robbond/confidence-engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

experiment(confidence-engine): isolate reconstruction observation

Browse Source
This commit is contained in:
robbond
2026-09-06 10:31:45 +01:00
parent 215c783d11
commit 0bb2f01100
5 changed files with 60 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/current-handoff.md
+6
View File
@@ -145,6 +145,12 @@
- OpenAI strict transport projection now derives structure from canonical Zod-generated JSON Schema and canonical input optionality from the Zod contract, including `.optional().default(...)`. Canonically omittable fields are required-but-nullable for transport; null placeholders are omitted only for those fields before canonical Zod validation, while genuinely required fields remain non-nullable.
- `reconstructionV2Schema` and Ollama behavior remain unchanged. `START_CASE_EXPERIMENT_PROVIDER=openai` is experiment-only selection; default behavior remains Ollama/Qwen. Deterministic closeout passed with zero live calls; next boundary is exactly one live GPT-5.6 Terra reconstruction.
## Reconstruction-only OpenAI observation seam
- The first Terra observation was blocked before execution because normal `startCase()` would continue into Ollama-backed question generation. No Terra inference occurred.
- An experiment-only reconstruction-only seam now returns after initial reconstruction and graph construction, before `determineGraphBackedQuestion`. Normal production/default `startCase()` behavior remains unchanged; explicit OpenAI helper selection guarantees one selected reconstruction provider call and zero downstream question-generation calls.
- Zero live calls occurred while implementing and verifying this seam. Next boundary: exactly one live Terra reconstruction with no retry.
## Current product architecture
Three distinct routes, not a single page: