B.push("pigeonhole");if(/(?:test|positive|negative)/i.test(Q)&&/(?:1\s*in\s*\d+|rare|uncommon|\d+%\s*(?:of|have))/i.test(Q))E+=0.75,B.push("base_rate");if(/\d+!/i.test(Q)&&/(?:zero|factor|digit)/i.test(Q))E+=0.7,B.push("factorial_counting");if(/clock/i.test(Q)&&/(?:overlap|coincide|same position)/i.test(Q))E+=0.8,B.push("clock_overlap");if(/(?:given|if|knowing|after)/i.test(Q)&&/(?:probability|chance|what.*odds)/i.test(Q))E+=0.6,B.push("conditional_probability");if(/(?:more likely|which.*probable|what.*probability)/i.test(Q)&&/(?:and|both|also)/i.test(Q)&&/(?:bank teller|feminist|active|personality|description)/i.test(Q))E+=0.8,B.push("conjunction_fallacy");if(/(?:door|box|curtain|envelope)/i.test(Q)&&/(?:switch|stay|change|keep)/i.test(Q)&&/(?:reveal|open|show)/i.test(Q))E+=0.85,B.push("monty_hall");if(/(?:estimate|guess|how (?:many|much|long))/i.test(Q)&&/(?:spin|wheel|number|digit|wrote|shown)/i.test(Q))E+=0.6,B.push("anchoring");if(/(?:already|spent|invested|paid|cost)/i.test(Q)&&/(?:should|continue|keep|stop|quit|abandon|walk away)/i.test(Q))E+=0.75,B.push("sunk_cost");if((/(?:save|saved|survive|lives?)/i.test(Q)||/(?:die|death|lost|killed)/i.test(Q))&&/(?:program|option|choice|treatment|plan) [ab]/i.test(Q))E+=0.7,B.push("framing_effect");return{required:E>=0.6,score:Math.min(1,E),categories:B}}function D8(A,Q){let B=acA(A),E=tcA(Q),I=A.toLowerCase(),C=FpA(I,Q)||gpA(I,E,Q)||NpA(I,Q)||JpA(I,Q);if(C)return C;if(E===null||B.length===0)return vYA();return ApA(I,B,E)||QpA(I,B,E)||BpA(I,B,E)||IpA(I,B,E)||EpA(I,E)||CpA(I,B,E)||DpA(I,B,E)||UpA(I,B,E)||YpA(I,E)||vYA()}function vYA(){return{passed:!0,warning:null,hint:null,trapType:null,confidence:0.5}}function aB(A,Q,B,E){return{passed:!1,warning:Q,hint:B,trapType:A,confidence:E}}function ApA(A,Q,B){if(!/(?:total|sum|together|cost).*?(?:more|less)\s*than/i.test(A))return null;if(Q.length<2)return null;let E=[...Q].sort((g,N)=>N-g),I=E[0],C=E[1];if(I===void 0||C===void 0)return null;let D=I-C,U=(I-C)/2,Y=Math.abs(B-D)<0.01||Math.abs(B-D*100)<0.5||Math.abs(B/100-D)<0.01,F=Math.abs(B-U)<0.01||Math.abs(B-U*100)<0.5||Math.abs(B/100-U)<0.01;if(Y&&!F)return aB("additive_system",`Potential trap: ${B} might be (${I} - ${C}) without solving the system`,`This is a system: x + y = ${I}, x - y = ${C}. Solve: y = (${I} - ${C}) / 2 = ${U}`,0.85);return null}function QpA(A,Q,B){if(!/doubles?|triples?/i.test(A))return null;if(!/half|quarter/i.test(A))return null;let E=Q.filter((C)=>C>5&&Number.isInteger(C));if(E.length===0)return null;let I=Math.max(...E);if(Math.abs(B-I/2)<0.5)return aB("nonlinear_growth",`Potential trap: ${B} is ${I}/2, but exponential growth doesn't work linearly`,`If something doubles each period and is full at time ${I}, it was half-full at time ${I-1}`,0.9);return null}function BpA(A,Q,B){if(!/machines?|workers?|people|printers?/i.test(A))return null;if(!/minutes?|hours?|seconds?/i.test(A))return null;let E=scA(Q),I=null;for(let[D,U]of E)if(U>=3){I=D;break}if(I===null)return null;let C=Q.filter((D)=>D!