rune/scripts/gen_memory_store_truth2.py

#!/usr/bin/env python3
"""gen_memory_store_truth2.py — memory-store batch 2 truth battery
(contract BUILDER-2026-10-07T18:05:47Z-152c16).

The LIVE _py_* oracles of memory_store.py. Cases: FTS conjunctive builder
(stopword filtering, fallback, 8-cap), chunk_text (recombine + hard-split),
rank fusion (hybrid/dense-only/lexical-only), project_chunk_rows,
map_reduce, fuse_results (dedup + scale-alignment).

Usage: python3 scripts/gen_memory_store_truth2.py   (from repo root)
"""

import json
import sys
from pathlib import Path

REPO = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(REPO / "src"))
sys.path.insert(0, str(REPO / "src" / "rune"))

import memory_store as ms  # noqa: E402


def main():
    cases = []

    def add(fn, name, inp, expected):
        cases.append({"fn": fn, "name": name, "input": inp, "expected": expected})

    # ── FTS conjunctive ───────────────────────────────────────────────────
    add("fts", "significant_terms",
        {"query": "the curator gates memory writes"},
        {"q": ms._py_fts_query_conjunctive("the curator gates memory writes")})
    add("fts", "stopwords_dropped",
        {"query": "the and was with nick"},
        {"q": ms._py_fts_query_conjunctive("the and was with nick")})
    add("fts", "short_fallback",
        {"query": "go is fun ai"},
        {"q": ms._py_fts_query_conjunctive("go is fun ai")})
    long_q = "alpha beta gamma delta epsilon zeta eta theta iota kappa"
    add("fts", "cap_eight",
        {"query": long_q},
        {"q": ms._py_fts_query_conjunctive(long_q)})
    add("fts", "empty_returns_raw",
        {"query": "a an of"},
        {"q": ms._py_fts_query_conjunctive("a an of")})

    # ── chunk_text ────────────────────────────────────────────────────────
    small = "one paragraph only"
    add("chunk", "under_size", {"text": small, "size": 1500, "overlap": 150},
        {"chunks": ms._py_chunk_text(small, 1500, 150)})
    para = "\n\n".join(f"paragraph {i} " + "x" * 40 for i in range(30))
    add("chunk", "recombine", {"text": para, "size": 400, "overlap": 150},
        {"chunks": ms._py_chunk_text(para, 400, 150)})
    noline = "y" * 3000
    add("chunk", "hard_split", {"text": noline, "size": 1000, "overlap": 150},
        {"chunks": ms._py_chunk_text(noline, 1000, 150)})
    add("chunk", "empty", {"text": "", "size": 1500, "overlap": 150},
        {"chunks": ms._py_chunk_text("", 1500, 150)})

    # ── rank fusion ───────────────────────────────────────────────────────
    dense = [[1, 0.9], [2, 0.8], [3, 0.7]]
    bm25 = [[2, 12.0], [4, 8.0], [1, 3.0]]
    add("fuse_ranks", "hybrid", {"dense": dense, "bm25": bm25},
        ms._py_rank_fusion(dense, bm25))
    add("fuse_ranks", "dense_only", {"dense": dense, "bm25": []},
        ms._py_rank_fusion(dense, []))
    add("fuse_ranks", "lexical_only", {"dense": [], "bm25": bm25},
        ms._py_rank_fusion([], bm25))
    add("fuse_ranks", "neither", {"dense": [], "bm25": []},
        ms._py_rank_fusion([], []))

    # ── project_chunk_rows ────────────────────────────────────────────────
    rows = [
        {"rowid": 1, "chunk_id": "c1", "text": "t" * 500, "wing": "w1", "room": "r1",
         "source_path": "a.md", "page": 3},
        {"rowid": 2, "chunk_id": "c2", "text": "short", "wing": "w1", "room": "r1",
         "source_path": "b.md"},
    ]
    ranks = {"2": 0.0625, "1": 0.05}
    add("project", "budget_and_shape", {"ranks": ranks},
        {"rows": ms._py_project_chunk_rows(ranks, rows, 6)})

    # ── map_reduce ────────────────────────────────────────────────────────
    entries = [{"id": i, "relevance": i * 0.1} for i in range(25)]
    add("map_reduce", "top3_per_chunk", {"entries": entries, "chunk_size": 10},
        {"mapped": ms._py_map_reduce(entries, 10)})
    add("map_reduce", "under_threshold", {"entries": entries[:5], "chunk_size": 10},
        {"mapped": ms._py_map_reduce(entries[:5], 10)})

    # ── fuse_results ──────────────────────────────────────────────────────
    ents = [
        {"id": 1, "text": "nick prefers coffee", "topic": "human",
         "relevance": 2.5, "confidence": 0.9},
        {"id": 2, "text": "NICK  PREFERS COFFEE", "topic": "human",
         "relevance": 1.1, "confidence": 0.8},  # dedup hit
        {"id": 3, "text": "rune gates memory writes", "topic": "project",
         "relevance": 5.6, "confidence": 0.95},
    ]
    res = [
        {"chunk_id": "c1", "text": "the spine rebuild parity", "wing": "research",
         "rrf": 0.0625, "confidence": 0.7},
        {"chunk_id": "c2", "text": "curator provenance recovery", "wing": "research",
         "rrf": 0.05},
    ]
    add("fuse_results", "dedup_scale_align", {"entries": ents, "research": res},
        {"merged": ms._py_fuse_results(ents, res)})

    out = {
        "oracle": "src/rune/memory_store.py _py_* oracles",
        "generated": "2026-10-07",
        "cases": cases,
    }
    target = REPO / "src/go/memory-store-deep" / "truth_recall.json"
    target.write_text(json.dumps(out, indent=1, ensure_ascii=False) + "\n")
    print(f"wrote {target} ({len(cases)} cases)")


if __name__ == "__main__":
    main()