> For the complete documentation index, see [llms.txt](https://help.delpha.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.delpha.io/core-concepts/how-delpha-works-resolution-evidence-confidence-credits.md). # How Delpha works — resolution, evidence, confidence, credits The concepts behind every Delpha check: the resolution as the unit of work, evidence on every verdict, confidence and review states, and credits as the single meter across channels. Delpha is the deterministic source of commercial truth for AI agents — the verification an agent (or a person) calls before acting on a company. The input is always your own record. Delpha doesn't ship you a database: you bring the account as it exists in your CRM or pipeline, and Delpha verifies it against external reality. Anything Delpha adds comes back as evidence or as a correction — never as a feed. Every check runs the same pipeline, whichever channel you use: **Anchor on the website → Identify the company and its parent → Reconcile against your record → Verdict + evidence.** Delpha anchors identity on the validated website, not the company name. Name-matching is how a CRM merges Bolt (SF, USA) with Bolt (Estonia); the website is the key that keeps them apart. ## Resolution — the unit of work A **resolution** is one run through the Delpha engine: a Playground check, an API call or batch, an MCP tool call, or an Audit file. Every resolution is recorded in your console and can be re-opened later, free — the verdict, the evidence, and the input it was given. Resolutions are deterministic: the same input reconciled against the same evidence returns the same verdict. Reproducible, evidenced, auditable — not a model's best guess. Verifications run asynchronously: you submit, Delpha investigates live sources, you retrieve the result. Each channel page explains its own submit-and-retrieve mechanics. ## Evidence — every verdict carries its proof Delpha never returns a bare answer. Each verdict ships with the evidence behind it: which sources were consulted, what they showed, and when. When Delpha proposes a correction to your record, the correction carries the same evidence — nothing is overwritten silently. This is what makes a Delpha verdict auditable: anyone re-opening a resolution sees exactly why the engine decided what it decided. ## Confidence — Delpha tells you how sure it is Every verdict includes a confidence score. When the evidence clears the bar, the result is **Verified**. When it doesn't, Delpha flags it **Needs review** rather than guessing — a human decides, with the evidence in front of them. A resolution moves through five states: **Processing**, then **Verified**, **Needs review**, **Unresolved** (no reliable answer exists in the sources), or **Failed** (the run itself errored and is not charged). ## Credits — one meter across every channel Credits are the single meter for everything Delpha does. Each resolution consumes credits from your workspace balance — shared across the app, batch uploads, API and MCP, not per-seat. Audit shows you the credit cost of a file before you run it. Subscriptions add a monthly credit allocation that rolls over while the plan is active; pay-as-you-go packs top the balance up anytime. When the balance is empty, new resolutions pause until you top up. See [delpha.ai](https://delpha.ai/) for current packs and tiers. ## One engine, four doors Salesforce, API, MCP, and app.delpha.ai all call the same engine and return the same JSON payload — one resolution record wherever you ran it. The MCP and API doors make Delpha consumable from any agent environment: Claude, ChatGPT and OpenAI agents, Cursor, VS Code, Agentforce, or your own custom agents — including stacks running on AWS, Snowflake, or Databricks. If it speaks MCP or HTTPS, it can verify a company before it acts on one. The use-case pages explain what each verdict means; the channel pages cover how to trigger it and where results land. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://help.delpha.io/core-concepts/how-delpha-works-resolution-evidence-confidence-credits.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.