==I&&(E.get(D)??0)>=2);for(let D of C)if(Math.abs(B-D)<0.1)return aB("rate_pattern",`Potential trap: ${B} assumes time scales with quantity`,`If ${I} machines make ${I} widgets in ${I} min, each machine makes 1 widget in ${I} min. More machines = same time, more output.`,0.85);return null}function IpA(A,Q,B){if(!/average\s*speed|speed.*average/i.test(A))return null;if(!/(?:round\s*trip|return|back)/i.test(A))return null;let E=Q.filter((Y)=>Y>0&&Y<200);if(E.length<2)return null;let I=E[0],C=E[1];if(I===void 0||C===void 0||I===C)return null;let D=(I+C)/2,U=2*I*C/(I+C);if(Math.abs(B-D)<0.5&&Math.abs(B-U)>1)return aB("harmonic_mean",`Potential trap: ${B} is the arithmetic mean (${I}+${C})/2`,`For round trips over fixed distance, use harmonic mean: 2×${I}×${C}/(${I}+${C}) = ${U.toFixed(1)}`,0.9);return null}function EpA(A,Q){if(!/coin|dice?|flip|roll/i.test(A))return null;if(!/(?:row|consecutive|straight|times)/i.test(A))return null;if(!/(?:probability|chance|likely)/i.test(A))return null;if(!(Math.abs(Q-50)<2||Math.abs(Q-0.5)<0.02)&&Q>0&&Q<100)return aB("independence","Potential gambler's fallacy: previous outcomes don't affect independent events","Each flip/roll is independent. Past results don't change future probability.",0.8);return null}function CpA(A,Q,B){if(!/(?:minimum|least|fewest)/i.test(A))return null;if(!/(?:guarantee|ensure|certain)/i.test(A))return null;if(!/(?:match|pair|same)/i.test(A))return null;if(B===2)return aB("pigeonhole","Potential trap: 2 items could all be different","Pigeonhole principle: with N categories, you need N+1 items to guarantee a match.",0.85);let E=Math.max(...Q.filter((I)=>I<100));if(B>E&&E>2)return aB("pigeonhole","Potential trap: you don't need majority, just one more than categories",`With ${E} categories (if that's the count), you need at most ${E+1} to guarantee a match.`,0.7);return null}function DpA(A,Q,B){if(!/(?:test|positive|negative)/i.test(A))return null;if(!/(?:probability|chance)/i.test(A))return null;let E=A.match(/1\s*(?:in|out of)\s*(\d+)/i);if(!E||!E[1])return null;let I=parseInt(E[1],10),C=1/I,U=Q.filter((M)=>M>=90&&M<=100)[0];if(U===void 0)return null;let Y=U/100,F=Y,g=1-Y,N=F*C+g*(1-C),J=F*C/N,G=B>1?B:B*100;if(G>80&&J<0.2)return aB("base_rate",`Potential base rate neglect: ${G.toFixed(0)}% ignores the low prevalence (1 in ${I})`,`Apply Bayes: P(disease|positive) ≈ ${(J*100).toFixed(0)}%, not ${(Y*100).toFixed(0)}%`,0.85);return null}function UpA(A,Q,B){if(!/trailing.*zero|zero.*trailing/i.test(A))return null;if(!/\d+!/i.test(A))return null;let E=A.match(/(\d+)!/);if(!E||!E[1])return null;let I=parseInt(E[1],10),C=0,D=5;while(D<=I)C+=Math.floor(I/D),D*=5;let U=Math.floor(I/5),Y=Math.floor(I/10);if(Math.abs(B-U)<0.5&&U!==C)return aB("factorial_counting",`Potential trap: ${B} only counts single factors of 5`,`Count ALL factors of 5: ⌊n/5⌋ + ⌊n/25⌋ + ⌊n/125⌋ + ... = ${C}`,0.85);if(Math.abs(B-Y)<0.5&&Y!==C)return aB("factorial_counting","Potential trap: trailing zeros come from factors of 5 (not 10)",`Since 2s are abundant, count factors of 5: ${C}`,0.8);return null}function YpA(A,Q){if(!/clock/i.test(A))return null;if(!/(?:overlap|coincide)/i.test(A))return null;if(!/(?:how many|times)/i.test(A))return null;if(/12\s*hours?