# Journey agents and checks

A journey runs on one to six agents, each with one model, and grades every answer with checks.

## Agents and models

The first model listed for each agent is its default. Models are shown by the vendor's ID.

| Agent | Type | Journey kinds | Default model | Other models | Reports its sources |
| --- | --- | --- | --- | --- | --- |
| ChatGPT | AI assistant | Research | `gpt-6.1-sol` | `gpt-6-astra`, `gpt-6-luna`, `gpt-5.6-sol`, `gpt-4o` | Yes |
| Claude | AI assistant | Research | `claude-sonnet-5-5` | `claude-opus-5-5`, `claude-haiku-4-5-20251001`, `claude-sonnet-5` | Yes |
| Perplexity | AI assistant | Research | `sonar` | `sonar-pro`, `sonar-reasoning-pro`, `sonar-deep-research` | Yes |
| Gemini | AI assistant | Research | `gemini-3.8-flash` | `gemini-3.1-pro-preview`, `gemini-3.5-flash-lite`, `gemini-2.5-flash` | Yes |
| Claude Code | Coding agent | Research, Agent Task | `claude-sonnet-5-5` | `claude-opus-5-5`, `claude-haiku-4-5-20251001` | Yes |
| Codex | Coding agent | Research, Agent Task | `gpt-5.3-codex` | `gpt-6.1-sol`, `gpt-5.2` | No |

| Setting | Default |
| --- | --- |
| Kind of a new journey | Research |
| Agent | ChatGPT for Research, Claude Code for Agent Task |
| **Number of runs** | 1 |
| Web search | On, at most 5 searches |
| Checks on a new journey | **it hit no contradictions**, **it never gave up or asked for a human** |

## Web search by agent

The web search control appears on Research journeys only.

| Agent | Search can be turned off | Applies the search limit |
| --- | --- | --- |
| ChatGPT | Yes | No |
| Claude | Yes | Yes, as one budget for the whole run |
| Perplexity | No | No |
| Gemini | Yes | No |
| Claude Code | Yes | No |
| Codex | Yes | No |

## The grading model

Judged checks are graded by `claude-sonnet-5`. You cannot change it.

## Library checks

Any check can be added to either kind of journey. **Suggested for** is advice only.

| Check | Kind | Suggested for | Needs | Setting |
| --- | --- | --- | --- | --- |
| **it completed** | Judged | Agent Task | | |
| **the documented order worked** | Judged | Agent Task | | |
| **it hit no contradictions** | Judged | Both | | |
| **it never gave up or asked for a human** | Measured | Both | | |
| **it found an answer** | Measured | Research | | |
| **the answer is correct** | Judged | Research | | |
| **it answered the question asked** | Judged | Research | | |
| **you appear at turn 1** | Measured | Research | Brand names | |
| **you survive to the last turn** | Measured | Research | Brand names | |
| **you are recommended, not just mentioned** | Judged | Research | Brand names | |
| **it finished within the turn ceiling** | Measured | Both | | **Turn ceiling**, 1 to 100 |
| **it stayed under the cost ceiling** | Measured | Both | | **Cost ceiling**, $0.01 to $10,000 |

Brand names come from **Brand names** in [Company Profile](/agent/company-profile/).

## Custom checks

| Type | Kind | Label it gets | Passes when |
| --- | --- | --- | --- |
| **meets a rubric you write** | Judged | Your rubric | The judge finds the rubric met |
| **mentions a term** | Measured | it mentions "term" | The final answer contains the term, ignoring case |
| **mentions a term**, set to **never mentions** | Measured | it never mentions "term" | The final answer does not contain the term |
| **cites a page** | Measured | it cites page | A URL the answer cites contains the text |

## Check results

| Result | Scorecard square | In a collection's counts |
| --- | --- | --- |
| Passed | Green | Graded, passed |
| Failed | Red | Graded, not passed |
| Not checked | White, outlined | Left out |
| Errored run (no results) | Grey | Left out |

| A check is not checked when | Example |
| --- | --- |
| It needs brand names and Company Profile has none | **you are recommended, not just mentioned** |
| Its setting is empty | A ceiling with no value |
| The judge finds the evidence not enough to decide | A rubric about a page the agent never opened |
| The check could not run | An error while grading |

## Runs per question

| How it runs | Answers per question |
| --- | --- |
| **Run N times** or **Run now** | **Number of runs** for each agent |
| **Try it once** | One for each agent |
| **Run collection** | Each journey's own agents and **Number of runs** |
| The free report | 10 on each of ChatGPT, Claude, Perplexity and Gemini (40 per question, 600 for 15 questions) |

## Collection results

| Share of a question's graded answers that passed | Result |
| --- | --- |
| 80% or more | **You win** |
| More than 20% and less than 80% | **Split** |
| 20% or less | **You lose** |
| No graded answers | **No result** |

| The question's checks include | An answer passes when |
| --- | --- |
| **you are recommended, not just mentioned** | That check passed |
| **What it says about you is accurate** (free report questions that name you) | That check passed |
| Neither | Every check that ran passed, and at least one ran |

## Limits

| Item | Limit |
| --- | --- |
| **What do you want to find out?** | 2,000 characters |
| Title | 200 characters |
| Turns | 1 to 20, each up to 2,000 characters |
| Persona | 2,000 characters, or a persona saved in Company Profile |
| Checks per journey | 50 |
| Agents per journey | 1 to 6, each at most once |
| **Number of runs** | 1, 3, 5 or 10 in the form. The app accepts 1 to 25. |
| Web search limit | No limit, 2, 5, 10 or 20 searches in the form. The app accepts 1 to 50. |
| Journeys per collection | 1 to 50, published journeys only (a journey is published after its first run that finishes) |
| Collection name and description | 100 and 500 characters |
| Collection runs | One at a time per collection |
| Collection run length | Closes after 60 minutes without a new answer, using the answers collected |
| Free report | One per organization |