# DEAD AGAIN 1.17 — SPEEDRUNNER DAVE

## Mission

Robot Dave stopped asking “is this solvable?” and started asking “which of the designer’s assumptions can I delete?”

The attack runs through the real `Game` engine. Speedrunner Mode deliberately tries fewer Ghost Anchors than the current certified solution, enumerates switch subsets, and asks state-space search to complete the room anyway.

## Result

The first full exploit pass found four meaningful sequence breaks. Two were boring bypasses and were patched. Two were clever, reproducible consequences of the world rules and were canonised.

### PATCHED — Level 7: TWO OF ME

Robot Dave reduced 2 deaths to 1 simply by vaulting the first door. That did not create an interesting alternate solution; it deleted half the lesson. Door #1 now owns its lane. The repaired room has been dynamically re-certified at **2 deaths**, and the lower-death subset attack no longer completes it.

### PATCHED — Level 10: AGAIN

Robot Dave reduced 2 deaths to 1 because the automatic lift could route around Door #2. Again, this was a geometry bypass rather than an emergent ghost trick. Door #2 is now full-height. The room has been dynamically re-certified at **2 deaths**, and the one-death attack is closed.

### CANONISED — Level 14: MACHINE ROOM

Robot Dave found a **zero-death** route through the existing machinery/geometry. It is a genuine executable sequence break rather than score laundering or collision corruption. We are keeping it. `perfectDeaths` is now **0** and the hint quietly acknowledges that not every moving part must be used.

### CANONISED — Level 42: SIX

The hidden room appeared to demand a procession of previous lives. Robot Dave found a repeatable **2-death short-circuit** using only the first two deliberate anchors and present-world interactions. It is weird, difficult and consistent with the rules, so it survives. `perfectDeaths` is now **2**.

## Final-level attack

ALIVE AGAIN remains certified at 4 deaths. Speedrunner Dave separately attacked every three-anchor subset (and all smaller subsets during the normal pass). None completed the room. This is strong evidence that the current 4-death certificate is robust under the present anchor-subset attack model.

## Design law reinforced

- Patch an exploit when it merely bypasses a named lesson or falsifies mastery.
- Keep an exploit when it is understandable, reproducible, and emerges from the world’s actual rules.
- When Robot Dave beats PERFECT legitimately, PERFECT moves.

## Important boundary

The attack is broad and executable, but it is not a mathematical proof of minimum deaths for every non-zero room. The next escalation is route-time optimisation and more general action-search that can discover lower-death solutions without expressing them primarily as Ghost Anchor subsets.