> ## Documentation Index > Fetch the complete documentation index at: https://docs.meetsquad.ai/llms.txt > Use this file to discover all available pages before exploring further. # The evidence pipeline > How Squad gets from a customer complaint to a decision you can defend. The whole chain, described once. The evidence pipeline The evidence pipeline Squad runs one chain. Feedback arrives from your connected tools, gets broken into individual claims, those claims group themselves by meaning, the groups get synthesised into findings, and the findings turn into work. Every screen in the product is a window onto one stage of that chain. This page is the map. Each stage links to the page that covers it properly. A source hands Squad raw text. That might be a Slack thread, a GitHub issue, a Gong call transcript, an app store review, a page an agent read, or a document you uploaded. Squad reads it and pulls out the specific claims it contains, one per **data signal**. A signal keeps the verbatim quote, a one-line summary, a type (pain point, feature request, bug report, and so on), a sentiment, a strength between 0 and 1, and a link back to where it came from. Nothing gets paraphrased away, so you can always read the customer's actual words. See [Data signals](/signals). The moment a batch of signals lands, each one is compared against the existing clusters of the same type. Close enough to one, it joins. Close to nothing, it starts a cluster of its own. The comparison is on meaning rather than wording, so "the export takes forever" and "CSV downloads time out" can end up in the same place. Once a cluster reaches three members, Squad reads them and writes a theme label plus a confidence score for how well the grouping holds together. Under three members a cluster stays unnamed, which is why a new workspace shows an empty Signal Clusters tab for a while. See [Signal clusters](/signal-clusters). When a clustering pass finishes, Squad looks at every active cluster with at least two members and synthesises what they add up to. A one-member cluster also qualifies if the signal behind it was captured deliberately by you or a teammate rather than picked up passively. The output is an **insight**: a title, a description, and a link back to every signal supporting it. Before it is saved, the new insight is checked against the ones you already have. A near-identical match folds into the existing insight instead of duplicating it. A close but distinct match creates a new insight with a "related to" link to its neighbour. See [Insights](/insights). Every insight the extraction pass touched is rescored. The score has two halves. **Strength** measures how well evidenced the insight is. It combines volume (how many signals), statistical confidence in that volume, how many distinct customers those signals came from, and how many different kinds of source corroborate it. A brand new insight with two signals is pulled toward your workspace's baseline rather than being crushed to zero, so new workspaces are not structurally penalised. **Momentum** measures whether the theme is alive. Recent signals count for more than old ones, a sudden burst of arrivals lifts it, and the trend over the last four weeks is read as accelerating, steady, or decelerating. The number you see on the Insights page is `strength * 0.6 + momentum * 0.4`. It moves when new evidence reaches the insight. It does not drift on its own between extraction runs. An insight that scores 0.65 or above is handed to Squad for action proposals, and you get a short ranked list of concrete next steps on the insight's detail page under **Squad recommends**. A nightly pass sweeps active insights over the same bar in case one crossed it since the last run. You can also trigger ranking by hand on any insight, which ignores the score bar entirely. See [Actions](/actions). Nothing gets written up automatically. When an action's real job is a document, you choose **Generate PRD** or **Generate decision brief** and Squad drafts it in the background from the evidence behind the action. You can also generate straight from an insight, or from a prompt of your own. See [One-pagers](/one-pagers). ## Where goals fit Goals steer the judgement calls. They do not touch the arithmetic. During extraction, Squad decides which of your goals each new insight supports and writes those links. When it proposes actions for an insight, it is shown your goals and their importance (1 to 5) so that relevance to what you actually care about shapes the ranking it returns. A goal never changes an insight's score. Raising a goal's importance from 3 to 5 will not move a single number on the Insights page. What it changes is how Squad weighs relevance when it is choosing and ordering work. See [Goals](/goals). ## What holds the chain together Each stage writes a link rather than a copy. A signal points at its cluster and at the insights it supports. An insight points at its goals. A one-pager points at everything its body cites. That is why you can open any one-pager and walk back to the customer sentence that started it. Some of those links are yours to make and break. See [Relationships](/relationships). ## What runs itself and what does not Signal extraction, clustering, labelling, insight synthesis, scoring, and action proposal all run without you. Squad also links insights to goals on its own. Creating goals, deciding an action's priority and owner, generating a one-pager, and making the final call on a decision brief are yours. Squad does not write a document or close a decision unprompted. ## Every stage, in order Individual claims, each with its verbatim quote and source Themes formed from signals that mean the same thing Scored findings, and how strength and momentum combine Concrete next steps, in priority order PRDs and decision briefs, backed by cited evidence The objectives that steer what Squad recommends How entities link, and how to trace a decision back to a customer sentence