andreagrandi / andreagrandi/draftomen

Local pair matcher with residual routing to batched validation

Closed
#527 0 comments 0 reactions 0 assignees View on GitHub
draftomen enhancement size: M
Dominant language
Python
Stars
0
Forks
0
Avg merge
1h 18m
Merged PRs (30d)
197

Description

## Problem

Every role-compatible candidate pair costs a model call even when structured capability parameters already determine compatibility. The per-pair stage dominates enrichment cost (~$3.00 of ~$3.85 per set on HOB).

## Proposed change

Deterministic local matcher resolves pairs from v2 capability parameters (zone/action/qualifier compatibility); only unresolvable pairs enter the batched validation queue (20 per batch, implemented on branch enrichment-validation-batching); candidate construction records the resolution basis (local vs model) per pair.

## Acceptance Criteria

- [ ] AC1: >= 80% of HOB candidate pairs resolve locally (measured on stored benchmark results)
- [ ] AC2: Benchmark gate unchanged and green: accepted/rejected verdicts for model-validated residue match benchmark.json
- [ ] AC3: Matcher is pure/deterministic; omissions aggregate per mechanism with a distinct reason
- [ ] AC4: Engine sends only residual pairs to validation; full HOB re-run <= 60 validation calls

## Dependencies

Blocked by #526 (needs v2 structured parameters to match on).

## Implementation classification

- **Estimated size:** M
- **Orchestration risk:** High
- **Reason:** New matching logic whose precision directly affects recommendation quality; benchmark-gated.

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reviewing the v2 structured parameters from dependency #526 and the enrichment-validation-batching branch, then inspect the candidate construction and benchmark.json results. Implement and evaluate deterministic local resolution plus residual validation against the HOB acceptance criteria: at least 80% local resolution, unchanged verdicts, aggregated omission reasons, and no more than 60 validation calls.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.