infospace pipeline for wealth of nations example

This commit is contained in:
2026-05-14 18:04:38 +02:00
parent 8804461ca3
commit a729a7643e
26 changed files with 1124 additions and 32 deletions

View File

@@ -0,0 +1,39 @@
# Wealth/VSM Generation Pilot Report
## Summary
This pilot proves one-chapter regeneration for the Adam Smith Wealth/VSM
infospace using explicit `infospace-bench` workflows, not the legacy process
script.
This is not the legacy process script; it is the successor workflow shape.
## One-Chapter Target
- Source: Book I Chapter III
- Legacy alignment: `markitect-main/examples/infospace-with-history`
- Successor pilot: `infospaces/wealth-vsm-generation-pilot`
## Workflow Shape
- `wealth-vsm-extract-entities` produces a generated entity bundle and splits
it into stable entity artifacts.
- `wealth-vsm-map-and-analyze` produces a relation artifact and generated
chapter analysis.
- `wealth-vsm-evaluate-entities` produces evaluation Markdown files using
successor `artifact_id` semantics.
- `infospace-bench check` merges deterministic collection checks and generated
evaluations into metrics history.
## Comparison To Legacy
The legacy implementation centered on a large `process_chapters.py` script with
implicit provider execution and output conventions. This pilot keeps provider
work explicit through assisted requests and records generated outputs through
the manifest or evaluation history.
## Scale-Up Notes
Full-corpus generation should happen only after reviewing one live
provider-backed run. Cost, model, prompt version, retry settings, and resume
strategy should be recorded before processing more chapters.