Prompt strategies

# Prompt strategies

Prompt variants used in the benchmark grid. Each file is the full prompt template for one strategy; the matrix runner substitutes the topic title (and for `pK`, a per-topic target entry count derived from the reference).

## Active set (the shared 8-prompt grid)

| ID | File | Strategy |
|---|---|---|
| `pA` | `pA.txt` | Simple baseline |
| `pD` | `pD_cot.txt` | Silent plan-then-write |
| `pE` | `pE_historian_role.txt` | Historian role |
| `pF` | `pF_1shot.txt` | One-shot example |
| `pH` | `pH.txt` | Decompose + phase coverage |
| `pK` | `pK_helping_exact.txt` | Exact count control (per-topic `k` injected) |
| `pO` | `pO_information_dense_historian_hybrid.txt` | Dense historian hybrid |
| `pQ` | `pQ_information_dense_historian_expanded_coverage.txt` | Dense expanded coverage |

## Retired variants

`pB`, `pC`, `pG`, `pI`, `pJ`, `pL`, `pM`, `pN`, `pP`, `pR` were used in earlier experiment stages and are retained for provenance; `EXPERIMENT_10TL.md` and `RUN_CHECKLIST.md` document the original smaller-scale experiment design they belong to.

`topic_query_titles.json` maps topic slugs to the natural-language titles substituted into prompts; `topic_categories.json` holds the domain-category labels used in the grouped analyses.