VIDRAFT_LAB

Agreed on the framing, and the push is fair.

25.7 (single-stream) and 601 (aggregate under concurrency) aren't the same axis — you're right. The 23.4× is a system-level number, and part of it is batching that any MoE gets for free. That isn't the VKAE claim, and we'll present it as what it is rather than fold it into the speedup.

What the paper leads with is the controlled comparison: identical concurrency, batch, and harness, VKAE on vs. off. Single-stream and the throughput ceiling stay as separate, labeled context.

Same discipline on quality — the tail, not the mean: long-context recall and the hard subset, measured against VKAE-off on identical inputs. If it doesn't hold there, it doesn't ship.

To your question directly: VKAE-on vs. VKAE-off at identical concurrency. That's the headline.

You've pinned it exactly. Surprise rate turns the denominator — the modeled surface — from an assertion into a measurement: it makes the coverage number falsifiable.

Honest answer to your question: today Chitos records what Phase 3 fired (the numerator), but it does not yet formally diff "nodes actually reached" against "nodes the static model predicted reachable" and emit that as a surprise rate. That's precisely the gap you've named, and it's the right next step.

Here's how we intend to fold it in: every time a Phase 3 payload lands on a node the static model marked unreachable, we log it as a surprise, feed it back into the enumerator (the surface model gets falsified by its own execution), and report it per phase and per vuln class alongside each "not demonstrated." A low surprise rate is then what earns the coverage number, rather than us asserting it.

One extension: surprise rate is most meaningful stratified by vuln class (1% on the SQLi surface and 1% on the XSS surface don't mean the same thing), and to separate "accurate model" from "blunt prober" it has to be read together with a payload-diversity measure — a prober incapable of surprising will trivially report a low rate. Thanks for this; it's the kind of push that keeps the tool honest.