[{"data":1,"prerenderedAt":473},["ShallowReactive",2],{"doc-\u002Fdocs\u002Fworking-with-the-agent\u002Fconfidence-gaps-and-iteration-limits":3,"docs-nav":350},{"id":4,"title":5,"body":6,"description":16,"extension":343,"meta":344,"navigation":345,"path":346,"seo":347,"stem":348,"__hash__":349},"content\u002Fdocs\u002F4.working-with-the-agent\u002F4.confidence-gaps-and-iteration-limits.md","Confidence, Gaps & Iteration Limits",{"type":7,"value":8,"toc":322},"minimark",[9,13,17,22,30,52,57,60,91,95,98,118,121,125,128,132,139,142,146,153,156,167,170,174,181,207,210,214,217,228,232,235,246,253,257,260,286,289,293,302,306],[10,11,5],"h1",{"id":12},"confidence-gaps-iteration-limits",[14,15,16],"p",{},"The agent is built to be honest about uncertainty and to stop when more iteration won't help. This page covers the three mechanisms that keep it that way: per-claim confidence scores, gap classification, and the hard 3-loop cap on refinement.",[18,19,21],"h2",{"id":20},"confidence-scores","Confidence scores",[14,23,24,25,29],{},"Every claim the Researcher produces gets a ",[26,27,28],"strong",{},"0–100 confidence score",", attached to the node. The score reflects:",[31,32,33,40,46],"ul",{},[34,35,36,39],"li",{},[26,37,38],{},"Source strength."," How authoritative are the documents the claim was drawn from?",[34,41,42,45],{},[26,43,44],{},"Consensus."," Do multiple sources agree, or is this one paper's view?",[34,47,48,51],{},[26,49,50],{},"Specificity."," Is the claim narrowly tied to the cited passages, or is it a broader generalization the passages only partially support?",[53,54,56],"h3",{"id":55},"reading-the-scores","Reading the scores",[14,58,59],{},"Rough heuristic for interpreting confidence:",[31,61,62,68,74,80],{},[34,63,64,67],{},[26,65,66],{},"85–100"," — well-supported, multiple sources in agreement, claim narrowly matches what's cited.",[34,69,70,73],{},[26,71,72],{},"65–84"," — solid, but worth a glance. Often a single strong source or two weaker corroborating ones.",[34,75,76,79],{},[26,77,78],{},"40–64"," — the agent had to reach. Inspect the citations and decide whether the claim is defensible as written.",[34,81,82,85,86,90],{},[26,83,84],{},"Below 40"," — flagged for your review. Often these ",[87,88,89],"em",{},"are"," the most interesting claims (contested ground), but they shouldn't go into your writing without a manual pass.",[53,92,94],{"id":93},"what-to-do-with-low-confidence-claims","What to do with low-confidence claims",[14,96,97],{},"Don't reflexively delete them. Three productive responses:",[31,99,100,106,112],{},[34,101,102,105],{},[26,103,104],{},"Soften."," Rewrite the claim to match what the evidence actually shows. \"X is the dominant view\" → \"X is one of several competing views.\"",[34,107,108,111],{},[26,109,110],{},"Add sources."," If the claim is right but undersupported, find more evidence in your KB (or add it to the KB).",[34,113,114,117],{},[26,115,116],{},"Counter-claim."," If the low confidence reflects genuine controversy, create a sibling node for the opposing view. The contradiction is the right representation.",[14,119,120],{},"Confidence scores are visible on each node and surfaced in the workflow summary so you can find them without hunting.",[18,122,124],{"id":123},"gap-classification","Gap classification",[14,126,127],{},"When the Researcher comes up short — either too few results, or results that all repeat the same thin material — it doesn't just give up. The RePlanner classifies the shortfall into one of two