The thesis in eleven steps
Each chapter asks one question and answers it with the stored evidence: a decision to try, a route to step through, the model's own reasoning, or the result cells behind a thesis table. Every chapter links to the full data.
- 01 · Sections 4.1–4.2The gameWhat decides a floor of Magic Tower?open
- 02 · Sections 4.5.1, 6.4One decisionWhich of four plausible first moves keeps the floor winnable?open
- 03 · Section 4.5.1, Table 4.4The optimal routeHow little room for error does the simple floor leave?open
- 04 · Sections 5.1–5.3, Listing 5.1What the model seesHow do the engine and the model share the work?open
- 05 · Section 4.6The exact solverWhere does the optimum come from, and why was the published one wrong?open
- 06 · Sections 6.2–6.3 (RQ1)The main resultDoes the hybrid agent win where the heuristic and the bare model do not?open
- 07 · Section 6.4.1, Appendix EHidden overridesWho really won the early episodes?open
- 08 · Sections 6.4.2–6.4.4One-sided annotationsWhat took the same model and prompt from 0 of 5 to 10 of 10?open
- 09 · Sections 6.3, 6.4Where episodes are lostAt which turn does a losing episode leave the winning one, and how far ahead must the engine look?open
- 10 · Section 6.5 (RQ2)Prompt optimisationDoes evolving the prompt help once the observation is good?open
- 11 · Sections 6.6–6.7, 7.5 (RQ3)Planning and the towerDo longer plans help, and how far up the 24-floor tower does the agent get?open