Docs
How to use Consensus Logic.
Your first task in six steps, the words the app uses, how to read a report, and what each status means. The same words are used here, in the app and in support, so nothing needs translating twice.
Getting started
Your first task, in six steps.
- 01
Check your providers
Open Consensus Logic. The provider check shows whether Codex and Claude Code are installed and signed in on this machine. Both green for the full model; one is enough to look around.
- 02
Register a project
Point the app at a folder. Documentation projects work without Git. The project's record is created under your local application data and is isolated from every other project.
- 03
Describe a Task
Say what outcome you want, in your own words. Type it. Microphone input arrives in the full workflow version and is marked Not implemented until then.
- 04
Start review
The controller establishes project state, both AIs plan without seeing each other's plan, then challenge. If they disagree on something that matters, the decision comes to you with the trade-off stated. The result is a programme with a plain-English summary. Nothing has been built.
- 05
Choose creator and policies
Pick which AI creates. Set Supervision to Manual or Automatic and Repair to Decision required or Auto-bounded. Set the budget for the task, which the app calls the allowance, if the project default is not right.
- 06
Start development
Only available after the summary for this plan revision has been shown. Steps run one at a time inside the allowance. Stop is always available. When the task closes, read the report.
Settled vocabulary
The same words everywhere.
| Term | Meaning |
|---|---|
| Project, Task, Step | A Project is a registered folder. A Task is the outcome you ask for. A Step is a small, limited piece of work the product generates to reach it. |
| Start review | Authorises discovery, two independent plans, challenge and reconciliation. Nothing is built. |
| Start development | Separately authorises the displayed programme to run, after the plain-English summary has been shown. |
| Creator and Reviewer | You choose which AI creates; the other independently reviews. Once per task, not per step. |
| Supervision | Manual: approve each prepared step. Automatic: let the next steps continue within the authority and budget you already gave. |
| Repair | Decision required: a failure comes to you first. Auto-bounded: a limited number of corrections run within the recorded allowance. |
| Allowance | The finite budget of calls, time and repairs at account, project, task and step level. Reserved before each call, settled after. |
| Plain English and Technical | The two report tabs. Same evidence, two levels of detail. |
| Stop | Starts nothing new immediately and reports whether the work actually stopped. If it cannot confirm that, nothing else runs. |
| Where we left off | The resume screen: completed work, open findings, and what to do next. |
| Evidence class | Product: observed through the installed application. Mechanism: a test of the machinery. Never mixed. |
Reading the report
Five questions, two tabs, one set of evidence.
Every task closes with the same five questions answered on the Plain English tab. The Technical tab shows the records, the test output and the ledger the answers were built from. If the two ever tell a different story, that is a bug, and the build log will say so.
- 01
What did I get
What was delivered, in plain words. What was accepted, and what was refused.
- 02
Exactly where is it
Paths, revisions, and where the previous version was kept.
- 03
How do I try it
The shortest route to seeing it work yourself.
- 04
What is still wrong or unknown
Failed checks, blocked steps, and everything Not run, with the test it waits for.
- 05
How long did it take
Elapsed, active and waiting time, with overlap disclosed.
Status states
What each state means, and what to do about it.
| State | Means | What to do |
|---|---|---|
| Verified | Product evidence passed through the installed application. | Nothing. It is proved. |
| Failed | A check ran and did not pass. | Follow your repair policy. The failure is recorded either way; a model saying it is fixed does not change the state. |
| Blocked | A dependency is unverified, or a stop is unconfirmed. | Resolve the dependency or confirm the stop from Where we left off. Nothing new starts until then. |
| Pending | Awaiting your approval or decision. | Open the task. It shows what to do next. |
| Unknown | An outcome the controller cannot yet confirm. | Do not assume success. Check the artefact yourself; the Technical tab shows what was and was not observed. |
| Not run | A check exists and has not been exercised. | Nothing to fix. The report lists it so it is not mistaken for a pass. |
| Not implemented | A capability that arrives in a later version. | The state names the version. The Proof page shows when it closes. |
Resume
Close the laptop. Come back tomorrow.
Close the app after any step. When you open it again, Where we left off shows completed work, open findings, spent allowance and what to do next for every task. Artefacts, records and accepted work are intact. Nothing runs while you are away; there is no unattended scheduler.
Nothing to download yet. Once installed, these docs open from the Help menu.