categories:",[53,129,131],{"id":130},"retrievable-gaps","Retrievable gaps",[14,133,134,135,138],{},"The information probably ",[87,136,137],{},"is"," in the KB, but the queries didn't surface it. Maybe the terminology was off, maybe the question needed to be sliced differently.",[14,140,141],{},"The RePlanner's response: generate a revised, more targeted set of queries and send the Researcher back in for another loop. You'll see this happen as the workflow continues without intervention.",[53,143,145],{"id":144},"kb-limitations","KB limitations",[14,147,148,149,152],{},"The information ",[87,150,151],{},"isn't"," in your KB. No amount of query refinement is going to find it.",[14,154,155],{},"The RePlanner's response: stop searching. Include the gap in the user-facing summary with a note about what's missing. Typical suggestions:",[31,157,158,161,164],{},[34,159,160],{},"Add specific papers to the KB.",[34,162,163],{},"Relax the question.",[34,165,166],{},"Accept the gap and proceed with what's available.",[14,168,169],{},"Distinguishing these two cases automatically is one of the more important things the system does — it stops you from burning credits looking for things that aren't there.",[18,171,173],{"id":172},"the-3-loop-cap","The 3-loop cap",[14,175,176,177,180],{},"The RePlanner can decide to refine the search at most ",[26,178,179],{},"three times per instruction",". The loop structure:",[31,182,183,189,195,201],{},[34,184,185,188],{},[26,186,187],{},"Initial."," FrontDesk hands off to the Researcher for broad discovery.",[34,190,191,194],{},[26,192,193],{},"Loop 1."," RePlanner narrows the scope to specific identified gaps.",[34,196,197,200],{},[26,198,199],{},"Loop 2."," Further narrowing — focusing on what loop 1 still missed.",[34,202,203,206],{},[26,204,205],{},"Loop 3."," Final targeted retrieval for edge cases.",[14,208,209],{},"After the third iteration, the system hardcodes a stop: the Editor applies whatever's best, the RePlanner writes a final summary, and the workflow ends. This is non-negotiable — there's no \"let it keep trying\" override.",[53,211,213],{"id":212},"why-three","Why three?",[14,215,216],{},"In practice, retrieval-augmented systems hit sharply diminishing returns past loop 2. The 3-loop cap exists because:",[31,218,219,222,225],{},[34,220,221],{},"Most genuine gaps are either resolved by loop 2 or are KB limitations.",[34,223,224],{},"Past loop 3, the system would mostly be paraphrasing earlier results back at itself.",[34,226,227],{},"Unbounded loops would let a single instruction burn arbitrary credits.",[53,229,231],{"id":230},"what-happens-at-the-cap","What happens at the cap",[14,233,234],{},"When the cap is reached, the workflow doesn't fail — it terminates cleanly:",[31,236,237,240,243],{},[34,238,239],{},"Best available results are applied to the mindmap.",[34,241,242],{},"The summary explains explicitly that the cap was reached.",[34,244,245],{},"Specific limitations or unresolved gaps are listed with follow-up suggestions.",[14,247,248,249,252],{},"If you want to push further on the same topic, the right move is another instruction with a ",[87,250,251],{},"narrower"," scope, not a way to extend the original loop.",[18,254,256],{"id":255},"reading-the-summary","Reading the summary",[14,258,259],{},"Every workflow ends with a textual summary that includes:",[31,261,262,268,274,280],{},[34,263,264,267],{},[26,265,266],{},"What changed"," on the mindmap.",[34,269,270,273],{},[26,271,272],{},"Completion score"," — the RePlanner's 0–100 assessment of how well the instruction was