/i.test(A)&&Math.abs(Q-12)<0.5)return aB("clock_overlap","Potential trap: hands overlap 11 times in 12 hours, not 12","The 12:00 overlap is shared. Hands overlap every ~65.45 minutes → 11 times per 12 hours.",0.9);if(/24\s*hours?/i.test(A)&&Math.abs(Q-24)<0.5)return aB("clock_overlap","Potential trap: hands overlap 22 times in 24 hours, not 24","11 overlaps per 12-hour period × 2 = 22 total.",0.9);return null}function FpA(A,Q){if(!/(?:more likely|which.*probable|what.*probability)/i.test(A))return null;let B=Q.toLowerCase();if(!/(?:and|both|as well)/i.test(A))return null;if(/\b[bB]\b/.test(Q)||/(?:and|both|as well)/i.test(B))return aB("conjunction_fallacy","Potential conjunction fallacy: P(A and B) ≤ P(A) always","A conjunction cannot be more probable than either of its parts. The more specific option is LESS likely.",0.85);return null}function gpA(A,Q,B){if(!(/monty\s*hall/i.test(A)||/(?:door|box|curtain)/i.test(A)&&/(?:switch|stay|change|keep)/i.test(A)&&/(?:reveal|open|show|goat)/i.test(A)))return null;let I=B.toLowerCase();if(/(?:probability|chance)/i.test(A)&&Q!==null){if(Math.abs(Q-50)<2||Math.abs(Q-0.5)<0.02)return aB("monty_hall","Potential Monty Hall trap: it's NOT 50/50 after a door is revealed","Switching wins 2/3 of the time, staying wins 1/3. The reveal gives you information.",0.9)}if(/(?:should|better|strategy)/i.test(A)){if(/(?:stay|keep|doesn't matter|50.?50|same|either)/i.test(I))return aB("monty_hall","Potential Monty Hall trap: switching is actually the better strategy","Switching wins 2/3 of the time. The host's reveal changes the odds.",0.85)}return null}function NpA(A,Q){if(!/(?:already|spent|invested|paid|cost)/i.test(A))return null;if(!/(?:should|continue|keep|stop|quit|abandon|walk away|finish)/i.test(A))return null;let B=Q.toLowerCase(),E=/(?:already spent|already invested|can't waste|too much invested|come this far|so much into)/i.test(B)||/(?:continue|keep going|finish).*(?:because|since).*(?:spent|invested|paid)/i.test(B),I=/^(?:yes|continue|keep|finish)/i.test(B.trim())&&!/(?:future value|expected return|profitable going forward|worth it regardless)/i.test(B)||/(?:wasted|thrown away|lost|for nothing)/i.test(B);if(E||I)return aB("sunk_cost","Potential sunk cost fallacy: past investment shouldn't influence future decisions","Sunk costs are gone - focus on whether FUTURE benefits justify FUTURE costs. What's already spent is irrelevant.",0.8);return null}function JpA(A,Q){if(!((/(?:save|saved|survive|lives?)/i.test(A)||/(?:die|death|lost|killed)/i.test(A))&&/(?:program|option|choice|treatment|plan) [ab]/i.test(A)))return null;let E=Q.toLowerCase(),I=/(?:save|saved|survive)/i.test(A)&&!/(?:die|death|killed)/i.test(A),C=/(?:die|death|killed)/i.test(A)&&!/(?:save|saved|survive)/i.test(A);if(I||C){if(!