Each row shows what a condition supplies to the model and what it asks for. Rows with “→ continues” are later stages of the same chat as the referenced condition — the model retains everything from that earlier turn. C6+ and C6++ are exploratory extensions past the 7 core conditions (C1–C7); C6++'s own prompt asked the model to implement its C6+ self-critique; the resulting designs are included in the ATWL-side analysis.
| Condition | Supplied | Asked for | Prompt |
|---|---|---|---|
| C1 C1 | Task description only. | Free-form plan, no ATWL. | — (task file is the prompt) |
| C2 C2 | ATWL spec + one worked example (HITL). No library. | Prose explanation + full ATWL design. | txt prompt |
| C3 C3 | ATWL spec + example + 54-workflow library, pre-formalised in ATWL. | Design citing library workflows by name, then a source-paper list. | txt prompt |
| C5 C5 | → continues C3. Source papers for cited workflows are uploaded. | Method-level enrichment table per phase (structure held fixed), plus non-incorporated refinement notes. | txt prompt |
| C4 C4 | 54-workflow library, natural-language summaries only. No ATWL spec at this stage. | Prose-only design citing library workflows, then a source-paper list. | txt prompt |
| C6 C6 | → continues C4. ATWL spec + example now supplied. | Faithful ATWL translation of the earlier prose design — explicitly not a redesign — plus a self-report of added vs. translated structure. | txt prompt |
| C6+ C6+ | → continues C6. | Self-critique: does the ATWL representation reveal design gaps (unspecified loops, missing assess steps, implicit criteria)? Answered as a numbered list of recommended, not-yet-applied modifications. | txt prompt |
| C6++ C6++ | → continues C6+. | Attempt to implement the self-critique as a revised ATWL representation, with a changelog of what was/wasn't implemented and why. | txt prompt |
| C7 C7 | → continues C4 (alternative to C6). | Fresh ATWL redesign — explicitly not a translation of the prose version — then a summary of structural departures. | txt prompt |