Decision log¶
Status: Draft. Record label choices, deferred scope, and open questions as the documentation set evolves.
How to use this log¶
Each entry should state the decision or open question, alternatives considered, and consequences. Link to the chapter or checklist row affected. Reverse prior entries with a new dated note rather than silent edits.
2026-05-11: Top-level split: harness versus runtime¶
Decision (provisional): Keep harness and runtime as the primary architectural split in Harness versus runtime.
Alternatives: Single “platform” layer merging config and hosting; or three layers (product, harness, runtime).
Consequences: Production checklist rows map mainly to runtime chapters; domain quality remains harness-owned. May revisit if Tayne packaging blurs the boundary.
2026-05-11: Neutral labels over vendor names¶
Decision (provisional): Draft capability chapters use neutral headings; LangChain/LangSmith names appear in reference notes and mapping tables only.
Alternatives: Mirror vendor terminology throughout for faster onboarding on one stack.
Consequences: Extra translation step for readers deep in LangSmith docs; better portability for Transmute-Data discussions.
2026-05-11: One draft chapter per checklist row¶
Decision (provisional): Ten files under runtime-capabilities/ aligned to 01-production-requirements-runtime.md.
Alternatives: Single long runtime document; or appendix-only treatment.
Consequences: Easier parallel editing and linking; index maintenance overhead.
2026-05-11: Transmute-Data fit stays non-normative¶
Decision (provisional): Transmute-Data fit describes attachment points without committing STN/Tayne to specific runtime products.
Alternatives: Immediate binding matrix to shipped components.
Consequences: Avoids stale implementation claims; follow-up work needed for concrete mappings.
Open questions (unresolved)¶
| ID | Question | Touches |
|---|---|---|
| OQ-1 | Merge harness and runtime under one “platform” heading in the outline? | 00-outline.md, 03-harness-vs-runtime.md |
| OQ-2 | Which checklist rows stay first-class chapters versus appendices? | runtime-capabilities/ |
| OQ-3 | Where do offline evals end and online production evals begin? | 05-operations.md, 07-transmute-data-fit.md |
| OQ-4 | Default double-texting policy for chat products built on shared runtime? | real-time-interaction.md |
| OQ-5 | Retention defaults for traces versus long-term memory store? | 06-product-and-compliance.md, memory.md |