How it works
One verdict, from the team to the audit.
The charter has your agents ask, before touching a file, for the team’s rules and decisions that apply to it. A check that is not a model refuses what no longer has its proof. Anyone recomputes the verdict.
During development
Your agents work from your decisions, not from their memory.
The same frame serves the AI agent while it builds, not only on delivery day. runward writes the charter the agent reads; whether it follows it depends on your tool, and the check refuses the rest.
- 01The decisions, to read before acting
runward gives the agent the team’s decisions, their status and what they require to be proven, before it touches the code.
- 02A contract for what they write
An accepted decision becomes the contract for what they write. What is not decided remains to be decided, not guessed.
- 03A thread that does not depend on its memory
After a break, the written decisions are where they were.
- 04A decision does not change quietly
The rule: an accepted decision is not modified, a new one is written. When the team has sealed the record, the check refuses a cited decision that changed.
- 05What no longer has its proof is refused
A sealed document that changed, a deviation resting on an unendorsed decision, evidence that no longer points to the code: the check refuses, without AI, and says where.
runward keeps the trace of the decisions and the seal of the evidence; your tests prove the behaviour.
The plain-language reading
From the check’s detail to one sentence.
The sentence comes from a closed dictionary: a kind of gap, a document of the record. The check’s detail stays below, as written.
What the check writes
- kind
- evidence-refused
- file
- runward/governance/threat-model.md
- line
- 50
- problem
- typed pointer file:code/src/core/domain/guard.ts#guardFields — symbol "guardFields" not found in the file (moved or renamed? update the pointer)
What you read
Security assessment: a cited piece of evidence is refused by the check.
check of 10 Nov, 15:52
Five states, one shape and one word each. Never colour alone.
- Refused by the check
- Required level not met
- No check received
- Content not shared
- Nothing is blockingaccording to the team’s requirements
Without taking our word for it
Check it yourself
It is not our screen saying so. The same check, run again on your side on the same commit, gives the same application, the same version, the same state and the same gaps.
- The check runs in the team’s delivery chain.
- The engine is under the MIT licence: anyone can read it and run it again.
- Your audit redoes the calculation on its own machine, without going through the team or through us. It needs read access to the code.
- Same applicationrunward-demo-renaming
- Same versionrunward 0.46.0
- Same stateRefused by the check
- Same gaps2 points to fix
- Repository
- stranxik/runward-demo-renaming
- Commit
- 97fce96d901d0d07cad8a230950d391b223736af
- Public run
- no. 15
Try it on an example, in one command.
Try it on an example