| DOES IT REPEAT | SEALED RE-RUN |
|---|
| Same page, asked twice | A different answer | The same verdict |
|---|
| Fix one section, re-run | A fresh review of everything | Only the part you edited can move |
|---|
| Vendor upgrades the model | Last month’s answer silently changes | Verdicts frozen under the version that made them |
|---|
| DOES IT SHOW ITS SOURCE | EVIDENCE LAYER |
|---|
| Quote traced to a document | Sounds right. Nobody checked | Matched against the fetched page, or dropped |
|---|
| Knows who wrote the evidence | No. A rival’s page and a buyer review are both text | Author class on every source, carried into the prompt |
|---|
| Flags a single-source finding | No | Yes, marked as such |
|---|
| DOES IT READ LIKE YOUR BUYER | BEHAVIOURAL STATES |
|---|
| Reader mindset | Patient. Reads to the bottom. Cooperative | Distracted, skeptical and ideal, mixed to match your channel |
|---|
| Reading behaviour | Careful reading only | Scans, hunts, abandons |
|---|
| How roles combine | One voice in several hats | Veto rules, not averages. One blocking role sinks it |
|---|
| Handoffs between roles | Not modelled | Seven chains, each scored healthy, at risk or broken |
|---|
| DOES IT REMEMBER | SIGNAL IDENTITY |
|---|
| Is this last week’s finding | No way to know there was a last week | Matched by meaning, so identity survives new wording |
|---|
| Do the buyers stay current | A saved prompt is frozen the day you write it | Re-compiled from live evidence on every run |
|---|
| Ever graded for accuracy | No. Nobody checks | Sealed before launch, then scored HIT or MISS by you |
|---|
| BOTH | BOTH |
|---|
| Time to an answer | Seconds | About 30 minutes |
|---|
| Writes your copy | Yes, and well | No. Not what we do |
|---|
| Best used for | Drafting, tightening, variants | The go or no-go call before you spend |
|---|