Reading your report and fixing what it finds
How to read a Verigent report. The composite and tier, the four pillars, the class radar, the bands, and the fix list ranked by how much each item would move your score. Then how a fix shows up on its own.
The report has one job, to tell you what to fix next. The headline is a composite from 0 to 100 and the tier it lands in. Under that, the score splits into 4 pillars, then into 32 dimensions, each with a number, a band, and the method that produced it. The fix list at the end ranks your weaknesses by how much fixing each one would move the composite.
The composite and the tier
The composite is a weighted sum of every dimension. Read it as the absolute ladder that lets you compare across different kinds of agent. Next to it is the proof status, Current, Ageing or Stale, which tells you how fresh the evidence is. How good and how recent are kept as two separate signals.
The pillars
Each pillar has its own band. The weighting tells you where a fix pays most. A fail inside the Agent pillar, which carries 50% of the composite, drags the number far harder than the same fail in a base-brain pillar.
On a free run the Sovereignty pillar reads as not measured. That is honest, not broken. Nothing was tested there, so nothing is scored.
The class radar
Your agent is drawn as a 12-spoke radar in a fixed order, so the shape always means the same thing. The strongest spoke is your class, and your standing is measured against other agents in that class rather than against everyone. Behind the main outline are fainter lines, the individual task scores. A tight cluster means the agent does that thing consistently. A wide spread means it depends on the task.
Bands and methods
Every dimension shows a number, a band (pass, warn, fail, or not measured) and how it was scored. Machine check, judged transcript, tripwire, or a real action the agent performed. Colour only ever means a measured state.
When your agent says done and it isn't
A lot of agents produce the words of completion without the thing having happened. The reservation "updated" with no record behind it, the file "written" that isn't there. Verigent doesn't take the agent's word for any of it. Where the task has a checkable outcome we check the outcome, and a claim without one scores zero. If your agent has this habit it shows up in the report as a fail on the dimensions that test it — outcome residual, false-positive resistance, workflow execution — with the tasks named, so you can go to the behaviour instead of guessing.
The fix list
A list of everything wrong isn't much use. An ordered list of what to fix first is. The fix list ranks each weakness by expected impact, how much moving that dimension out of its current band would move the composite. Each entry names the evidence, which tasks failed and on which dimension.
Most fixes are harness changes, not model swaps. A retry with backoff on a failed tool call. A re-plan step when a fact contradicts the current plan. Holding a correct position under pushback. The report tells you which behaviour a dimension tests. The fix is on your side.
Seeing a fix land
You don't trigger a retest. Continuous verification comes round on its own, about 5 pulls a day, a few dimensions each, and every dimension is re-sat within the week. When the affected dimension comes up again, the change shows. Scores are never rewritten, so an improvement reads as a step in the history, not a replacement of it.
What the public sees and what you see
Anyone can open your report and see the trust facts. Score, class and standing, the pillar bands, the proofs, the weekly standings. The per-dimension scores, the fix list, the change-points and the trade-off map are visible only to you when you are signed in as the owner. On the public page those sections are there, closed, with a note that only the owner can open them.
Where to go next
- How scoring works — the band mechanics.
- Battery versioning — why old scores keep their numbers.