docs(rfc): Spike for reconstructing the narrative

[robertmitchellv]

🔀 PULL REQUEST

💡 Summary

Adds RFC proposing how the refiner will rebuild a section's human-readable
<text> from the entries that survive pruning, starting with Results, Problems,
Medications, and Immunizations.

This is documentation only--no application code changes. The RFC records the
architecture decision and the reasoning behind it; implementation follows in a
later PR once the approach is reviewed.

In short, the chosen approach ("Option C") splits the work into three layers:
shared mechanical primitives written once (a typed-value renderer, a field
extractor, a table builder); per-section field maps as data (keyed by
template_id); and short per-section join functions as code that handle the
structural reach into ancestors/siblings. The RFC explains why this beats both a
fully-hardcoded approach and a fully-data-driven one, and how the field maps
become a down payment on the future template-aware matching engine without
building it now.

A runnable companion script (reconstruction_toy.py) demonstrates the design
end-to-end on a small eICR fragment.

Please review the RFC itself for the full context, options comparison, and
deferred questions:

• docs/rfc/0010-narrative-reconstruction.md

🔗 Related Issue

Fixes #1241

• [SPIKE] Narrative reconstruction for Results #1241

✅ Acceptance Criteria

🧪 How to test

You can run the toy example folded into the RFC if you'd like to test things out.

ℹ️ Additional Information

[github-actions]

🔒 Security Scan Results

⚠️ Found 15 vulnerabilities

Severity	Total
🟠 High	12
🟡 Medium	3

📦 refiner-app

✅ No vulnerabilities found

📦 refiner-lambda

Severity	Count
🟠 High	4

📦 refiner-ops

Severity	Count
🟠 High	8
🟡 Medium	3

---

View detailed results: Security tab
Last updated: 2026-06-11 16:17:34 UTC

[jakewheeler]

Great work Robert 🎉