/(?:equivalent|same|framing|mathematically|expected value|doesn't matter)/i.test(E)){if(/\b[ab]\b/i.test(E))return aB("framing_effect","Potential framing effect: check if options are mathematically equivalent","The way choices are presented (lives saved vs lives lost) often triggers different intuitive responses to identical expected outcomes. Calculate expected values to decide rationally.",0.7)}}return null}var TYA={minConfidence:0.7,maxCombined:1,excludeTypes:[]};var GpA={additive_system:"⚠️ System of equations detected. Define variables x,y and solve algebraically.",nonlinear_growth:"⚠️ Exponential growth. Work backwards from the end state, not forwards.",rate_pattern:"⚠️ Rate problem. Calculate rate per unit first, then scale.",harmonic_mean:"⚠️ Round trip speed. Use harmonic mean: 2ab/(a+b), not arithmetic.",independence:"⚠️ Independent events. Past outcomes don't affect future probability.",pigeonhole:"⚠️ Guarantee problem. Consider worst case: need categories + 1.",base_rate:"⚠️ Rare condition + test. Apply Bayes' theorem with base rate.",factorial_counting:"⚠️ Factorial zeros. Count factors of 5: ⌊n/5⌋ + ⌊n/25⌋ + ...",clock_overlap:"⚠️ Clock hands overlap 11 times per 12 hours, not 12.",conditional_probability:"⚠️ Conditional probability. Use P(A|B) = P(A∩B)/P(B).",conjunction_fallacy:"⚠️ Conjunction trap. P(A and B) ≤ P(A) always.",monty_hall:"⚠️ Revealed information changes odds. Switching wins 2/3.",anchoring:"⚠️ Ignore irrelevant numbers. Base estimate on actual data only.",sunk_cost:"⚠️ Sunk cost trap. Past spending is irrelevant to future decisions.",framing_effect:"⚠️ Check framing. Calculate expected values for both options."};function kYA(A,Q){let B=typeof Q==="number"?{...TYA,maxCombined:Q}:{...TYA,...Q},E=ecA(A);if(!E.required||E.categories.length===0)return{shouldPrime:!1,trapTypes:[],primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:"no_traps_detected"};if(E.score<B.minConfidence)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`confidence_below_threshold:${E.score.toFixed(2)}<${B.minConfidence}`};let I=E.categories.filter((Y)=>!B.excludeTypes.includes(Y));if(I.length===0)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`all_types_excluded:${E.categories.join(",")}`};let C=I.slice(0,B.maxCombined),D=[];for(let Y of C){let F=GpA[Y];if(F)D.push(F)}if(D.length===0)return{shouldPrime:!1,trapTypes:E.categories,primedTypes:[],primingPrompt:null,allPrompts:[],confidence:E.score,skippedReason:`no_prompts_for_types:${C.join(",")}`};let U=null;if(D.length===1)U=D[0]??null;else if(D.length>1)U=D.map((Y,F)=>`${F+1}. ${Y.replace("⚠️ ","")}`).join(`
 `);return{shouldPrime:!0,trapTypes:E.categories,primedTypes:C,primingPrompt:U,allPrompts:D,confidence:E.score,skippedReason:null}}var MpA=190,_YA=1e4;function WpA(A){switch(A){case"Low":return 1;case"Moderate":return 2;case"High":return 4;case"Very Hard":return 6;case"Almost Impossible":return 8}}function RD(A,Q,B){let E=WA.getThoughts(A,Q),I=[];for(let D of E)if(D.verification?.confidence!==void 0)I.push(D.verification.confidence);if(B!==void 0)I.push(B);let C=I.length>0?I.reduce((D,U)=>D+U,0)/I.length:0;return{stepConfidence:B,chainConfidence:C,stepsWithConfidence:I.length}}function pg(A,Q,B){if(B)return"complete";if(A>=Q)return"threshold_reached";if(A>=Q*0.8)return"review";return"continue"}function bYA(A,Q){switch(A){case"complete":return"Reasoning chain complete.";case"threshold_reached":return`Confidence ${(Q*100).toFixed(0)}% reached threshold. Consider completing or add one more verification step.