addressed.",[34,275,276,279],{},[26,277,278],{},"Limitations encountered"," — KB gaps, low-confidence areas, anything the agent flagged.",[34,281,282,285],{},[26,283,284],{},"Follow-up suggestions"," — usually 1–3 specific next instructions.",[14,287,288],{},"The summary is short by design (under ~300 words). Read it. The follow-up suggestions in particular often save a turn — they're written with the workflow's full context, which yours might be missing.",[18,290,292],{"id":291},"token-and-budget-enforcement","Token and budget enforcement",[14,294,295,296,301],{},"Independent of the 3-loop cap, there are configurable per-operation and total token limits. If a workflow would exceed your budget, it's stopped early and the partial result is reported transparently. See ",[297,298,300],"a",{"href":299},"..\u002Fgetting-started\u002Fcredits-and-billing","Credits & Billing"," for the credit-side view.",[18,303,305],{"id":304},"whats-next","What's next",[31,307,308,315],{},[34,309,310,314],{},[297,311,313],{"href":312},".\u002Fversion-history-and-provenance","Version History & Provenance"," — inspecting the basis for any specific claim.",[34,316,317,321],{},[297,318,320],{"href":319},".\u002Fwriting-instructions","Writing Instructions"," — phrasing that helps the agent terminate well.",{"title":323,"searchDepth":324,"depth":324,"links":325},"",4,[326,332,336,340,341,342],{"id":20,"depth":327,"text":21,"children":328},2,[329,331],{"id":55,"depth":330,"text":56},3,{"id":93,"depth":330,"text":94},{"id":123,"depth":327,"text":124,"children":333},[334,335],{"id":130,"depth":330,"text":131},{"id":144,"depth":330,"text":145},{"id":172,"depth":327,"text":173,"children":337},[338,339],{"id":212,"depth":330,"text":213},{"id":230,"depth":330,"text":231},{"id":255,"depth":327,"text":256},{"id":291,"depth":327,"text":292},{"id":304,"depth":327,"text":305},"md",{},true,"\u002Fdocs\u002Fworking-with-the-agent\u002Fconfidence-gaps-and-iteration-limits",{"title":5,"description":16},"docs\u002F4.working-with-the-agent\u002F4.confidence-gaps-and-iteration-limits","VlvDoC2DctlxazqWkCNxPoG4G0us0_lPFir5_JorX7M",[351,374,410,442,461],{"slug":352,"title":353,"items":354},"getting-started","Getting Started",[355,359,363,367,371],{"path":356,"title":357,"stem":358},"\u002Fdocs\u002Fgetting-started\u002Fintroduction","Introduction","docs\u002F1.getting-started\u002F1.introduction",{"path":360,"title":361,"stem":362},"\u002Fdocs\u002Fgetting-started\u002Fquickstart","Quickstart","docs\u002F1.getting-started\u002F2.quickstart",{"path":364,"title":365,"stem":366},"\u002Fdocs\u002Fgetting-started\u002Fthe-interface","The Interface","docs\u002F1.getting-started\u002F3.the-interface",{"path":368,"title":369,"stem":370},"\u002Fdocs\u002Fgetting-started\u002Fzotero-integration","Zotero Integration","docs\u002F1.getting-started\u002F4.zotero-integration",{"path":372,"title":300,"stem":373},"\u002Fdocs\u002Fgetting-started\u002Fcredits-and-billing","docs\u002F1.getting-started\u002F5.credits-and-billing",{"slug":375,"title":376,"items":377},"core-concepts","Core Concepts",[378,382,386,390,394,398,402,406],{"path":379,"title":380,"stem":381},"\u002Fdocs\u002Fcore-concepts\u002Fprojects","Projects","docs\u002F2.core-concepts\u002F1.projects",{"path":383,"title":384,"stem":385},"\u002Fdocs\u002Fcore-concepts\u002Fknowledge-bases","Knowledge Bases","docs\u002F2.core-concepts\u002F2.knowledge-bases",{"path":387,"title":388,"stem":389},"\u002Fdocs\u002Fcore-concepts\u002Frag-and-retrieval","RAG and Retrieval","docs\u002F2.core-concepts\u002F3.rag-and-retrieval",{"path":391,"title":392,"stem":393},"\u002Fdocs\u002Fcore-concepts\u002Flabeled-items","Labeled Items","docs\u002F2.core-concepts\u002F4.labeled-items",{"path":395,"title":396,"stem":397},"\u002Fdocs\u002Fcore-concepts\u002Fmindmaps-nodes-and-claims","Mindmaps, Nodes & Claims","docs\u002F2.core-concepts\u002F5.mindmaps-nodes-and-claims",{"path":399,"title":400,"stem":401},"\u002Fdocs\u002Fcore-concepts\u002Fthe-multi-agent-system","The Multi-Agent System","docs\u002F2.core-concepts\u002F6.the-multi-agent-system",{"path":403,"title":404,"stem":405},"\u002Fdocs\u002Fcore-concepts\u002Fterminology-graph-explorer","Terminology Graph Explorer","docs\u002F2.core-concepts\u002F7.terminology-graph-explorer",{"path":407,"title":408,"stem":409},"\u002Fdocs\u002Fcore-concepts\u002Fai-verification","AI Verification","docs\u002F2.core-concepts\u002F8.ai-verification",{"slug":411,"title":412,"items":413},"research-workflows","Research Workflows",[414,418,422,426,430,434,438],{"path":415,"title":416,"stem":417},"\u002Fdocs\u002Fresearch-workflows\u002Fliterature-review","Literature Review","docs\u002F3.research-workflows\u002F1.literature-review",{"path":419,"title":420,"stem":421},"\u002Fdocs\u002Fresearch-workflows\u002Fhypothesis-generation-and-testing","Hypothesis Generation & Testing","docs\u002F3.research-workflows\u002F2.hypothesis-generation-and-testing",{"path":423,"title":424,"stem":425},"\u002Fdocs\u002Fresearch-workflows\u002Fthematic-analysis","Thematic Analysis","docs\u002F3.research-workflows\u002F3.thematic-analysis",{"path":427,"title":428,"stem":429},"\u002Fdocs\u002Fresearch-workflows\u002Fcomparative-analysis","Comparative Analysis","docs\u002F3.research-workflows\u002F4.comparative-analysis",{"path":431,"title":432,"stem":433},"\u002Fdocs\u002Fresearch-workflows\u002Fnode-transformation","Node Transformation","docs\u002F3.research-workflows\u002F5.node-transformation",{"path":435,"title":436,"stem":437},"\u002Fdocs\u002Fresearch-workflows\u002Foutlining-a-paper","Outlining a Paper","docs\u002F3.research-workflows\u002F6.outlining-a-paper",{"path":439,"title":440,"stem":441},"\u002Fdocs\u002Fresearch-workflows\u002Fgap-finding","Gap Finding","docs\u002F3.research-workflows\u002F7.gap-finding",{"slug":443,"title":444,"items":445},"working-with-the-agent","Working With The Agent",[446,449,453,457,458],{"path":447,"title":320,"stem":448},"\u002Fdocs\u002Fworking-with-the-agent\u002Fwriting-instructions","docs\u002F4.working-with-the-agent\u002F1.writing-instructions",{"path":450,"title":451,"stem":452},"\u002Fdocs\u002Fworking-with-the-agent\u002Fagent-context","Agent Context","docs\u002F4.working-with-the-agent\u002F2.agent-context",{"path":454,"title":455,"stem":456},"\u002Fdocs\u002Fworking-with-the-agent\u002Fconversational-refinement","Conversational Refinement","docs\u002F4.working-with-the-agent\u002F3.conversational-refinement",{"path":346,"title":5,"stem":348},{"path":459,"title":313,"stem":460},"\u002Fdocs\u002Fworking-with-the-agent\u002Fversion-history-and-provenance","docs\u002F4.working-with-the-agent\u002F5.version-history-and-provenance",{"slug":462,"title":463,"items":464},"reference","Reference",[465,469],{"path":466,"title":467,"stem":468},"\u002Fdocs\u002Freference\u002Fsettings-and-api-keys","Settings & API Keys","docs\u002F5.reference\u002F1.settings-and-api-keys",{"path":470,"title":471,"stem":472},"\u002Fdocs\u002Freference\u002Ftroubleshooting-and-faq","Troubleshooting & FAQ","docs\u002F5.reference\u002F2.troubleshooting-and-faq",1780503376376]