> ## Documentation Index > Fetch the complete documentation index at: https://docs.runwita.com/llms.txt > Use this file to discover all available pages before exploring further. # Topics > Threads of discussion that span engagements, with their own state. Topics are where Runwita stops being a notes app and starts being something else. A topic is a recurring thread, a thing that comes up in multiple conversations, that you want to track over time. *"Licensing model"*, *"Phase 2 timeline"*, *"Onboarding flow redesign"*. Each topic has a state (open, blocked, resolved), a category, and a chronological log of touchpoints across engagements. ## Why topics exist In conventional notes apps, every conversation is a flat island. You search for *"licensing"* and you get five separate transcripts, each with their own context, and you have to mentally stitch the arc back together. The thread isn't first-class. In Runwita, the thread *is* the object. The engagements are how the thread accumulates new state. When you scroll a topic, you see: * The current state (e.g. *Open* or *Blocked*). * Every prior touchpoint, in order, with the engagement it came from and what was said about the topic *in that conversation*. * A category (technical, commercial, governance, planning, relationship). * A status timeline, when it became blocked, when it cleared, when it was resolved. That's the value: the question *"where did we land on the licensing question?"* is one click, not a search. ## How topics get created You don't create topics manually. They're created by the AI during the topic-matching step that runs when you save an engagement. For each section in your extracted engagement, the AI looks at: * The section's heading and body. * The journey's existing topics (their titles, statuses, categories). It decides: *does this section continue an existing topic, or is it a new thread?* If it's continuing, the section becomes a new touchpoint on that topic. If it's new, a new topic is created. A few rules the matcher follows: * **Substance, not section title.** *"License discussion"* and *"Annual licensing renewal"* are the same topic, not two. The matcher looks at content overlap. * **No fragmenting.** A topic only appears once per engagement, even if multiple sections discussed it. The matcher merges them into one touchpoint. * **Logistics aren't topics.** Greetings, scheduling the next conversation, off-topic chitchat don't get topics. Only substantive recurring threads do. * **2 to 6 topics per engagement.** That's the matcher's target band. Not every section needs its own topic. ## Topic state Each topic has a status that reflects where it stands. The status is derived from the chronological history of touchpoints on the topic, ordered by engagement date. | State | What it means | | ------------ | ------------------------------------------------------------------------------------------------------------------ | | **Open** | Active discussion, no impediment. The default state for any new topic and any topic that's moving forward. | | **Blocked** | An impediment was raised and not yet cleared. *"We can't proceed until legal signs off."* | | **Resolved** | The topic is complete, no more discussion expected. Terminal. | | **Stale** | No touchpoint in a long time (configurable per topic). Auto-flagged so the journey doesn't quietly leak attention. | ### The blocked-is-sticky rule Status follows two simple rules: 1. **Resolved is terminal.** Once a topic is resolved, it stays resolved unless you explicitly reopen it. 2. **Blocked is sticky.** Once a touchpoint marks a topic blocked, the topic stays blocked until a *later* touchpoint actively clears it. A bare *discussed* touchpoint after blocked does not clear the block. The block is only cleared by a touchpoint with state *progressed*, *decided*, or *resolved*. This is intentional. Real blockers don't dissolve because the topic got mentioned again. *"We're still waiting on legal"* doesn't unblock anything; *"Legal signed off"* (a progressed or decided touchpoint) does. ### How touchpoint states map The states a touchpoint can carry are slightly richer than the topic states themselves: | Touchpoint state | Effect on topic status | | ---------------- | ---------------------------------------------------------------------------------------------- | | `discussed` | Stays as-is. Won't clear a blocked topic. Sets a fresh topic to open. | | `progressed` | Sets to open. Clears blocked. | | `decided` | Sets to open. Clears blocked. The decision itself is recorded as a Decision on the engagement. | | `blocked` | Sets to blocked. | | `resolved` | Sets to resolved (terminal). | ### Out-of-order processing Topic status is computed from the touchpoints' **engagement dates**, not the order in which you processed them. If you process an Apr 22 conversation first and then later push a back-dated Apr 10 one, the older one doesn't clobber the newer state. The system always reads the chronological tail, regardless of when you got around to processing each piece. ## Categories Each topic has a category, used for filtering and visual grouping on the journey page. Five categories: * **Technical**, architecture, integration, infrastructure, data. * **Commercial**, pricing, licensing, contracts, terms. * **Governance**, compliance, security, privacy, audit. * **Planning**, timelines, milestones, scope, capacity. * **Relationship**, stakeholder dynamics, communication patterns, escalations. The category is set by the AI on creation. You can edit it on the topic detail page if it's wrong. ## Reading the topic timeline Click any topic on a journey page to see its touchpoint history. Each row is one engagement that touched the topic, with: * The date of that engagement. * The state at that touchpoint. * A brief note (one to two sentences from the AI summarising what was said about *this topic* in *that conversation*). Scroll the timeline and you read the arc of the topic, not the engagements. ## Editing topics Topics are editable. From a topic's detail page you can: * Rename the topic. * Change its category. * Manually flip its status (e.g. mark something resolved before the next engagement confirms it). * Edit individual touchpoint notes. * Delete the topic (which removes all its touchpoints; the underlying engagements stay intact). Stale-after-days is also editable per topic. Some topics move weekly, some are quarterly. The auto-stale threshold isn't one-size-fits-all. ## When topics go wrong Topic matching is imperfect. The two failure modes you'll see: **Over-fragmentation.** The matcher creates a new topic when it should have continued an existing one. Usually because the engagement section was worded differently from the existing topic's title. Fix: open both topics, edit one to match, and either re-run extraction (Reprocess with AI) or manually move touchpoints. **Over-merging.** The matcher attaches a section to the wrong existing topic because the surface words overlap but the substance is different. Less common but harder to spot, you usually catch it weeks later. Fix: split the touchpoint off into a new topic from the topic detail page. If you find the matcher is consistently off on a journey, two levers help: 1. Use a sharper, more capable model on the workhorse tier. Claude Haiku 4.5 and gpt-5 do better than gpt-5-nano on this task. 2. Give your topics more distinctive titles. *"Licensing"* matches everything; *"Annual licensing renewal terms"* matches only its actual thread. ## What's next Frontier and Workhorse, and which one runs topic matching.