`;case"review":return`Confidence ${(Q*100).toFixed(0)}% approaching threshold. Review recent steps for completeness.`;case"continue":return`Continue reasoning. Chain confidence: ${(Q*100).toFixed(0)}%`;case"verification_failed":return"Verification failed. Use revise, branch, or override to continue.";case"budget_exhausted":return"Token budget exhausted. Complete your reasoning or start a new session."}}async function XpA(A,Q,B){let E=WA.getThoughts(A,Q);if(E.length<3)return;let I=v_(E);if(I.pattern==="insufficient")return;if(I.unresolved)await B({type:"text",text:`
 ⚠️ **Early Drift Warning:** ${I.explanation}
 `+(I.suggestion?`   \uD83D\uDCA1 ${I.suggestion}
@@ -368,7 +368,7 @@ Calculation: ${C.toFixed(2)} × ${V.toFixed(2)} × ${L.toFixed(2)} = ${d.toFixed
 
 **Suggested:** revise
 `})}function KpA(A){let{sessionId:Q,branchId:B,stepNumber:E,thought:I,domain:C,verificationConfidence:D,compressionResult:U}=A;return{id:`${Q}:${B}:${E}`,step_number:E,thought:I,timestamp:Date.now(),branch_id:B,verification:{passed:!1,confidence:D,domain:C},compression:U?{input_bytes_saved:(U.original_tokens-U.compressed_tokens)*4,output_bytes_saved:0,context_bytes_saved:0,original_tokens:U.original_tokens,compressed_tokens:U.compressed_tokens}:void 0}}async function zpA(A,Q,B,E){if(!B)return{thought:A,result:null};let I=cg({thought:A,systemPrompt:Q});if(!I.hasComputations)return{thought:A,result:null};return await E({type:"text",text:`⚡ **Augmented** ${I.computations.length} computations (${I.domain})
-`}),{thought:I.augmented,result:{applied:!0,computations:I.computations.length,filtered:I.filteredCount,domain:I.domain}}}async function ZpA(A,Q,B,E){if(!(Q.compress||B||A.length>500&&r0A(A).shouldCompress))return{thought:A,result:null,autoCompressed:!1};let C=Q.compression_query||Q.context||"",U=iM(A,C,{target_ratio:B?0.4:0.6}),Y=B&&!Q.compress,F=Y?" [budget guard]":"";return await E({type:"text",text:`\uD83D\uDCE6 **Compressed** ${U.original_tokens}→${U.compressed_tokens} tokens (${(U.ratio*100).toFixed(0)}%)${F}
+`}),{thought:I.augmented,result:{applied:!0,computations:I.computations.length,filtered:I.filteredCount,domain:I.domain}}}async function ZpA(A,Q,B,E){if(!(Q.compress!==!1||B||A.length>500&&r0A(A).shouldCompress))return{thought:A,result:null,autoCompressed:!1};let C=Q.compression_query||Q.context||"",U=iM(A,C,{target_ratio:B?0.4:0.6}),Y=B&&!Q.compress,F=Y?" [budget guard]":"";return await E({type:"text",text:`\uD83D\uDCE6 **Compressed** ${U.original_tokens}→${U.compressed_tokens} tokens (${(U.ratio*100).toFixed(0)}%)${F}
 `}),{thought:U.compressed,result:{applied:!0,original_tokens:U.original_tokens,compressed_tokens:U.compressed_tokens,ratio:U.ratio},autoCompressed:Y}}async function RpA(A,Q,B,E){if(A.length>_YA){await E({type:"text",text:`⚠️ Question too long (${A.length} chars, max ${_YA}). Skipping trap detection.
 
 `});return}if(WA.setQuestion(Q,A),B!==1){await E({type:"text",text:`⚠️ Question provided at step ${B}. Trap priming only runs on step 1. Stored for spot-check at complete.
@@ -516,7 +516,7 @@ FLOW:
 3.[optional]challenge()→adversarial self-check of claims
 4.complete(final_answer="...")→auto spot-check against common traps
 5.if status=review→revise per reconsideration.suggested_revise
-`,parameters:$YA,annotations:{streamingHint:!0},execute:async(A,Q)=>{let B=WA.getActiveSession();if(!B)B=`s_${crypto.randomUUID()}`,WA.setActiveSession(B);try{if(A.hard_limit_tokens&&B){let U=WA.getTokenUsage(B);if(U&&U.total>=A.hard_limit_tokens){let Y={session_id:B,current_step:0,branch:"main",operation:A.operation,chain_confidence:0,confidence_threshold:A.confidence_threshold,steps_with_confidence:0,status:"budget_exhausted",suggested_action:"Token budget exhausted. Complete the reasoning chain with your current answer or start a new session.",session_tokens:U,budget_exhausted:{limit:A.hard_limit_tokens,current:U.total,exceeded_by:U.total-A.hard_limit_tokens,message:`Session has used ${U.total} tokens, exceeding hard limit of ${A.hard_limit_tokens}.`,recommendation:"Use complete operation to finalize your answer, or start a fresh session for new reasoning."}};return{content:[{type:"text",text:`
+`,parameters:$YA,annotations:{streamingHint:!0},execute:async(A,Q)=>{let B=WA.getActiveSession();if(!B)B=`s_${crypto.randomUUID()}`,WA.setActiveSession(B);try{if(A.hard_limit_tokens&&B){let U=WA.getTokenUsage(B);if(U&&U.total>=A.hard_limit_tokens){let Y={session_id:B,current_step:0,branch:"main",operation:A.operation,chain_confidence:0,confidence_threshold:A.confidence_threshold??0.8,steps_with_confidence:0,status:"budget_exhausted",suggested_action:"Token budget exhausted. Complete the reasoning chain with your current answer or start a new session.",session_tokens:U,budget_exhausted:{limit:A.hard_limit_tokens,current:U.total,exceeded_by:U.total-A.hard_limit_tokens,message:`Session has used ${U.total} tokens, exceeding hard limit of ${A.hard_limit_tokens}.`,recommendation:"Use complete operation to finalize your answer, or start a fresh session for new reasoning."}};return{content:[{type:"text",text:`
 \`\`\`json
 ${JSON.stringify(Y,null,2)}
 \`\`\``}]}}}let E;switch(A.operation){case"step":E=await $pA(A,Q,B);break;case"navigate":E=await vpA(A,Q,B);break;case"branch":E=await TpA(A,Q,B);break;case"revise":E=await kpA(A,Q,B);break;case"complete":E=await _pA(A,Q,B);break;case"augment":E=await bpA(A,Q,B);break;case"override":E=await ypA(A,Q,B);break;case"hint":E=await xpA(A,Q,B);break;case"mistakes":E=await fpA(A,Q,B);break;case"spot_check":E=await hpA(A,Q,B);break;case"challenge":E=await upA(A,Q,B);break;default:throw Error(`Unknown operation: ${A.operation}`)}let I=bC(A,E);if(E.compression?.applied){let{countTokens:U}=await Promise.resolve().then(() => (dM(),a0A)),Y=WA.get(E.session_id),g=Y?.thoughts[Y.thoughts.length-1]?.thought||"",N=U(g),J={...A,thought:void 0},G=U(JSON.stringify(J));I.input_tokens=N+G,I.total_tokens=I.input_tokens+I.output_tokens}E.tokens=I;let C=WA.get(E.session_id);if(C)C.tokenUsage.input+=I.input_tokens,C.tokenUsage.output+=I.output_tokens,C.tokenUsage.operations+=1;let D=WA.getTokenUsage(E.session_id)||{total_input:0,total_output:0,total:0,operations:0};if(E.session_tokens=D,A.warn_at_tokens&&D.total>A.warn_at_tokens)E.token_warning={threshold:A.warn_at_tokens,current:D.total,exceeded_by:D.total-A.warn_at_tokens,message:`Session token usage (${D.total}) exceeds threshold (${A.warn_at_tokens}). Consider completing or compressing.`};return{content:[{type:"text",text:`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "verifiable-thinking-mcp",
-  "version": "0.6.0",
+  "version": "0.6.1",
   "description": "MCP server for structured reasoning with cognitive trap detection",
   "author": "CoderDayton",
   "license": "MIT",

package/src/think/helpers.ts

@@ -247,7 +247,7 @@ export interface AugmentResult {
  * @returns Augmented thought and metadata, or null if disabled/no computations
  */
 export function tryAugment(args: ThinkArgs, thought: string): AugmentResult | null {
-  if (!args.augment_compute) {
+  if (args.augment_compute === false) {
     return null;
   }
 

package/src/think/scratchpad-schema.ts

@@ -42,14 +42,14 @@ export const ScratchpadSchema = z.object({
     .number()
     .min(0)
     .max(1)
-    .default(0.8)
-    .describe("Chain confidence threshold to suggest completion"),
+    .optional()
+    .describe("Chain confidence threshold to suggest completion (default: 0.8)"),
   token_budget: z
     .number()
     .int()
     .min(100)
-    .default(3000)
-    .describe("Max tokens before auto-compressing new steps"),
+    .optional()
+    .describe("Max tokens before auto-compressing new steps (default: 3000)"),
   warn_at_tokens: z
     .number()
     .int()
@@ -99,12 +99,12 @@ export const ScratchpadSchema = z.object({
       "Run domain verification. Auto-enabled for chains >3 steps. Set to false to disable.",
     ),
   domain: z.enum(["math", "logic", "code", "general"]).optional(),
-  local_compute: z.boolean().default(false).describe("Try local compute for math"),
+  local_compute: z.boolean().optional().describe("Try local compute for math (default: false)"),
   augment_compute: z
     .boolean()
-    .default(true)
+    .optional()
     .describe("Auto-inject computed values into thought (default: true)"),
-  compress: z.boolean().default(true).describe("Compress thought before storing"),
+  compress: z.boolean().optional().describe("Compress thought before storing (default: true)"),
   compression_query: z.string().optional().describe("Query for context-aware compression"),
   max_step_tokens: z
     .number()
@@ -112,7 +112,10 @@ export const ScratchpadSchema = z.object({
     .min(10)
     .optional()
     .describe("Max tokens for this step. Rejects if exceeded (default: no limit)"),
-  force_large: z.boolean().default(false).describe("Allow step even if it exceeds max_step_tokens"),
+  force_large: z
+    .boolean()
+    .optional()
+    .describe("Allow step even if it exceeds max_step_tokens (default: false)"),
   preconditions: z
     .array(z.string())
     .optional()
@@ -127,7 +130,7 @@ export const ScratchpadSchema = z.object({
     ),
   step_id: z.number().int().min(1).optional().describe("Step number to view"),
   branch_id: z.string().optional().describe("Filter history by branch"),
-  limit: z.number().int().min(1).max(50).default(10).describe("Max steps to return"),
+  limit: z.number().int().min(1).max(50).optional().describe("Max steps to return (default: 10)"),
 
   // Branch operation fields
   from_step: z.number().int().min(1).optional().describe("Step to branch from (default: current)"),
@@ -161,7 +164,10 @@ export const ScratchpadSchema = z.object({
     .optional()
     .describe("Text containing math expressions to compute and inject (augment/mistakes)"),
   system_context: z.string().optional().describe("System prompt context for domain filtering"),
-  store_as_step: z.boolean().default(false).describe("Store augmented result as a reasoning step"),
+  store_as_step: z
+    .boolean()
+    .optional()
+    .describe("Store augmented result as a reasoning step (default: false)"),
 
   // Override operation fields
   acknowledge: z
@@ -183,9 +189,14 @@ export const ScratchpadSchema = z.object({
     .describe("Number of steps to reveal. Omit to auto-increment when continuing."),
   cumulative: z
     .boolean()
-    .default(true)
-    .describe("Show all steps up to reveal_count (true) or just the nth step (false)"),
-  reset: z.boolean().default(false).describe("Reset hint state and start from beginning"),
+    .optional()
+    .describe(
+      "Show all steps up to reveal_count (true) or just the nth step (false). Default: true",
+    ),
+  reset: z
+    .boolean()
+    .optional()
+    .describe("Reset hint state and start from beginning (default: false)"),
 
   // Spot check operation fields
   answer: z.string().optional().describe("The proposed answer to check for trap patterns"),

package/src/tools/scratchpad.ts

@@ -607,7 +607,7 @@ async function applyCompression(
   autoCompressed: boolean;
 }> {
   const shouldCompress =
-    args.compress ||
+    args.compress !== false ||
     budgetExceeded ||
     (thought.length > 500 && needsCompression(thought).shouldCompress);
 
@@ -2630,7 +2630,7 @@ FLOW:
             branch: "main",
             operation: args.operation,
             chain_confidence: 0,
-            confidence_threshold: args.confidence_threshold,
+            confidence_threshold: args.confidence_threshold ?? 0.8,
             steps_with_confidence: 0,
             status: "budget_exhausted",
             suggested_action: