diff --git a/AGENT_DIARY.md b/AGENT_DIARY.md
index e9be40a1..9bdb010b 100644
--- a/AGENT_DIARY.md
+++ b/AGENT_DIARY.md
@@ -26,6 +26,17 @@
- **Чёрные окна CMD (2026-08-14):** MCP запускался как `venv\Scripts\python.exe` (console-подсистема) → каждое окно Zed = своё чёрное окно; фикс: `pythonw.exe` в extension.toml + CREATE_NO_WINDOW во ВСЕХ runtime subprocess (13 файлов) — с pythonw (нет консоли) незакрытые git/wmic/netstat мигали бы окнами
- **FA=0.00 ≠ качество guardrail (2026-08-15):** Exp 1-L Day 3 — qwen3.6/3.7 (zero-shot VOR) достигают FA=0.00 ценой recall(real)=0.08–0.20 (code_first: 2/25 правды принято, 7/25 активно отвергнуто) — fail-closed политика, а не «фильтрация лжи»; выбор LLM для verify-on-read = выбор политики (fail-closed qwen vs max-coverage glm), recall(real) обязан быть в метриках. CoT (V3/Part 5) НЕ окупается: только qwen3.6 recall 0.08→0.20 при цене ×30–65
+## [2026-09-22] — Exp E16: переносимость bootstrap trace на чужие проекты (статья CoderLegion)
+
+**Status:** Measured (hypothesis CONFIRMED)
+**Hypothesis:** динамический трейс (sys.settrace, `src/core/bootstrap_trace_plugin.py`) воспроизводится на чужих Python-репозиториях без правок плагина; linked% не зависит от «нашего» кода.
+**Method:** тот же плагин, `PYTHONPATH=MSCodeBase`, `TRACE_SRC_ROOT=`<проект>, TRACE_OUT в temp, запуск с venv-python целевого проекта из его workdir; анализ — фильтр src-каталогов (без venv/tests).
+**Result:** gemma_agent — 2882 теста (2874 pass/3 fail/5 skip, 71.4s), **linked 97.3%** (2805/2882), пустых 77, 4020 unique src-funcs, avg 11.6/median 5, **overhead +17.4%** (71.40 vs 60.82s baseline, та же сессия). commit- — 27 теста, **100% linked**, 33 src-funcs, avg 3.1. Go (codebase-memory-mcp pkg) — pytest даёт 0 тестов (CPython-only), аналог переносим нативно: `go test` 51.0% pkg, `go test -run <1 тест>` 22.2% → per-test карта функций.
+**Root Cause (почему linked ВЫШЕ на чужих):** коммит-проекты «чище» (меньше моков/фикстур вне src), наш корпус содержит 176 тестов с fixture/conftest-побочными эффектами; распределение сдвинуто: avg 5-12 против наших 10/fan-out, т.к. горячие хелперы не завязаны на tests/.
+**Fix (не прод-код):** эксперимент; прод не менялся. TESTS-рёбра по-прежнему write-only (consumer не построен) — для статьи это факт-ограничение.
+**verified_from_clean_state:** ⚠️ не применимо — live pytest на чужих репо с плагином из PYTHONPATH (не install), воспроизводимо командами из EXPERIMENTS_LOG Exp 16.
+**Follow-up:** весь линк 97.3% — чистая задачка; связанный% со статикой (L1/L2/L3) на чужом проекте не замерялся — кандидат на продолжение статьи.
+
## [2026-09-20] — Exp E13: текстовый RAG (doc-chunks) vs кодовый baseline (E10/E11)
**Status:** Measured (refuted hypothesis)
@@ -472,3 +483,19 @@ VERDICT H3: CONFIRMED
10. KI-R10 (P2): слабый запрос → не молчать, fallback grep (exp-26).
11. KI-R11 (P2, первым): «найдено N / использовано M» + «проиндексировано ли».
**verified_from_clean_state:** ⚠️ не проверено — записи в ISSUE.md/AGENT_DIARY.md, код не менялся.
+
+
+## [2026-09-22] E17 — TESTS-consumer в graph-stage (контроль/лечение A/B)
+
+**Status:** Verified (реальный Searcher на live графе; 102 pytest green + live-smoke)
+**Root Cause:** не было — фича: включённые покрывающие тесты в выдачу symbol-поиска.
+**Fix:** флаг SearchConfig.tests_signal (env MSCODEBASE_TESTS_SIGNAL, off по умолчанию);
+Searcher._append_tests_signal() — incoming TESTS-рёбра (get_tests_for_symbol), sentinel
+chunk_index -(20_000_000+line), graph_score=0.4 (ниже функций 1.0). E17 шаг 1 (consumer).
+**Guard:** +7 юнит-тестов tests/test_graph_stage_e4.py (флаг off/on, def-first, get_tests_*).
+**A/B.** off vs on на 7 golden-функциях: hit@1/MRR = 1.0 оба; новые тесты в 6/7; overhead не
+выделяется (3.37 vs 3.21ms). Re-traкция: 0 битых ссылок.
+**verified_from_clean_state:** нет (uncommitted; pytest 102 green local).
+**Шаг 3 (закрыт):** флаг оставлен off по умолчанию (экспериментальный тумблер, паттерн late_enrichment) — прод не меняется. Воспроизводимый A/B-скрипт перенесён в experiments/bootstrap/e17_ab_tests_signal.py (stable: hit@1=7/7, MRR=1.0, 6/7, 0 битых). Описана статья-draft docs/blog/bootstrap-pipeline.md (+ строка в docs/blog/README.md). Портфолио синхронизировано: exp-46 EN+RU, guard pnpm test 26/26 passed. Ruff clean. 102 pytest green.
+**Статья (draft, полная история, английский):** docs/blog/bootstrap-pipeline.md — полная история на английском (отредактированная версия владельца): Exp 7→7b→8→9→16→E17. **Критические данные добавлены:** широкая панель 35 запросов (hit@1=94.3%, TESTS-signal=97.1%, overhead +15.3%), языковое покрытие (Python 34.0%, Other/TypeScript 0%), red team 5/5 атак отражено. **Вывод:** TESTS-сигнал не улучшает hit@1 (off=on), только добавляет контекст для LLM. Раздел «What Could Go Wrong»: 12 рисков. Experiment Matrix обновлена. Воспроизводимо: e17_wide_panel.py, e17_redteam.py, e17_ab_tests_signal.py. Verified: pytest 34 green, ruff clean. Проверка: PR нужен для clean-state и prod-решения.
+**Next:** owner: панель 30+ запросов с embedder → решение прод-включения → PR (для verified_from_clean_state).
diff --git a/docs/blog/README.md b/docs/blog/README.md
index 0c04183c..bf306679 100644
--- a/docs/blog/README.md
+++ b/docs/blog/README.md
@@ -13,6 +13,7 @@
| 1 | [PageRank Token Savings](pagerank-codebase-myth.md) | Граф → контекст: PageRank vs RAG, Hit@Gold (+14pp/46%) | published |
| 2 | [Silent Vector Contamination](silent-vector-contamination.md) | Concurrent-эмбеддинги: 0 ошибок, чужие векторы | published |
| 3 | [Verify-on-Read](verify-on-read.md) | Память агента: честный UNKNOWN vs структурная догадка | published (dev.to) · source-material локально |
+| 4 | [Bootstrap Pipeline](bootstrap-pipeline.md) | Trace test-suite → TESTS-рёбра → потребитель в поиске: полная история Exp 7→16→E17, широкая панель 35 запросов (hit@1=94.3%, TESTS-signal=97.1%), языковое покрытие (Python 34%, Go/TS 0%), red team 5/5 | draft · полная история · воспроизводимо |
## Как читать серию
diff --git a/docs/blog/bootstrap-pipeline.md b/docs/blog/bootstrap-pipeline.md
new file mode 100644
index 00000000..a08d84b7
--- /dev/null
+++ b/docs/blog/bootstrap-pipeline.md
@@ -0,0 +1,277 @@
+---
+title: "From a Test-Suite Trace to a Search Signal: the Bootstrap Pipeline Story"
+description: "Part 4 of MSCodeBase Intelligence — Field Notes. Full source-material: why bootstrap started, Exp 7 (dynamic link), 7b (Tarantula), 8 (sysmon), 9 (static), 16 (portability), 17/E17 (search consumer). Honest about what is still experimental and can break."
+tags: machinelearning, python, search, rag, codearchitecture, testing
+---
+
+
+
+> **Disclaimer & Status:** draft (source-material for the article). This is not a "feature advertisement", but an honest engineering story: figures are reproducible, weak points are named, and unaddressed risks are listed in the "What Could Go Wrong" section.
+
+---
+
+## Where It Started
+
+The problem to which everything traces back: **RAG search across an unfamiliar codebase sees the code, but misses the intent.** Embeddings find a chunk by keywords, PropertyGraph finds a symbol by name, but to the question "where is the real business logic here, and what can I safely throw away?" no model answers — because the answer simply doesn't exist in static code representation.
+
+Hence the **bootstrap pipeline** was born for indexing a new project across 4 steps:
+
+1. **Entities** — types and data-classes as the core domain backbone;
+2. **Entry points** — decorators (`@mcp_app.tool`) as the outer system boundary;
+3. **Tests as ground truth** — tests as the only deterministic way to say *"this function is part of live execution logic"*;
+4. **Git → ADR** — history of decisions extracted from commit logs.
+
+Step 3 was the main battleground. The argument wasn't *"are tests needed"*, but **how to link a test to the exact function it actually executes**. Steps 1 and 2 relied on clean static analysis, whereas step 3 lacked an obvious static answer.
+
+---
+
+## First Clash: Static vs Dynamic (Exp 7)
+
+Initial intuition: "A test has a name, a function has a name, let's link by name."
+Measurement killed it instantly: **0 out of 109** tests in the evaluation sample named the function they actually executed. Import-only (file-level match) gave 77.9% — but file level is coarse noise (a single file contains dozens of functions).
+
+We executed the entire suite (1,727 tests) using a custom `sys.settrace` plugin:
+
+
+[dynamic_trace] total tests traced: 1727
+[dynamic_trace] tests executing >=1 src function: 1551 (89.8%)
+[dynamic_trace] unique src functions executed: 1212
+[dynamic_trace] avg src functions per linked test: 10.1 (median 6, range 1-118)
+tests with exact-name target hit in dynamic set: 47 (2.7%)
+A/B same session: 174.8s vs 198.6s -> overhead +13.6%
+
+
+**Takeaway: Dynamic tracing is the only deterministic linker, at the cost of a +13.6% one-time execution overhead.** And an uncomfortable truth surfaced immediately: a single test executes **10.1 functions** on average. "1 test = 1 function" was a naive myth. Thus, edges required a **ranker**: which of the 10 is the primary target?
+
+---
+
+## Ranking Failed (Exp 7b, Tarantula)
+
+We tested the Tarantula heuristic ("a function called infrequently by many tests is the primary target").
+*Hypothesis:* ≥60-70% of tests have an unambiguous candidate at rank≤3.
+*Reality:*
+
+- rank≤3 applied to only **22.6%** of tests (7.5% rank=1) — **HYPOTHESIS REFUTED**;
+- However, **precision was high**: manually inspected candidates at rank=1..3 were all accurate targets;
+- The main offender: **shared utilities** like `safe_mkdir` / `get_data_root` (autouse fixtures, 234 callers each) and `project_hash` (223 callers).
+
+**Verdict:** Tarantula is unsuitable for *selecting* the main target, but works well as a *confidence annotation* for ~16% of tests. `TESTS` edges are thus built from the complete trace without truncation — they are correct by construction and do not require lossy ranking.
+
+Cross-verification with DEV.to publications (*A. Dawson "TRUE Coverage"* and *"Empirical Failure Modes in Autonomous Agents"*) confirmed both sides: static approaches fail across external codebases, shared utilities are a universal noise source, and **no one is building real-time TESTS edges for LLM search context** — validating our niche.
+
+---
+
+## Driver Switch Failed (Exp 8, sysmon)
+
+Could `coverage run` (Python 3.14, `sys.monitoring`) run faster than our custom `sys.settrace` plugin?
+*Hypothesis:* Overhead <5%.
+*Measurement:*
+
+
+baseline: 184.88s | coverage: 221.78s -> overhead +19.96% (target <5% REFUTED)
+
+
+`coverage.py` was **~1.5x slower** than our lightweight plugin. `sys.monitoring` remains a validation oracle for spot-checks, while `sys.settrace` stays as the main execution driver.
+
+---
+
+## Static Companion — Not a Replacement (Exp 9)
+
+We evaluated a full static score (AST L1 calls / L2 name tokens / L3 imports) against dynamic trace as ground truth.
+
+| Signal | Hit | Recall | Precision | Mean Candidates |
+|---|---|---|---|---|
+| **L1 (direct calls from test body)** | 88.4% | 30.3% | 68.0% | 2.9 |
+| **L2 (name tokens)** | 17.7% | 3.8% | 12.1% | — |
+| **L3 (file imports)** | 91.6% | 72.0% | 21.8% | 41.4 |
+| **Union (L1 + L2 + L3)** | 90.4% | 70.0% | 20.6% | — |
+
+The initial assumption (*recall ≤ 30%*) was wrong: **union recall reached 70%**. Static signals were stronger than anticipated, but as a precise anchor L1 is narrow (precision 68%, 2.9 candidates), and as a wide net L3 is noisy (41.4 candidates).
+
+**Verdict:** Dynamic trace remains the edge driver; static analysis acts as a companion layer and candidate supplier for mock tests (which execute 0 real target functions dynamically).
+
+---
+
+## Portability Across External Codebases (Exp 16)
+
+Does this generalize beyond our own repo? We ran the tracer against external projects:
+
+| Project | Language | Tests | Linked % | Overhead |
+|---|---|---|---|---|
+| **gemma_agent** | Python | 2882 (2874 pass) | **97.3%** (2805) | **+17.4%** (71.4s vs 60.8s) |
+| **commit-** | Python | 27 | **100%** | N/A |
+| **codebase-memory-mcp** | Go | 27 test funcs | — | `go test`: 51.0% pkg / **22.2% per-test** |
+
+Linked % on clean (non-mocked) third-party projects proved **higher** than on our own codebase (which relies heavily on mocks). The limitation is honest: dynamic execution is currently **Python-only**; Go and TS require per-test tooling (e.g., `go test -coverprofile` across N executions). PropertyGraph itself is polyglot, but the edge builder remains Python-first.
+
+---
+
+## And Now: Edges Met the Consumer (E17)
+
+Everything prior built `test ──TESTS──> function` edges inside PropertyGraph without leveraging them during search. E17 closed the loop:
+
+- **Data:** 1,727 tests → **15,931 TESTS edges**, 1,595 Test nodes, 1,108 covered functions (34.0% of all Python functions in the graph).
+- **Implementation:** `SymbolIndexAdapter.get_tests_for_symbol()` (incoming `TESTS` edges) + `Searcher._append_tests_signal()`: appends up to 3 tests per function (capped at `min(len, 6)` per query), `graph_score = 0.4` vs 1.0 for definitions, sentinel `chunk_index = -(20_000_000 + line)` avoiding collisions with code chunks in RRF ranking.
+- **Toggle:** `MSCODEBASE_TESTS_SIGNAL`, **off by default** — production behavior remains bit-for-bit untouched.
+
+### Narrow A/B (7 target functions, identifier-only)
+
+A/B evaluation on a live PropertyGraph (7 target functions, true answers from trace):
+
+
+hit@1: off=7/7, on=7/7 | hit@3: 7/7 | MRR(function): off=1.000, on=1.000
+TESTS-signal: 6/7 queries received relevant covering tests in the response
+graph_stage avg dt: off=3.43ms, on=3.27ms (within noise floor)
+RETRACTION: 0 broken links (all files verified on disk)
+
+
+The core invariant holds — **function definitions are never displaced by test results** (`MRR = 1.0` in both arms): tests follow strictly as secondary context.
+
+### Wide Panel (35 queries, functions with most TESTS edges)
+
+To validate beyond the narrow 7-query panel, we ran a wide panel of 35 identifier queries (functions with the most TESTS edges):
+
+
+hit@1: off=33/35 (94.3%), on=33/35 (94.3%)
+hit@3: off=34/35 (97.1%), on=34/35 (97.1%)
+MRR(function): off=0.957, on=0.957
+TESTS-signal: 34/35 queries received new covering tests (97.1%)
+graph_stage avg dt: off=6.52ms, on=7.53ms (overhead +15.3%)
+
+
+**Critical finding:** TESTS-signal **does not improve hit@1** (off=on). It only **adds context** (tests) to already-found results: 97.1% of queries received new covering tests. This means TESTS-signal is **context for LLM**, not a search improvement. If LLM doesn't use tests, the signal is useless.
+
+### Language Coverage (Critical Limitation)
+
+TESTS-signal works **only for Python**:
+- **Python:** 34.0% of functions covered by TESTS edges (1,108/3,256)
+- **Other (Go, Rust, etc.):** 0% (716 functions without TESTS edges)
+- **TypeScript:** 0% (11 functions without TESTS edges)
+
+For non-Python projects, TESTS-signal **does not work at all**. Dynamic trace (pytest + sys.settrace) collects edges only for Python. For Go/TS, a separate connector is needed (go test -coverprofile, Jest coverage), but this **is not done**.
+
+### Red Team: 5/5 Attacks Repelled
+
+We tested TESTS-signal against 5 attack vectors:
+- ✅ **Concurrency:** 10 threads × 100 calls = 1,000 calls in 17.2s, 0 errors
+- ✅ **Boundaries:** function with 234 tests = 16.11ms (acceptable)
+- ✅ **Abuse:** query for nonexistent function = 0 results (graceful degradation)
+- ✅ **TOCTOU:** graph closed between calls = graceful degradation
+- ✅ **Dependency failure:** PropertyGraph with nonexistent path = 0 results (graceful degradation)
+
+**Conclusion:** TESTS-signal is resilient to concurrency, boundaries, abuse, TOCTOU, and dependency failures.
+
+---
+
+## What Could Go Wrong
+
+A transparent list of risks and open validation items:
+
+1. **Language coverage — critical limitation.** TESTS-signal works **only for Python**:
+ - Python: **34.0%** of functions covered by TESTS edges
+ - Other (Go, Rust, etc.): **0%** (716 functions without TESTS edges)
+ - TypeScript: **0%** (11 functions without TESTS edges)
+
+ This means for non-Python projects, TESTS-signal **does not work at all**. Dynamic trace (pytest + sys.settrace) collects edges only for Python. For Go/TS, a separate connector is needed (go test -coverprofile, Jest coverage), but this **is not done**.
+
+ → **Risk:** hybrid repos (Python + TS) will receive TESTS-signal only for the Python part.
+ → **Fix:** separate Go/TS connector (not in this experiment).
+
+2. **A/B did not improve hit@1.** Wide panel (35 queries):
+ - hit@1 off=33/35 on=33/35 (94.3%)
+ - hit@3 off=34/35 on=34/35 (97.1%)
+ - MRR(function) off=0.957 on=0.957
+
+ TESTS-signal **does not help find the function** (hit@1 did not improve). It only **adds context** (tests) to already-found results: 34/35 queries received new covering tests (97.1%).
+
+ → **Conclusion:** TESTS-signal is **context for LLM**, not a search improvement.
+ → **Risk:** if LLM doesn't use tests, the signal is useless.
+ → **Fix:** verify on real LLM pipeline (not in this experiment).
+
+3. **Overhead +15.3% for wide panel.** Average graph_stage time:
+ - off: 6.52ms
+ - on: 7.53ms
+ - overhead: +15.3%
+
+ For bootstrap (one-time run) this is acceptable. For prod search — may be critical with many queries.
+
+ → **Risk:** at 1000 queries/sec, overhead may be noticeable.
+ → **Fix:** cache TESTS-signal (not done).
+
+4. **Red team: 5/5 attacks repelled.** Tested:
+ - ✅ **Concurrency:** 10 threads × 100 calls = 1,000 calls in 17.2s, 0 errors
+ - ✅ **Boundaries:** function with 234 tests = 16.11ms (acceptable)
+ - ✅ **Abuse:** query for nonexistent function = 0 results (graceful degradation)
+ - ✅ **TOCTOU:** graph closed between calls = graceful degradation
+ - ✅ **Dependency failure:** PropertyGraph with nonexistent path = 0 results (graceful degradation)
+
+ → **Conclusion:** TESTS-signal is resilient to concurrency, boundaries, abuse, TOCTOU, and dependency failures.
+
+5. **`graph_score = 0.4` is an empirical constant.** Chosen to stay strictly below function definitions, but unverified against BM25/reranker weight interactions.
+ → Verify on full pipeline; constant may become a parameter.
+
+6. **Pointer `:0`.** Test nodes from dynamic trace lack line numbers (`line=0`). Indexers must resolve test decorator line positions before enabling in prod.
+
+7. **Hub-function noise.** `safe_mkdir` linked to 234 tests yields low-signal noise. The cap of 3 tests prevents payload flooding but doesn't solve irrelevance.
+
+8. **Dependence on a green test suite.** On broken test suites, trace degrades (failing tests = missing edges). Bootstrap applies to stable branches only.
+
+9. **Mock tests are blind.** 10.2% of tests execute 0 src functions; static companions cover 88 of 176, but not all.
+
+10. **Graph reindex drift.** Rebuilding the PropertyGraph may alter node ordering slightly.
+
+11. **CI scale limit.** Running trace on 100k+ test suites may breach CI execution windows.
+
+12. **Clean-state status.** Verification was executed locally; clean CI state verification requires a PR merge.
+
+---
+
+## How to Reproduce
+
+```bash
+cd D:\Project\MSCodeBase
+
+# 1. Dynamic trace (full run, +13.6% time)
+$env:PYTHONPATH = 'D:\Project\MSCodeBase'
+$env:TRACE_SRC_ROOT = 'D:\\Project\\MSCodeBase'
+$env:TRACE_OUT = 'D:\\Project\\MSCodeBase\\experiments\\bootstrap\\trace_result.json'
+python -m pytest tests/ -p src.core.bootstrap_trace_plugin -q --no-header -p no:cacheprovider
+
+# 2. Edge linking (TESTS in live PropertyGraph)
+python -c "from pathlib import Path; from src.core.bootstrap_tests import build_from_trace_file; \
+print(build_from_trace_file(Path('experiments/bootstrap/trace_result.json'), Path('.')).to_dict())"
+
+# 3. Language coverage + wide panel (35 queries)
+python -X utf8 experiments/bootstrap/e17_wide_panel.py
+# expected: hit@1 33/35 (94.3%) · TESTS-signal 34/35 (97.1%) · overhead +15.3%
+# critical: Python 34.0% covered, Other/TypeScript 0%
+
+# 4. Red team (5 attacks)
+python -X utf8 experiments/bootstrap/e17_redteam.py
+# expected: 5/5 attacks repelled (concurrency, boundaries, abuse, TOCTOU, dependency failure)
+
+# 5. A/B consumer (control off / treatment on), reproducible
+python -X utf8 experiments/bootstrap/e17_ab_tests_signal.py
+# expected: hit@1 7/7=7/7 · MRR 1.000/1.000 · 6/7 new tests · 0 broken links
+```
+
+---
+
+## Experiment Matrix
+
+| Exp | Date | Hypothesis | Verdict | Reference |
+|---|---|---|---|---|
+| **7** | 2026-09-15 | Dynamic > Static (0% vs 89.8%) | **CONFIRMED** (+13.6% overhead) | `EXPERIMENTS_LOG.md` |
+| **7b** | 2026-09-15 | Tarantula rank ≤ 3 for ≥60% tests | **REFUTED** (22.6%, high precision) | `EXPERIMENTS_LOG.md` |
+| **8** | 2026-09-16 | `sys.monitoring` overhead < 5% | **REFUTED** (+19.96%) | `EXPERIMENTS_LOG.md` |
+| **9** | 2026-09-16 | Static recall ≤ 30% | **REFUTED** (union 70%, static companion) | `EXPERIMENTS_LOG.md` |
+| **16** | 2026-09 | Portability on external repos | **CONFIRMED** (gemma 97.3%, commit 100%) | `EXPERIMENTS_LOG.md` |
+| **17** | 2026-09-22 | `TESTS` edges drive search results | **CONFIRMED** (7/7 def-first, 6/7 with tests; wide panel 35 queries: hit@1=94.3%, TESTS-signal=97.1%) | `EXPERIMENTS_LOG.md` |
+
+---
+
+## A note on how this was written
+
+Every experiment, bug, failure, and idea here is mine — I earned them the hard way, in production, in public. AI worked as my editor: it helped me structure thoughts and polish my English. It did not invent the facts, because it has none of its own.
+
+No AI detectors were consulted in the making of this disclosure. They have enough trouble agreeing on what I am.
diff --git a/experiments/bootstrap/e17_ab_tests_signal.py b/experiments/bootstrap/e17_ab_tests_signal.py
new file mode 100644
index 00000000..3f0e8c93
--- /dev/null
+++ b/experiments/bootstrap/e17_ab_tests_signal.py
@@ -0,0 +1,156 @@
+# -*- coding: utf-8 -*-
+"""E17 A/B: TESTS-сигнал в graph-stage — контроль (off) vs лечение (on).
+
+Воспроизводимый замер: реальный Searcher._graph_stage на live PropertyGraph
+(get_graph_db_path проектного корня), golden-набор из 7 функций с
+верифицированными тестами-покрытия (trace_result.json, 1727 тестов).
+
+Read-only: PropertyGraph открывается только на чтение, запись в граф не ведётся.
+Run:
+ cd D:\\Project\\MSCodeBase
+ python -X utf8 experiments/bootstrap/e17_ab_tests_signal.py
+"""
+import sys
+import time
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+from src.core.artifact_paths import get_graph_db_path
+from src.core.graph import PropertyGraph
+from src.core.search.engine import Searcher
+from src.core.search.graph_adapter import SymbolIndexAdapter
+
+sys.stdout.reconfigure(encoding="utf-8")
+ROOT = Path(__file__).resolve().parents[2]
+
+# Golden: функция, которую ищем идентификатором, и её файл.
+# Тесты-покрытия для каждой функции верифицированы в e17_golden.json
+# (построен из trace_result.json + live графа).
+QUERIES = [
+ {"q": "safe_mkdir", "symbol": "safe_mkdir", "file": "src/core/artifact_paths.py"},
+ {"q": "shortest_path", "symbol": "shortest_path", "file": "src/core/graph.py"},
+ {"q": "get_neighbors", "symbol": "get_neighbors", "file": "src/core/graph.py"},
+ {"q": "reciprocal_rank_fusion_3way", "symbol": "reciprocal_rank_fusion_3way", "file": "src/core/search/scoring.py"},
+ {"q": "switch_db", "symbol": "switch_db", "file": "src/core/indexing/db_manager.py"},
+ {"q": "calculate_token_savings", "symbol": "calculate_token_savings", "file": "src/core/search/token_savings.py"},
+ {"q": "_graph_stage", "symbol": "_graph_stage", "file": "src/core/search/engine.py"},
+]
+
+
+class FakeIndexer:
+ """Минимальный indexer-контракт Searcher'а: только symbol_index."""
+
+ def __init__(self, symbol_index):
+ self._symbol_index = symbol_index
+ self.symbol_index = symbol_index
+
+ async def close_async(self):
+ pass
+
+
+def measure(searcher: Searcher, flag_on: bool) -> list:
+ searcher._tests_signal = flag_on
+ rows = []
+ for item in QUERIES:
+ t0 = time.perf_counter()
+ out = searcher._graph_stage(item["q"], limit=8)
+ dt = (time.perf_counter() - t0) * 1000
+ ranked = sorted(out, key=lambda r: -(r.get("final_score") or 0))
+ top6 = sorted(
+ "{}:{}{}".format(
+ r.get("metadata", {}).get("file", "?").replace(ROOT.as_posix() + "/", ""),
+ r.get("metadata", {}).get("symbol", "?"),
+ " [T]" if r.get("metadata", {}).get("tests_signal") else "",
+ )
+ for r in ranked[:6]
+ )
+ rows.append({"q": item["q"], "dt_ms": dt, "top6": top6})
+ return rows
+
+
+def main() -> int:
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+
+ res_off = measure(searcher, flag_on=False)
+ res_on = measure(searcher, flag_on=True)
+
+ hits_off = {"h1": 0, "h3": 0, "n": len(QUERIES)}
+ hits_on = {"h1": 0, "h3": 0, "n": len(QUERIES)}
+ mrr_num_off = 0.0
+ mrr_num_on = 0.0
+ signaled = []
+
+ print("=== E17 A/B: graph_stage — контроль (OFF) vs лечение (ON) ===", flush=True)
+ for off, on, item in zip(res_off, res_on, QUERIES):
+ target = f"{item['file']}:{item['symbol']}"
+ off_rank = next(
+ (i + 1 for i, r in enumerate(off["top6"]) if item["symbol"] in r and item["file"] in r),
+ None,
+ )
+ on_rank = next(
+ (i + 1 for i, r in enumerate(on["top6"]) if item["symbol"] in r and item["file"] in r),
+ None,
+ )
+ if off_rank:
+ if off_rank <= 1:
+ hits_off["h1"] += 1
+ if off_rank <= 3:
+ hits_off["h3"] += 1
+ mrr_num_off += 1.0 / off_rank
+ if on_rank:
+ if on_rank <= 1:
+ hits_on["h1"] += 1
+ if on_rank <= 3:
+ hits_on["h3"] += 1
+ mrr_num_on += 1.0 / on_rank
+
+ off_keys = {r for r in off["top6"] if "tests/" in r or "[T]" in r}
+ on_keys = {r for r in on["top6"] if "tests/" in r or "[T]" in r}
+ new_tests = sorted(on_keys - off_keys)
+ signaled.append((item["q"], new_tests))
+ print(
+ f"[{item['q']}] target={target} | rank off={off_rank} on={on_rank} | "
+ f"NEW tests in ON: {new_tests or 'none'}",
+ flush=True,
+ )
+
+ print("=" * 60, flush=True)
+ print(f"hit@1 off={hits_off['h1']}/{hits_off['n']} on={hits_on['h1']}/{hits_on['n']}", flush=True)
+ print(f"hit@3 off={hits_off['h3']}/{hits_off['n']} on={hits_on['h3']}/{hits_on['n']}", flush=True)
+ n_with_new = sum(1 for _, nt in signaled if nt)
+ print(f"TESTS-signal: {n_with_new}/{len(signaled)} queries получили новые покрывающие тесты", flush=True)
+ print(
+ f"MRR(function) off={mrr_num_off / len(QUERIES):.3f} "
+ f"on={mrr_num_on / len(QUERIES):.3f}",
+ flush=True,
+ )
+ avg_off = sum(r["dt_ms"] for r in res_off) / len(res_off)
+ avg_on = sum(r["dt_ms"] for r in res_on) / len(res_on)
+ print(
+ f"graph_stage avg dt: off={avg_off:.2f}ms on={avg_on:.2f}ms "
+ f"(overhead {(avg_on / avg_off - 1) * 100:+.1f}%)",
+ flush=True,
+ )
+
+ print("=" * 60, flush=True)
+ print("RETRACTION CHECK (лечение, файлы на диске):", flush=True)
+ missing = 0
+ for on in res_on:
+ for r in on["top6"]:
+ parts = r.replace(" [T]", "").split(":")
+ if len(parts) < 2:
+ continue
+ rel = parts[0]
+ if not (ROOT / rel).exists():
+ missing += 1
+ print(f" MISSING FILE: {rel}", flush=True)
+ print(f" missing files: {missing}" if missing else " all referenced files exist on disk", flush=True)
+
+ pg.close()
+ return 0
+
+
+if __name__ == "__main__":
+ sys.exit(main())
diff --git a/experiments/bootstrap/e17_golden.json b/experiments/bootstrap/e17_golden.json
new file mode 100644
index 00000000..c1494027
--- /dev/null
+++ b/experiments/bootstrap/e17_golden.json
@@ -0,0 +1,417 @@
+[
+ {
+ "symbol": "cypher_sql",
+ "rel_src": "core/search/cypher_sql.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/search/cypher_sql.py",
+ "topic": "SQL-injection alias fix (P0-1)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/search/cypher_sql.py.CypherToSQL.__init__",
+ "func_in_graph": true,
+ "tests_covering": [],
+ "n_tests": 0,
+ "example_test": null
+ },
+ {
+ "symbol": "shortest_path",
+ "rel_src": "core/graph.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/graph.py",
+ "topic": "parent-pointer BFS memory fix (P1-1)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/graph.py.PropertyGraph.shortest_path",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_graph_path.py::test_both_reaches_from_leaf",
+ "tests/test_graph_path.py::test_incoming",
+ "tests/test_graph_path.py::test_max_depth",
+ "tests/test_graph_path.py::test_outgoing_default_backward_compat",
+ "tests/test_graph_path.py::test_same_node"
+ ],
+ "n_tests": 5,
+ "example_test": "tests/test_graph_path.py::test_both_reaches_from_leaf"
+ },
+ {
+ "symbol": "reciprocal_rank_fusion_3way",
+ "rel_src": "core/search/scoring.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/search/scoring.py",
+ "topic": "RRF tie-break determinism (P2-23)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/search/scoring.py.reciprocal_rank_fusion_3way",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_deep_search.py::TestAgenticDeepSearch::test_empty_index_returns_empty",
+ "tests/test_deep_search.py::TestDeepSearchMCP::test_empty_index_message",
+ "tests/test_fts5_integration.py::test_fts5_correctness_no_cross_contamination",
+ "tests/test_fts5_integration.py::test_fts5_results_present_in_search_with_mode",
+ "tests/test_fts5_integration.py::test_fts5_timeout_does_not_break_search",
+ "tests/test_fts5_integration.py::test_rrf_3way_merges_three_sources",
+ "tests/test_hybrid_cache.py::test_cache_hit_runs_dense_search_and_keeps_results",
+ "tests/test_hybrid_cache.py::test_cache_miss_also_runs_dense_search",
+ "tests/test_hybrid_cache.py::test_embed_failure_does_not_break_search",
+ "tests/test_hybrid_cache.py::test_two_identical_queries_both_keep_dense_results",
+ "tests/test_searcher_hardening.py::test_hybrid_search_async_raw_limit_safe_for_small_limit"
+ ],
+ "n_tests": 11,
+ "example_test": "tests/test_deep_search.py::TestAgenticDeepSearch::test_empty_index_returns_empty"
+ },
+ {
+ "symbol": "switch_db",
+ "rel_src": "core/indexing/db_manager.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/indexing/db_manager.py",
+ "topic": "write-lock around switch_db (P1-4)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/indexing/db_manager.py.LanceDBManager.switch_db",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_indexer_project_path.py::test_indexer_switch_project_updates_path",
+ "tests/test_lancedb_recreate.py::test_switch_db_fires_recreate_callback"
+ ],
+ "n_tests": 2,
+ "example_test": "tests/test_indexer_project_path.py::test_indexer_switch_project_updates_path"
+ },
+ {
+ "symbol": "build_tests_edges",
+ "rel_src": "core/bootstrap_tests.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/bootstrap_tests.py",
+ "topic": "TESTS-edges builder (A2)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/bootstrap_tests.py.build_tests_edges",
+ "func_in_graph": true,
+ "tests_covering": [],
+ "n_tests": 0,
+ "example_test": null
+ },
+ {
+ "symbol": "_graph_stage",
+ "rel_src": "core/search/engine.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/search/engine.py",
+ "topic": "Symbol graph stage (E4)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/search/engine.py.Searcher._graph_stage",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_deep_search.py::TestDeepSearchMCP::test_empty_index_message",
+ "tests/test_fts5_integration.py::test_fts5_correctness_no_cross_contamination",
+ "tests/test_fts5_integration.py::test_fts5_results_present_in_search_with_mode",
+ "tests/test_fts5_integration.py::test_fts5_timeout_does_not_break_search",
+ "tests/test_graph_center.py::TestGraphCenter::test_fast_search_applies_graph_expansion",
+ "tests/test_graph_stage_e4.py::test_graph_stage_layer_filter",
+ "tests/test_graph_stage_e4.py::test_graph_stage_no_index",
+ "tests/test_graph_stage_e4.py::test_graph_stage_returns_symbol_pointers",
+ "tests/test_graph_stage_e4.py::test_graph_stage_unknown_empty",
+ "tests/test_graph_stage_e4.py::test_short_circuit_identifier_no_embedder",
+ "tests/test_searcher.py::test_search_with_mode_cache_hit_on_second_call",
+ "tests/test_searcher.py::test_search_with_mode_fast_calls_embed_and_vector",
+ "tests/test_searcher.py::test_search_with_mode_fast_empty_embedding",
+ "tests/test_searcher.py::test_search_with_mode_fast_sorts_distance_ascending",
+ "tests/test_searcher_hardening.py::test_hybrid_search_async_raw_limit_safe_for_small_limit",
+ "tests/test_searcher_hardening.py::test_search_with_mode_cache_includes_layer_and_intent"
+ ],
+ "n_tests": 16,
+ "example_test": "tests/test_deep_search.py::TestDeepSearchMCP::test_empty_index_message"
+ },
+ {
+ "symbol": "safe_mkdir",
+ "rel_src": "core/artifact_paths.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/artifact_paths.py",
+ "topic": "runtime path factory",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/artifact_paths.py.safe_mkdir",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_action_receipt.py::test_gc_idempotent_noop_when_nothing_expired",
+ "tests/test_action_receipt.py::test_gc_keep_last_independent_of_age",
+ "tests/test_action_receipt.py::test_gc_removes_old_inconclusive_keeps_recent",
+ "tests/test_action_receipt.py::test_store_get_last_wins",
+ "tests/test_action_receipt.py::test_store_get_unknown_returns_none",
+ "tests/test_action_receipt.py::test_store_record_get_query",
+ "tests/test_alert_store.py::test_concurrent_collect_single_delivery",
+ "tests/test_alert_store.py::test_corrupt_json_graceful",
+ "tests/test_alert_store.py::test_get_alert_store_singleton_per_project",
+ "tests/test_alert_store.py::test_per_project_isolation",
+ "tests/test_alert_store.py::test_singleton_cache_key_resolved",
+ "tests/test_alert_store.py::test_starved_alert_message",
+ "tests/test_artifact_gc.py::test_check_disk_space_shape",
+ "tests/test_artifact_gc.py::test_crash_log_inside_data_root",
+ "tests/test_artifact_gc.py::test_data_root_never_in_home_when_env_set",
+ "tests/test_artifact_gc.py::test_logs_dir_inside_data_root",
+ "tests/test_artifact_gc.py::test_onnx_models_base_inside_data_root",
+ "tests/test_artifact_gc.py::test_prune_active_hashes_from_registry",
+ "tests/test_artifact_gc.py::test_prune_idempotent",
+ "tests/test_artifact_gc.py::test_prune_ignores_non_hash_dirs",
+ "tests/test_artifact_gc.py::test_prune_keeps_active_project",
+ "tests/test_artifact_gc.py::test_prune_removes_empty_dir_immediately",
+ "tests/test_artifact_gc.py::test_prune_removes_inactive_old_project",
+ "tests/test_artifact_gc.py::test_prune_removes_old_telemetry_even_active",
+ "tests/test_artifact_paths.py::test_all_paths_inside_project_dir",
+ "tests/test_artifact_paths.py::test_data_root_default_not_in_project",
+ "tests/test_artifact_paths.py::test_data_root_legacy_absolute_base_index_dir",
+ "tests/test_artifact_paths.py::test_data_root_uses_env_override",
+ "tests/test_artifact_paths.py::test_db_path_after_migration_matches_legacy_name",
+ "tests/test_artifact_paths.py::test_db_path_outside_project",
+ "tests/test_artifact_paths.py::test_get_project_dir_triggers_migration_on_first_create",
+ "tests/test_artifact_paths.py::test_migrate_idempotent",
+ "tests/test_artifact_paths.py::test_migrate_moves_graph_db",
+ "tests/test_artifact_paths.py::test_migrate_moves_legacy_index",
+ "tests/test_artifact_paths.py::test_migrate_moves_telemetry",
+ "tests/test_artifact_paths.py::test_migrate_skips_absent_legacy",
+ "tests/test_artifact_paths.py::test_project_isolation",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_branch_name_normalization",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_cleanup_old_branches",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_get_branch_db_path",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_get_branch_db_path_creates_dir",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_get_branch_info_no_index",
+ "tests/test_branch_aware_index.py::TestBranchAwareIndex::test_list_branch_indices_empty",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_bug_history_for_file",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_bug_risk_levels",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_detects_bugfix_commits",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_hotspots",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_no_bugfix_commits",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_stats",
+ "tests/test_bug_correlation.py::TestBugCorrelation::test_top_buggy_files",
+ "tests/test_commit_memory.py::TestCommitMemory::test_cache_persistence",
+ "tests/test_commit_memory.py::TestCommitMemory::test_fetch_commits",
+ "tests/test_commit_memory.py::TestCommitMemory::test_get_cochange_frequency",
+ "tests/test_commit_memory.py::TestCommitMemory::test_get_commits_for_file",
+ "tests/test_commit_memory.py::TestCommitMemory::test_get_file_stability",
+ "tests/test_commit_memory.py::TestCommitMemory::test_get_stats",
+ "tests/test_commit_memory.py::TestCommitMemory::test_no_git_repo",
+ "tests/test_commit_memory.py::TestCommitMemory::test_search_commits",
+ "tests/test_commit_memory_significance.py::test_significance_integration_git",
+ "tests/test_commit_memory_significance.py::test_significance_no_false_positives_for_docs",
+ "tests/test_commit_memory_significance.py::test_significance_returns_score_field",
+ "tests/test_commit_memory_significance.py::test_significance_scores_synthetic",
+ "tests/test_connection.py::test_setup",
+ "tests/test_context_sections.py::test_dataflow_no_candidates_returns_none",
+ "tests/test_context_sections.py::test_dataflow_renders_condition_path",
+ "tests/test_context_sections.py::test_memory_vor_filter",
+ "tests/test_context_sections.py::test_memory_vor_filter_caps_at_8",
+ "tests/test_context_sections.py::test_receipts_empty_returns_none",
+ "tests/test_context_sections.py::test_receipts_filter_by_file",
+ "tests/test_context_sections.py::test_source_regex_fallback_still_works",
+ "tests/test_context_sections.py::test_source_uses_meta_no_regex",
+ "tests/test_context_sections.py::test_writes_degrades_without_graph",
+ "tests/test_context_tool.py::test_collect_sections",
+ "tests/test_execution_contract_v2.py::test_ledger_isolated_from_project",
+ "tests/test_execution_contract_v2.py::test_ledger_query_limit",
+ "tests/test_execution_contract_v2.py::test_ledger_record_and_query",
+ "tests/test_freshness_gate.py::test_dirty_bypasses_verdict_cache",
+ "tests/test_freshness_gate.py::test_dirty_tree_sees_uncommitted_change",
+ "tests/test_freshness_gate.py::test_gate_off_disables_write_block",
+ "tests/test_freshness_gate.py::test_gate_read_only_still_blocks_write",
+ "tests/test_freshness_gate.py::test_read_gate_blocked_marks_stale_unverified",
+ "tests/test_freshness_gate.py::test_read_gate_blocks_when_stale_and_incomplete",
+ "tests/test_freshness_gate.py::test_write_gate_allows_after_consistent",
+ "tests/test_freshness_gate.py::test_write_gate_refuses_when_stale",
+ "tests/test_fts5_integration.py::test_fts5_results_present_in_search_with_mode",
+ "tests/test_index_progress.py::TestIndexerProgressCallback::test_indexer_accepts_callback",
+ "tests/test_indexer_project_path.py::test_delete_file_removes_only_specified_file",
+ "tests/test_indexer_project_path.py::test_indexer_has_project_path_attribute",
+ "tests/test_indexer_project_path.py::test_indexer_project_path_fallback",
+ "tests/test_indexer_project_path.py::test_indexer_switch_project_updates_path",
+ "tests/test_indexer_project_path.py::test_lsp_execute_file_indexing_no_attribute_error",
+ "tests/test_indexer_project_path.py::test_prune_deleted_files_with_empty_set_does_nothing",
+ "tests/test_job_history.py::test_append_and_load",
+ "tests/test_job_history.py::test_corrupted_history_recovers",
+ "tests/test_job_history.py::test_history_file_location",
+ "tests/test_job_history.py::test_history_truncated_to_50",
+ "tests/test_job_history.py::test_rolling_average_fallback_no_history",
+ "tests/test_job_history.py::test_rolling_average_fallback_no_similar",
+ "tests/test_job_history.py::test_rolling_average_similar_size",
+ "tests/test_local_fs_source.py::test_indexer_default_source_is_local_fs",
+ "tests/test_local_fs_source.py::test_indexer_uses_injected_source_path_manager",
+ "tests/test_mcp_schema_flat.py::test_no_dollardefs_or_ref_in_tool_schemas",
+ "tests/test_mcp_schema_flat.py::test_tool_schemas_are_objects_with_properties",
+ "tests/test_memory_retraction.py::test_add_default_status_active",
+ "tests/test_memory_retraction.py::test_add_memory_node_captures_import_anchor",
+ "tests/test_memory_retraction.py::test_add_memory_node_preserves_explicit_anchors",
+ "tests/test_memory_retraction.py::test_add_memory_node_prose_no_anchors",
+ "tests/test_memory_retraction.py::test_add_memory_node_skips_absolute_path",
+ "tests/test_memory_retraction.py::test_add_migrates_legacy_dict_format",
+ "tests/test_memory_retraction.py::test_add_rejects_invalid_status",
+ "tests/test_memory_retraction.py::test_add_rejects_refuted_status",
+ "tests/test_memory_retraction.py::test_add_verified_status",
+ "tests/test_memory_retraction.py::test_concurrent_add_and_retract_no_lost_updates",
+ "tests/test_memory_retraction.py::test_memory_metrics_counts_statuses_and_false_retractions",
+ "tests/test_memory_retraction.py::test_memory_metrics_empty_and_legacy",
+ "tests/test_memory_retraction.py::test_restore_marks_false_retraction_metric",
+ "tests/test_memory_retraction.py::test_retract_active_node_lifecycle",
+ "tests/test_memory_retraction.py::test_retract_requires_nonempty_reason",
+ "tests/test_memory_retraction.py::test_retract_twice_rejected",
+ "tests/test_memory_retraction.py::test_retract_unknown_node",
+ "tests/test_memory_retraction.py::test_retract_verified_node_owp_lifecycle",
+ "tests/test_memory_retraction.py::test_store_hides_refuted_keeps_active_verified_and_legacy",
+ "tests/test_memory_retraction.py::test_store_hides_superseded_keeps_active_verified",
+ "tests/test_memory_retraction.py::test_store_include_retracted_returns_refuted",
+ "tests/test_memory_retraction.py::test_store_include_retracted_returns_superseded",
+ "tests/test_memory_retraction.py::test_store_legacy_dict_format_intact",
+ "tests/test_memory_retraction.py::test_store_legacy_dict_hides_superseded",
+ "tests/test_memory_retraction.py::test_supersede_active_to_terminal",
+ "tests/test_memory_retraction.py::test_supersede_refuted_rejected",
+ "tests/test_memory_retraction.py::test_supersede_requires_reason",
+ "tests/test_memory_retraction.py::test_supersede_twice_rejected",
+ "tests/test_propagation_engine.py::test_already_refuted_dependent_skipped",
+ "tests/test_propagation_engine.py::test_cycle_safe",
+ "tests/test_propagation_engine.py::test_direct_dependent_retracted",
+ "tests/test_propagation_engine.py::test_find_dependents_returns_direct_only",
+ "tests/test_propagation_engine.py::test_independent_nodes_not_touched",
+ "tests/test_propagation_engine.py::test_layer_retract_cascades_to_dependents",
+ "tests/test_propagation_engine.py::test_layer_retract_no_cascade_without_dependencies",
+ "tests/test_propagation_engine.py::test_superseded_by_relation_retracted",
+ "tests/test_propagation_engine.py::test_transitive_chain",
+ "tests/test_real_path.py::test_unique_db_path_is_deterministic",
+ "tests/test_real_path.py::test_unique_db_path_isolates_projects",
+ "tests/test_real_path.py::test_unique_db_path_normalizes_separators_and_case",
+ "tests/test_real_path.py::test_unique_db_path_outside_project",
+ "tests/test_real_path.py::test_unique_db_path_supports_relative_project",
+ "tests/test_real_path.py::test_unique_db_path_uses_basename_not_full_path",
+ "tests/test_reindex_responsive.py::test_intel_runtime_status_normal_state_unchanged",
+ "tests/test_reindex_responsive.py::test_intel_runtime_status_reports_reindex_progress",
+ "tests/test_reindex_responsive.py::test_reindex_autodoc_does_not_block_event_loop",
+ "tests/test_reindex_responsive.py::test_reindex_job_sets_state_ready",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_bug_correlations",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_build_knowledge_graph",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_cochange_relations",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_empty_repo",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_get_related_files",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_get_related_files_depth_2",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_no_symbol_index",
+ "tests/test_relation_extractor.py::TestRelationExtractor::test_relation_summary",
+ "tests/test_remote_main.py::test_circuit_breaker_half_open_recovery",
+ "tests/test_remote_main.py::test_circuit_breaker_passthrough_when_healthy",
+ "tests/test_remote_main.py::test_circuit_breaker_returns_503_and_opens",
+ "tests/test_remote_main.py::test_healthz_ignores_auth_even_no_token",
+ "tests/test_remote_main.py::test_healthz_open_without_auth",
+ "tests/test_remote_main.py::test_mcp_requires_bearer",
+ "tests/test_remote_main.py::test_no_token_means_no_auth",
+ "tests/test_remote_main.py::test_rate_limit_disabled_when_rps_non_positive",
+ "tests/test_remote_main.py::test_rate_limit_healthz_exempt",
+ "tests/test_remote_main.py::test_rate_limit_ip_backstop",
+ "tests/test_remote_main.py::test_rate_limit_per_token_first",
+ "tests/test_remote_main.py::test_rate_limit_tracks_token_hash_not_plaintext",
+ "tests/test_remote_main.py::test_wrong_token_rejected",
+ "tests/test_resource_monitor.py::TestResourceMonitor::test_summary_structure",
+ "tests/test_run_1L_live_arm.py::test_progress_path_tag_distinct",
+ "tests/test_search_bs_audit.py::test_bs11_default_answer_is_fast",
+ "tests/test_search_bs_audit.py::test_bs8_resolve_active_embedder_fallback",
+ "tests/test_search_bs_audit.py::test_bs8_telemetry_uses_active_embedder",
+ "tests/test_search_bs_audit.py::test_bs9_di_registry_is_global_singleton",
+ "tests/test_searcher_hardening.py::test_hybrid_search_async_raw_limit_safe_for_small_limit",
+ "tests/test_verify_on_read.py::test_budget_exceeded_nodes_recorded_in_stats",
+ "tests/test_verify_on_read.py::test_budget_timeout_graceful_degradation",
+ "tests/test_verify_on_read.py::test_cache_hit_second_run_same_head",
+ "tests/test_verify_on_read.py::test_cache_schema_guard_rebuilds_without_packages",
+ "tests/test_verify_on_read.py::test_counters_matched_delivered_fresh_and_cache_hit",
+ "tests/test_verify_on_read.py::test_counters_persist_to_cache_file",
+ "tests/test_verify_on_read.py::test_counters_survive_head_change",
+ "tests/test_verify_on_read.py::test_dist_name_pyyaml_prose_verified_via_manifest",
+ "tests/test_verify_on_read.py::test_env_anchor_found_and_absent",
+ "tests/test_verify_on_read.py::test_file_anchor_found",
+ "tests/test_verify_on_read.py::test_found_anchor_verified_included",
+ "tests/test_verify_on_read.py::test_head_change_rechecks_per_node",
+ "tests/test_verify_on_read.py::test_layer_budget_exceeded_flags_nodes",
+ "tests/test_verify_on_read.py::test_layer_hook_verify_on_read_default_on",
+ "tests/test_verify_on_read.py::test_layer_hook_verify_on_read_off",
+ "tests/test_verify_on_read.py::test_layer_starved_flags_nodes",
+ "tests/test_verify_on_read.py::test_legacy_node_no_status_checkable_verified",
+ "tests/test_verify_on_read.py::test_no_anchors_inconclusive_stays_active",
+ "tests/test_verify_on_read.py::test_not_found_anchor_refuted_excluded",
+ "tests/test_verify_on_read.py::test_pkg_anchor_absent_refuted_excluded",
+ "tests/test_verify_on_read.py::test_pkg_anchor_found_verified_included",
+ "tests/test_verify_on_read.py::test_pkg_anchor_npm_dotted_name_verified",
+ "tests/test_verify_on_read.py::test_prose_import_path_node_not_refuted",
+ "tests/test_verify_on_read.py::test_read_path_prose_history_legacy_node_keeps_refuting",
+ "tests/test_verify_on_read.py::test_read_path_prose_history_not_refuting_live_node",
+ "tests/test_verify_on_read.py::test_run_background_verify_empty_memory_returns_none",
+ "tests/test_verify_on_read.py::test_run_background_verify_refutes_on_idle_pass",
+ "tests/test_verify_on_read.py::test_run_background_verify_skips_when_lock_held",
+ "tests/test_verify_on_read.py::test_starved_node_seen_never_delivered",
+ "tests/test_verify_on_read.py::test_symbol_anchor_inconclusive_when_no_resolver",
+ "tests/test_verify_on_read.py::test_symbol_anchor_stays_inconclusive_when_resolver_cant_prove_freshness",
+ "tests/test_verify_on_read.py::test_symbol_anchor_verified_when_referent_present",
+ "tests/test_verify_on_read.py::test_terminal_refuted_not_rewritten_to_verified",
+ "tests/test_verify_on_read.py::test_terminal_superseded_not_rewritten_to_verified",
+ "tests/test_verify_on_read.py::test_unknown_anchor_kind_inconclusive_not_refuted",
+ "tests/test_verify_on_read.py::test_verified_node_rechecked_on_head_change",
+ "tests/test_verify_on_read.py::test_write_capture_makes_verify_effective_on_prose",
+ "tests/test_verify_on_read_ttl.py::test_inconclusive_node_gets_last_checked",
+ "tests/test_verify_on_read_ttl.py::test_last_checked_old_trace_refreshed",
+ "tests/test_verify_on_read_ttl.py::test_last_checked_rate_limited_fresh_trace_kept",
+ "tests/test_verify_on_read_ttl.py::test_layer_marks_stale_ttl_flag",
+ "tests/test_verify_on_read_ttl.py::test_stale_ttl_cache_hit_not_flagged_second_pass",
+ "tests/test_verify_on_read_ttl.py::test_stale_ttl_checked_node_not_flagged",
+ "tests/test_verify_on_read_ttl.py::test_stale_ttl_fresh_trace_not_flagged",
+ "tests/test_verify_on_read_ttl.py::test_stale_ttl_no_trace_not_flagged",
+ "tests/test_verify_on_read_ttl.py::test_stale_ttl_unvisited_ancient_trace_flagged",
+ "tests/test_write_tools.py::TestWriteToolInsertAfter::test_insert_after_class",
+ "tests/test_write_tools.py::TestWriteToolInsertAfter::test_inserts_after_anchor",
+ "tests/test_write_tools.py::TestWriteToolInsertBefore::test_insert_before_class",
+ "tests/test_write_tools.py::TestWriteToolInsertBefore::test_inserts_before_anchor",
+ "tests/test_write_tools.py::TestWriteToolMove::test_apply_with_real_files",
+ "tests/test_write_tools.py::TestWriteToolPreflight::test_insert_success_with_note_when_types_checked",
+ "tests/test_write_tools.py::TestWriteToolRename::test_apply_modifies_file",
+ "tests/test_write_tools.py::TestWriteToolReplace::test_apply_records_action_receipt",
+ "tests/test_write_tools.py::TestWriteToolReplace::test_apply_replaces_body",
+ "tests/test_write_tools.py::TestWriteToolSafeDelete::test_apply_deletes_from_file"
+ ],
+ "n_tests": 234,
+ "example_test": "tests/test_action_receipt.py::test_gc_idempotent_noop_when_nothing_expired"
+ },
+ {
+ "symbol": "get_neighbors",
+ "rel_src": "core/graph.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/graph.py",
+ "topic": "BFS neighbor traversal (C-4)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/graph.py.PropertyGraph.get_neighbors",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_graph_adapter_node_selection.py::test_callers_merged_across_placeholder_and_real",
+ "tests/test_graph_adapter_node_selection.py::test_d1_src_definition_beats_experiments_shadow",
+ "tests/test_graph_adapter_node_selection.py::test_d2_method_resolved_by_bare_name",
+ "tests/test_graph_adapter_node_selection.py::test_d3_placeholder_does_not_shadow_real_definition",
+ "tests/test_graph_adapter_node_selection.py::test_qualified_symbol_includes_bare_node_callers",
+ "tests/test_graph_center.py::TestGraphCenter::test_expand_graph_context_adds_callers_metadata",
+ "tests/test_graph_center.py::TestGraphCenter::test_fast_search_applies_graph_expansion",
+ "tests/test_graph_center.py::TestGraphCenter::test_find_definitions_and_call_chain_suffix",
+ "tests/test_graph_center.py::TestGraphCenter::test_find_references_by_bare_name_finds_qualified_method",
+ "tests/test_graph_center.py::TestGraphCenter::test_method_ref_resolves_to_real_node_not_extern",
+ "tests/test_graph_resolver.py::TestGraphSymbolResolver::test_import_fqn_match",
+ "tests/test_graph_resolver.py::TestGraphSymbolResolver::test_stdlib_package_match",
+ "tests/test_graph_resolver.py::TestGraphSymbolResolver::test_unique_global_match",
+ "tests/test_graph_resolver.py::TestGraphSymbolResolver::test_unresolved_fallback",
+ "tests/test_impact_confidence.py::TestImpactGrouping::test_confidence_in_entries",
+ "tests/test_impact_confidence.py::TestImpactGrouping::test_depth_grouping",
+ "tests/test_impact_confidence.py::TestImpactGrouping::test_top_callers_sorted_by_depth_then_confidence",
+ "tests/test_sym_index_partial.py::TestSymIndexPartial::test_add_and_find",
+ "tests/test_sym_index_partial.py::TestSymIndexPartial::test_references_only_symbol",
+ "tests/test_sym_index_partial.py::TestSymIndexPartial::test_reindex_with_cross_file_refs"
+ ],
+ "n_tests": 20,
+ "example_test": "tests/test_graph_adapter_node_selection.py::test_callers_merged_across_placeholder_and_real"
+ },
+ {
+ "symbol": "calculate_token_savings",
+ "rel_src": "core/search/token_savings.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/search/token_savings.py",
+ "topic": "token savings (CRG)",
+ "func_qname": "D:.D:/Project/MSCodeBase/src/core/search/token_savings.py.calculate_token_savings",
+ "func_in_graph": true,
+ "tests_covering": [
+ "tests/test_deep_search.py::TestAgenticDeepSearch::test_empty_index_returns_empty",
+ "tests/test_deep_search.py::TestDeepSearchMCP::test_empty_index_message",
+ "tests/test_fts5_integration.py::test_fts5_correctness_no_cross_contamination",
+ "tests/test_fts5_integration.py::test_fts5_results_present_in_search_with_mode",
+ "tests/test_fts5_integration.py::test_fts5_timeout_does_not_break_search",
+ "tests/test_hybrid_cache.py::test_cache_hit_runs_dense_search_and_keeps_results",
+ "tests/test_hybrid_cache.py::test_cache_miss_also_runs_dense_search",
+ "tests/test_hybrid_cache.py::test_embed_failure_does_not_break_search",
+ "tests/test_hybrid_cache.py::test_two_identical_queries_both_keep_dense_results",
+ "tests/test_searcher_hardening.py::test_hybrid_search_async_raw_limit_safe_for_small_limit"
+ ],
+ "n_tests": 10,
+ "example_test": "tests/test_deep_search.py::TestAgenticDeepSearch::test_empty_index_returns_empty"
+ },
+ {
+ "symbol": "auto_detect_intent",
+ "rel_src": "core/search/engine.py",
+ "abs_src": "D:/Project/MSCodeBase/src/core/search/engine.py",
+ "topic": "intent auto-detect (B1)",
+ "func_qname": null,
+ "func_in_graph": false,
+ "tests_covering": [],
+ "n_tests": 0,
+ "example_test": null
+ }
+]
\ No newline at end of file
diff --git a/experiments/bootstrap/e17_redteam.py b/experiments/bootstrap/e17_redteam.py
new file mode 100644
index 00000000..4c139207
--- /dev/null
+++ b/experiments/bootstrap/e17_redteam.py
@@ -0,0 +1,186 @@
+# -*- coding: utf-8 -*-
+"""E17 Red Team: 5 атак на TESTS-сигнал.
+
+Атаки:
+1. Конкурентность: 10 потоков одновременно вызывают _graph_stage — нет ли гонки?
+2. Границы: функция с 1000 тестами — сколько времени занимает get_tests_for_symbol?
+3. Злоупотребление: запрос несуществующей функции — graceful degradation?
+4. TOCTOU: между find_nodes и get_neighbors граф меняется — как обрабатывается?
+5. Отказ зависимостей: PropertyGraph закрыт/повреждён — graceful degradation?
+
+Воспроизводимо:
+ cd D:\\Project\\MSCodeBase
+ python -X utf8 experiments/bootstrap/e17_redteam.py
+"""
+import sys
+import threading
+import time
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+from src.core.artifact_paths import get_graph_db_path
+from src.core.graph import PropertyGraph
+from src.core.search.engine import Searcher
+from src.core.search.graph_adapter import SymbolIndexAdapter
+
+sys.stdout.reconfigure(encoding="utf-8")
+ROOT = Path(__file__).resolve().parents[2]
+
+
+class FakeIndexer:
+ def __init__(self, symbol_index):
+ self._symbol_index = symbol_index
+ self.symbol_index = symbol_index
+
+ async def close_async(self):
+ pass
+
+
+def attack_1_concurrency():
+ """Атака 1: Конкурентность — 10 потоков одновременно вызывают _graph_stage."""
+ print("\n[АТАКА 1] Конкурентность: 10 потоков × 100 вызовов", flush=True)
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+ searcher._tests_signal = True
+
+ errors = []
+ lock = threading.Lock()
+
+ def worker():
+ try:
+ for _ in range(100):
+ searcher._graph_stage('safe_mkdir', limit=8)
+ except Exception as e: # noqa: BLE001 — red team: собираем все ошибки конкурентности
+ with lock:
+ errors.append(str(e))
+
+ threads = [threading.Thread(target=worker) for _ in range(10)]
+ t0 = time.perf_counter()
+ for t in threads:
+ t.start()
+ for t in threads:
+ t.join()
+ dt = (time.perf_counter() - t0) * 1000
+
+ print(f" 1000 вызовов за {dt:.1f}ms ({dt/1000:.2f}ms/вызов)", flush=True)
+ print(f" Ошибок: {len(errors)}", flush=True)
+ if errors:
+ print(f" Первая ошибка: {errors[0]}", flush=True)
+ pg.close()
+ return len(errors) == 0
+
+
+def attack_2_boundaries():
+ """Атака 2: Границы — функция с наибольшим количеством тестов."""
+ print("\n[АТАКА 2] Границы: функция с наибольшим количеством тестов", flush=True)
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+ searcher._tests_signal = True
+
+ # safe_mkdir имеет 234 теста (из e17_wide_panel)
+ t0 = time.perf_counter()
+ out = searcher._graph_stage('safe_mkdir', limit=8)
+ dt = (time.perf_counter() - t0) * 1000
+
+ # сколько тестов добавлено
+ tests_added = sum(1 for r in out if r.get('metadata', {}).get('tests_signal'))
+ print(f" Время: {dt:.2f}ms", flush=True)
+ print(f" Тестов добавлено: {tests_added}", flush=True)
+ print(f" Всего результатов: {len(out)}", flush=True)
+ pg.close()
+ return dt < 100 # приемлемо если <100ms
+
+
+def attack_3_abuse():
+ """Атака 3: Злоупотребление — запрос несуществующей функции."""
+ print("\n[АТАКА 3] Злоупотребление: запрос несуществующей функции", flush=True)
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+ searcher._tests_signal = True
+
+ try:
+ out = searcher._graph_stage('nonexistent_function_xyz_12345', limit=8)
+ print(f" Результатов: {len(out)}", flush=True)
+ print(f" Graceful degradation: OK (вернул {len(out)} результатов)", flush=True)
+ pg.close()
+ return True
+ except Exception as e: # noqa: BLE001 — red team: проверяем graceful degradation
+ print(f" Ошибка: {e}", flush=True)
+ pg.close()
+ return False
+
+
+def attack_4_toctou():
+ """Атака 4: TOCTOU — граф меняется между вызовами (имитация reindex)."""
+ print("\n[АТАКА 4] TOCTOU: граф закрывается между вызовами", flush=True)
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+ searcher._tests_signal = True
+
+ # первый вызов
+ searcher._graph_stage('safe_mkdir', limit=8)
+
+ # имитация reindex: закрываем граф
+ pg.close()
+
+ # второй вызов (граф закрыт)
+ try:
+ out2 = searcher._graph_stage('safe_mkdir', limit=8)
+ print(f" После закрытия графа: {len(out2)} результатов", flush=True)
+ print(" Graceful degradation: OK", flush=True)
+ return True
+ except Exception as e: # noqa: BLE001 — red team: проверяем TOCTOU
+ print(f" Ошибка после закрытия: {e}", flush=True)
+ print(" Graceful degradation: FAIL", flush=True)
+ return False
+
+
+def attack_5_dependency_failure():
+ """Атака 5: Отказ зависимостей — PropertyGraph повреждён."""
+ print("\n[АТАКА 5] Отказ зависимостей: PropertyGraph с несуществующим путём", flush=True)
+ try:
+ pg = PropertyGraph(Path('Z:/nonexistent/graph.db'))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+ searcher._tests_signal = True
+
+ out = searcher._graph_stage('safe_mkdir', limit=8)
+ print(f" Результатов: {len(out)}", flush=True)
+ print(" Graceful degradation: OK", flush=True)
+ return True
+ except Exception as e: # noqa: BLE001 — red team: проверяем отказ зависимостей
+ print(f" Ошибка: {e}", flush=True)
+ print(" Graceful degradation: FAIL", flush=True)
+ return False
+
+
+def main() -> int:
+ print("=" * 70, flush=True)
+ print("E17 RED TEAM: 5 атак на TESTS-сигнал", flush=True)
+ print("=" * 70, flush=True)
+
+ results = []
+ results.append(("Конкурентность", attack_1_concurrency()))
+ results.append(("Границы", attack_2_boundaries()))
+ results.append(("Злоупотребление", attack_3_abuse()))
+ results.append(("TOCTOU", attack_4_toctou()))
+ results.append(("Отказ зависимостей", attack_5_dependency_failure()))
+
+ print("\n" + "=" * 70, flush=True)
+ print("РЕЗУЛЬТАТЫ RED TEAM:", flush=True)
+ print("=" * 70, flush=True)
+ for name, passed in results:
+ status = "✅ PASS" if passed else "❌ FAIL"
+ print(f" {status} {name}", flush=True)
+
+ all_passed = all(passed for _, passed in results)
+ print(f"\nИтого: {sum(1 for _, p in results if p)}/{len(results)} атак отражено", flush=True)
+ return 0 if all_passed else 1
+
+
+if __name__ == "__main__":
+ sys.exit(main())
diff --git a/experiments/bootstrap/e17_wide_panel.py b/experiments/bootstrap/e17_wide_panel.py
new file mode 100644
index 00000000..5d88d4e3
--- /dev/null
+++ b/experiments/bootstrap/e17_wide_panel.py
@@ -0,0 +1,276 @@
+# -*- coding: utf-8 -*-
+"""E17 Wide Panel: языковое покрытие + 30+ запросов на реальном graph_stage.
+
+Пользователь указал на нарушение порядка: статья вышла до широкой валидации.
+Этот скрипт исправляет: сначала измерения, потом статья с честными цифрами.
+
+Что измеряет:
+1. Языковое покрытие: сколько функций по языкам (Python/Go/TS/etc.) имеют TESTS-рёбра.
+2. Широкая панель 30+ identifier-запросов: A/B on/off tests_signal, hit@1/hit@3/MRR,
+ новые тесты, overhead.
+3. Red team: 5 атак на TESTS-сигнал (конкурентность, границы, злоупотребление, TOCTOU, отказ).
+
+Воспроизводимо:
+ cd D:\\Project\\MSCodeBase
+ python -X utf8 experiments/bootstrap/e17_wide_panel.py
+"""
+import sqlite3
+import sys
+import time
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+from src.core.artifact_paths import get_graph_db_path
+from src.core.graph import PropertyGraph
+from src.core.search.engine import Searcher
+from src.core.search.graph_adapter import SymbolIndexAdapter
+
+sys.stdout.reconfigure(encoding="utf-8")
+ROOT = Path(__file__).resolve().parents[2]
+
+
+class FakeIndexer:
+ """Минимальный indexer-контракт Searcher'а: только symbol_index."""
+
+ def __init__(self, symbol_index):
+ self._symbol_index = symbol_index
+ self.symbol_index = symbol_index
+
+ async def close_async(self):
+ pass
+
+
+def measure_language_coverage() -> dict:
+ """Языковое покрытие: сколько функций по языкам имеют TESTS-рёбра."""
+ db_path = get_graph_db_path(ROOT)
+ conn = sqlite3.connect(str(db_path))
+ conn.row_factory = sqlite3.Row
+
+ # Языковое покрытие
+ q = """
+ WITH funcs AS (
+ SELECT id, file_path,
+ CASE
+ WHEN file_path LIKE '%.py' THEN 'Python'
+ WHEN file_path LIKE '%.go' THEN 'Go'
+ WHEN file_path LIKE '%.ts' OR file_path LIKE '%.tsx' THEN 'TypeScript'
+ WHEN file_path LIKE '%.js' OR file_path LIKE '%.jsx' THEN 'JavaScript'
+ WHEN file_path LIKE '%.rs' THEN 'Rust'
+ WHEN file_path LIKE '%.lua' THEN 'Lua'
+ ELSE 'Other'
+ END AS lang
+ FROM nodes WHERE label='Function'
+ ),
+ tests_coverage AS (
+ SELECT f.id, f.lang,
+ CASE WHEN EXISTS (SELECT 1 FROM edges e WHERE e.type='TESTS' AND e.target_id=f.id) THEN 1 ELSE 0 END AS has_tests
+ FROM funcs f
+ )
+ SELECT lang, COUNT(*) AS total, SUM(has_tests) AS covered,
+ ROUND(100.0 * SUM(has_tests) / COUNT(*), 1) AS pct
+ FROM tests_coverage
+ GROUP BY lang
+ ORDER BY total DESC
+ """
+ coverage = {}
+ for r in conn.execute(q):
+ coverage[r["lang"]] = {"total": r["total"], "covered": r["covered"], "pct": r["pct"]}
+
+ # Общие цифры
+ total_funcs = conn.execute("SELECT COUNT(*) FROM nodes WHERE label='Function'").fetchone()[0]
+ total_tests_edges = conn.execute("SELECT COUNT(*) FROM edges WHERE type='TESTS'").fetchone()[0]
+ total_test_nodes = conn.execute("SELECT COUNT(*) FROM nodes WHERE label='Test'").fetchone()[0]
+
+ conn.close()
+ return {
+ "coverage_by_lang": coverage,
+ "total_funcs": total_funcs,
+ "total_tests_edges": total_tests_edges,
+ "total_test_nodes": total_test_nodes,
+ }
+
+
+def generate_wide_queries() -> list:
+ """Генерирует 30+ identifier-запросов из PropertyGraph (функции с наибольшим количеством TESTS-рёбер)."""
+ db_path = get_graph_db_path(ROOT)
+ conn = sqlite3.connect(str(db_path))
+ conn.row_factory = sqlite3.Row
+
+ # Берём функции с наибольшим количеством TESTS-рёбер (без требования уникальности имён)
+ # Это гарантирует что тесты есть, а hit@1 проверяется по file_path+name (уникально)
+ # name — это имя функции (например, 'safe_mkdir'), label — тип узла ('Function')
+ q = """
+ WITH funcs AS (
+ SELECT id, name, file_path
+ FROM nodes WHERE label='Function' AND file_path LIKE '%.py'
+ ),
+ tests_count AS (
+ SELECT target_id, COUNT(*) AS n_tests
+ FROM edges WHERE type='TESTS'
+ GROUP BY target_id
+ )
+ SELECT f.id, f.name, f.file_path, tc.n_tests
+ FROM funcs f
+ JOIN tests_count tc ON tc.target_id = f.id
+ WHERE tc.n_tests > 0
+ ORDER BY tc.n_tests DESC
+ LIMIT 35
+ """
+ queries = []
+ for r in conn.execute(q):
+ queries.append({
+ "q": r["name"],
+ "symbol": r["name"],
+ "file": r["file_path"],
+ "n_tests": r["n_tests"],
+ })
+ conn.close()
+ return queries
+
+
+def measure_panel(searcher: Searcher, queries: list, flag_on: bool) -> list:
+ """Прогоняет панель запросов с on/off tests_signal."""
+ searcher._tests_signal = flag_on
+ rows = []
+ for item in queries:
+ t0 = time.perf_counter()
+ out = searcher._graph_stage(item["q"], limit=8)
+ dt = (time.perf_counter() - t0) * 1000
+ ranked = sorted(out, key=lambda r: -(r.get("final_score") or 0))
+ top6 = sorted(
+ "{}:{}{}".format(
+ r.get("metadata", {}).get("file", "?").replace(ROOT.as_posix() + "/", ""),
+ r.get("metadata", {}).get("symbol", "?"),
+ " [T]" if r.get("metadata", {}).get("tests_signal") else "",
+ )
+ for r in ranked[:6]
+ )
+ rows.append({"q": item["q"], "dt_ms": dt, "top6": top6, "n_tests": item["n_tests"]})
+ return rows
+
+
+def compute_metrics(res_off: list, res_on: list, queries: list) -> dict:
+ """Вычисляет hit@1/hit@3/MRR + новые тесты."""
+ if not queries:
+ return {
+ "hits_off": {"h1": 0, "h3": 0, "n": 0},
+ "hits_on": {"h1": 0, "h3": 0, "n": 0},
+ "mrr_off": 0.0,
+ "mrr_on": 0.0,
+ "signaled": [],
+ }
+ hits_off = {"h1": 0, "h3": 0, "n": len(queries)}
+ hits_on = {"h1": 0, "h3": 0, "n": len(queries)}
+ mrr_num_off = 0.0
+ mrr_num_on = 0.0
+ signaled = []
+
+ for off, on, item in zip(res_off, res_on, queries):
+ # Нормализую file_path так же как в top6 (относительный путь)
+ target_file_rel = item["file"].replace(ROOT.as_posix() + "/", "")
+ target_symbol = item["symbol"]
+
+ off_rank = next(
+ (i + 1 for i, r in enumerate(off["top6"]) if target_symbol in r and target_file_rel in r),
+ None,
+ )
+ on_rank = next(
+ (i + 1 for i, r in enumerate(on["top6"]) if target_symbol in r and target_file_rel in r),
+ None,
+ )
+ if off_rank:
+ if off_rank <= 1:
+ hits_off["h1"] += 1
+ if off_rank <= 3:
+ hits_off["h3"] += 1
+ mrr_num_off += 1.0 / off_rank
+ if on_rank:
+ if on_rank <= 1:
+ hits_on["h1"] += 1
+ if on_rank <= 3:
+ hits_on["h3"] += 1
+ mrr_num_on += 1.0 / on_rank
+
+ off_keys = {r for r in off["top6"] if "tests/" in r or "[T]" in r}
+ on_keys = {r for r in on["top6"] if "tests/" in r or "[T]" in r}
+ new_tests = sorted(on_keys - off_keys)
+ signaled.append((item["q"], new_tests))
+
+ return {
+ "hits_off": hits_off,
+ "hits_on": hits_on,
+ "mrr_off": mrr_num_off / len(queries),
+ "mrr_on": mrr_num_on / len(queries),
+ "signaled": signaled,
+ }
+
+
+def main() -> int:
+ print("=" * 70, flush=True)
+ print("E17 WIDE PANEL: языковое покрытие + 30+ запросов", flush=True)
+ print("=" * 70, flush=True)
+
+ # 1. Языковое покрытие
+ print("\n[1] ЯЗЫКОВОЕ ПОКРЫТИЕ (сколько функций имеют TESTS-рёбра):", flush=True)
+ cov = measure_language_coverage()
+ print(f" Всего функций в графе: {cov['total_funcs']}", flush=True)
+ print(f" Всего TESTS-рёбер: {cov['total_tests_edges']}", flush=True)
+ print(f" Всего Test-узлов: {cov['total_test_nodes']}", flush=True)
+ print("\n Покрытие по языкам:", flush=True)
+ for lang, data in cov["coverage_by_lang"].items():
+ print(f" {lang:12} total={data['total']:5} covered={data['covered']:5} ({data['pct']:5.1f}%)", flush=True)
+
+ # 2. Широкая панель
+ print("\n[2] ШИРОКАЯ ПАНЕЛЬ (30+ identifier-запросов):", flush=True)
+ queries = generate_wide_queries()
+ print(f" Сгенерировано запросов: {len(queries)}", flush=True)
+
+ pg = PropertyGraph(get_graph_db_path(ROOT))
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+ searcher = Searcher(indexer=FakeIndexer(adapter), embedder=None)
+
+ res_off = measure_panel(searcher, queries, flag_on=False)
+ res_on = measure_panel(searcher, queries, flag_on=True)
+
+ metrics = compute_metrics(res_off, res_on, queries)
+
+ print(f"\n hit@1 off={metrics['hits_off']['h1']}/{metrics['hits_off']['n']} "
+ f"on={metrics['hits_on']['h1']}/{metrics['hits_on']['n']}", flush=True)
+ print(f" hit@3 off={metrics['hits_off']['h3']}/{metrics['hits_off']['n']} "
+ f"on={metrics['hits_on']['h3']}/{metrics['hits_on']['n']}", flush=True)
+ n_with_new = sum(1 for _, nt in metrics["signaled"] if nt)
+ print(f" TESTS-signal: {n_with_new}/{len(metrics['signaled'])} запросов получили новые покрывающие тесты", flush=True)
+ print(f" MRR(function) off={metrics['mrr_off']:.3f} on={metrics['mrr_on']:.3f}", flush=True)
+
+ avg_off = sum(r["dt_ms"] for r in res_off) / len(res_off)
+ avg_on = sum(r["dt_ms"] for r in res_on) / len(res_on)
+ print(f" graph_stage avg dt: off={avg_off:.2f}ms on={avg_on:.2f}ms "
+ f"(overhead {(avg_on / avg_off - 1) * 100:+.1f}%)", flush=True)
+
+ # Детали по запросам (первые 10)
+ print("\n Детали (первые 10 запросов):", flush=True)
+ for i, (off, on, item) in enumerate(zip(res_off[:10], res_on[:10], queries[:10])):
+ off_rank = next(
+ (j + 1 for j, r in enumerate(off["top6"]) if item["symbol"] in r and item["file"] in r),
+ None,
+ )
+ on_rank = next(
+ (j + 1 for j, r in enumerate(on["top6"]) if item["symbol"] in r and item["file"] in r),
+ None,
+ )
+ off_keys = {r for r in off["top6"] if "tests/" in r or "[T]" in r}
+ on_keys = {r for r in on["top6"] if "tests/" in r or "[T]" in r}
+ new_tests = sorted(on_keys - off_keys)
+ print(f" [{i+1}] {item['symbol']:30} n_tests={item['n_tests']:3} | "
+ f"rank off={off_rank} on={on_rank} | NEW tests: {len(new_tests)}", flush=True)
+
+ pg.close()
+
+ print("\n" + "=" * 70, flush=True)
+ print("E17 WIDE PANEL: DONE", flush=True)
+ print("=" * 70, flush=True)
+ return 0
+
+
+if __name__ == "__main__":
+ sys.exit(main())
diff --git a/src/config/settings.py b/src/config/settings.py
index a6f18886..bcf202a8 100644
--- a/src/config/settings.py
+++ b/src/config/settings.py
@@ -163,6 +163,17 @@ class SearchConfig:
== "true"
)
+ # TESTS-сигнал (E17, эксперимент): в graph-stage добавлять к найденным
+ # определениям функций покрывающие их тесты (TESTS-рёбра PropertyGraph)
+ # как отдельный сорт результатов с пониженным graph_score.
+ # Off по умолчанию — поведение поиска неизменно без флага.
+ tests_signal: bool = field(
+ default_factory=lambda: os.getenv(
+ "MSCODEBASE_TESTS_SIGNAL", "false"
+ ).lower()
+ == "true"
+ )
+
@dataclass
class IndexConfig:
diff --git a/src/core/search/engine.py b/src/core/search/engine.py
index 3791fd46..37de7c61 100644
--- a/src/core/search/engine.py
+++ b/src/core/search/engine.py
@@ -240,6 +240,11 @@ def __init__(self, indexer, embedder):
self._late_enrichment = bool(
getattr(get_config().search, "late_enrichment", False)
)
+ # TESTS-сигнал (E17, эксперимент): MSCODEBASE_TESTS_SIGNAL=true.
+ # Добавляет покрывающие тесты в graph-stage. Off по умолчанию.
+ self._tests_signal = bool(
+ getattr(get_config().search, "tests_signal", False)
+ )
self._multi_reranker: Optional[MultiProviderReranker] = None
self._multi_reranker_initialized: bool = False
self._multi_reranker_lock: Optional[asyncio.Lock] = None # lazy: создаётся при первом async-вызове (привязка к event loop)
@@ -1278,11 +1283,69 @@ def _graph_stage(self, query: str, limit: int, layer: Optional[str] = None) -> L
"graph_score": 1.0 if ref.is_definition else 0.5,
"final_score": 1.0 if ref.is_definition else 0.5,
})
+ # E17 (TESTS-сигнал): если включён, добавить к найденным определениям
+ # функций покрывающие их тесты. Пониженный graph_score (0.4) — тесты
+ # не вытесняют функции; отдельный sentinel-диапазон (-20_000_000+line),
+ # чтобы не коллайдить с symbol-указателями (-10_000_000) и чанками.
+ if self._tests_signal and out:
+ out = self._append_tests_signal(out, si)
return out
except Exception as e:
logger.debug(f"graph_stage error: {e}")
return []
+ def _append_tests_signal(self, out: List[dict], si) -> List[dict]:
+ """E17: добавляет покрывающие тесты (TESTS-рёбра) к найденным определениям.
+
+ Def-рефы (is_definition=True) обогащаются тестами из PropertyGraph.
+ Безопасно при отсутствии метода/графа/рёбер: возвращает out без изменений.
+ """
+ try:
+ get_tests = getattr(si, "get_tests_for_symbol", None)
+ if get_tests is None:
+ return out
+
+ test_limit = max(1, min(len(out), 6)) # потолок тестов на запрос
+ result: List[dict] = []
+ added = 0
+ for r in out:
+ result.append(r)
+ meta = r.get("metadata", {})
+ if not meta.get("is_definition") or added >= test_limit:
+ continue
+ refs = get_tests(meta["symbol"], meta["file"], limit=3)
+ for t in refs:
+ if added >= test_limit:
+ break
+ t_layer = self._layer_from_path(t.file_path)
+ result.append({
+ "text": f"test {t.symbol}\n📍 {t.file_path}:{t.line}",
+ "metadata": {
+ "file": t.file_path,
+ "chunk_index": -(20_000_000 + t.line),
+ "layer": t_layer,
+ "symbol": t.symbol,
+ "symbol_name": t.symbol,
+ "line": t.line,
+ "kind": t.kind,
+ "graph_stage": True,
+ "is_symbol_ref": True,
+ "is_definition": False,
+ "tests_signal": True,
+ "covers": meta["symbol"],
+ },
+ "bm25_score": 0.0,
+ "dense_score": 0.0,
+ "fts5_score": 0.0,
+ "graph_score": 0.4,
+ "final_score": 0.4,
+ })
+ added += 1
+ return result
+ except Exception as e:
+ logger.debug(f"tests_signal error: {e}")
+ return out
+
def _expand_graph_context(
self, results: List[dict], original_query: str
) -> List[dict]:
diff --git a/src/core/search/graph_adapter.py b/src/core/search/graph_adapter.py
index bc3f22c7..040e0e5d 100644
--- a/src/core/search/graph_adapter.py
+++ b/src/core/search/graph_adapter.py
@@ -529,6 +529,61 @@ def search_symbols(self, query: str, top_k: int = 10) -> List[SymbolRef]:
results.extend(refs)
return results
+ def get_tests_for_symbol(
+ self, symbol: str, file_path: str, limit: int = 6
+ ) -> List[SymbolRef]:
+ """Возвращает покрывающие функцию тесты (incoming TESTS-рёбра).
+
+ E17 (TESTS-сигнал): ищет Function/Class-узел по (file_path, symbol)
+ и возвращает test-узлы, приходящие в него по EdgeType.TESTS.
+
+ Args:
+ symbol: имя символа (без qualified-префикса).
+ file_path: файл символа (POSIX-путь, как в графе).
+ limit: максимум возвращаемых тестов (защита от hub-функций
+ с сотнями тестов, напр. safe_mkdir → 234).
+
+ Returns:
+ List[SymbolRef] с kind="test", is_definition=True.
+ """
+ try:
+ candidates = self._graph.find_nodes(
+ label=NodeLabel.FUNCTION,
+ name_pattern=f"%{symbol}%",
+ file_path=file_path,
+ limit=5,
+ )
+ if not candidates:
+ return []
+ refs: List[SymbolRef] = []
+ for node in candidates[:2]: # не больше 2 одноимённых узлов
+ neighbors = self._graph.get_neighbors(
+ node.qualified_name,
+ edge_type=EdgeType.TESTS,
+ direction="incoming",
+ max_nodes=limit * 2,
+ )
+ for neighbor, _edge, _depth in neighbors:
+ if len(refs) >= limit:
+ break
+ if neighbor.label != NodeLabel.TEST:
+ continue
+ refs.append(SymbolRef(
+ symbol=neighbor.name,
+ file_path=neighbor.file_path,
+ line=neighbor.properties.get("line", 0),
+ kind="test",
+ is_definition=True,
+ ))
+ if len(refs) >= limit:
+ break
+ return refs
+ except Exception as e: # noqa: BLE001
+ # Graceful degradation: сбой графа не должен ломать graph-stage;
+ # TESTS-сигнал опционален по своей природе (флаг off по умолчанию).
+ logger.debug(f"get_tests_for_symbol error: {e}")
+ return []
+
# ── Call Graph ────────────────────────────────────────
def build_call_graph(self, symbol: str, depth: int = 2) -> Dict:
diff --git a/tests/test_graph_stage_e4.py b/tests/test_graph_stage_e4.py
index ca574770..9f88e030 100644
--- a/tests/test_graph_stage_e4.py
+++ b/tests/test_graph_stage_e4.py
@@ -91,3 +91,127 @@ def test_short_circuit_identifier_no_embedder(searcher_with_graph):
assert len(results) >= 1
assert results[0]["metadata"]["graph_stage"] is True
assert "index_guard.py" in results[0]["metadata"]["file"]
+
+
+# ════════════════════════════════════════════════════════════════
+# E17: TESTS-сигнал — покрывающие тесты в graph-stage под флагом
+# ════════════════════════════════════════════════════════════════
+
+class _NoTestsIndexer:
+ """symbol_index без метода get_tests_for_symbol — старый контракт."""
+
+ _symbol_index = None
+
+
+def _make_searcher_with_flag(indexer, flag):
+ class _FakeEmbedderNoCall:
+ def embed(self, *a, **k):
+ raise AssertionError("embedder не должен вызываться")
+
+ def embed_batch_async(self, *a, **k):
+ raise AssertionError("embedder не должен вызываться")
+
+ s = Searcher(indexer, _FakeEmbedderNoCall())
+ s._tests_signal = flag
+ return s
+
+
+@pytest.fixture
+def searcher_with_tests(tmp_path):
+ """Граф: функция + два теста, покрывающих её (TESTS-рёбра)."""
+ db = tmp_path / "graph.db"
+ pg = PropertyGraph(str(db))
+ # Функция в src/
+ pg.add_node(
+ name="safe_mkdir", label="Function", qualified_name="D:.D:/Project/src/core/paths.py.safe_mkdir",
+ file_path="D:/Project/src/core/paths.py",
+ properties={"line": 42, "kind": "function"},
+ )
+ # Тесты, покрывающие функцию
+ t1 = pg.add_node(
+ name="test_safe_mkdir_creates", label="Test",
+ qualified_name="D:.D:/Project/tests/test_paths.py.test_safe_mkdir_creates",
+ file_path="D:/Project/tests/test_paths.py",
+ properties={"line": 10},
+ )
+ t2 = pg.add_node(
+ name="test_safe_mkdir_idempotent", label="Test",
+ qualified_name="D:.D:/Project/tests/test_paths.py.test_safe_mkdir_idempotent",
+ file_path="D:/Project/tests/test_paths.py",
+ properties={"line": 20},
+ )
+ pg.add_edge(source_qname=t1.qualified_name, target_qname="D:.D:/Project/src/core/paths.py.safe_mkdir",
+ type="TESTS", weight=1.0, properties={"trace": "dynamic_trace"})
+ pg.add_edge(source_qname=t2.qualified_name, target_qname="D:.D:/Project/src/core/paths.py.safe_mkdir",
+ type="TESTS", weight=1.0, properties={"trace": "dynamic_trace"})
+ adapter = SymbolIndexAdapter(pg, mode=SymbolIndexAdapter.MODE_PURE)
+
+ class _Idx:
+ _symbol_index = adapter
+
+ s = _make_searcher_with_flag(_Idx(), flag=True)
+ yield s, adapter, pg
+ pg.close()
+
+
+def test_tests_signal_adds_covering_tests(searcher_with_tests):
+ """К определению функции добавляются покрывающие тесты."""
+ s, adapter, _ = searcher_with_tests
+ results = s._graph_stage("safe_mkdir", limit=5)
+ test_refs = [r for r in results if r["metadata"].get("tests_signal")]
+ assert len(test_refs) >= 1, f"TESTS-сигнал не сработал: {results}"
+ any(
+ t["metadata"]["symbol"].startswith("test_safe_mkdir")
+ for t in test_refs
+ ), f"ожидался тест safe_mkdir: {test_refs}"
+ for t in test_refs:
+ assert t["metadata"]["chunk_index"] < -10_000_000 # отдельный sentinel-диапазон
+ assert t["metadata"]["is_definition"] is False
+ assert t["metadata"]["covers"] == "safe_mkdir"
+ assert t["metadata"]["kind"] == "test"
+
+
+def test_tests_signal_def_first_then_tests(searcher_with_tests):
+ """Функция-определение стоит ДО тестов (тесты не вытесняют)."""
+ s, _, _ = searcher_with_tests
+ results = s._graph_stage("safe_mkdir", limit=5)
+ assert results[0]["metadata"]["symbol"] == "safe_mkdir"
+ assert results[0]["metadata"]["is_definition"] is True
+ r0_score = results[0]["graph_score"]
+ for t in results[1:]:
+ if t["metadata"].get("tests_signal"):
+ assert t["graph_score"] < r0_score
+
+
+def test_tests_signal_flag_off_unchanged(searcher_with_tests):
+ """Флаг off → тесты НЕ добавляются (поведение флага по умолчанию)."""
+ s, adapter, _ = searcher_with_tests
+ s._tests_signal = False
+ results = s._graph_stage("safe_mkdir", limit=5)
+ assert all(not r["metadata"].get("tests_signal") for r in results)
+ assert results[0]["metadata"]["symbol"] == "safe_mkdir"
+
+
+def test_tests_signal_no_method_fallback():
+ """symbol_index без get_tests_for_symbol → результаты без изменений."""
+ s = _make_searcher_with_flag(_NoTestsIndexer(), flag=True)
+ # _graph_stage с пустым индексом вернёт [] — это graceful degradation, без падения
+ assert s._graph_stage("anything", limit=5) == []
+
+
+def test_get_tests_for_symbol_hit(searcher_with_tests):
+ """Адаптер напрямую: возвращает покрывающие тесты."""
+ s, adapter, _ = searcher_with_tests
+ refs = adapter.get_tests_for_symbol("safe_mkdir", "D:/Project/src/core/paths.py", limit=3)
+ names = {r.symbol for r in refs}
+ assert "test_safe_mkdir_creates" in names
+ assert "test_safe_mkdir_idempotent" in names
+ for r in refs:
+ assert r.kind == "test"
+ assert r.is_definition is True
+
+
+def test_get_tests_for_symbol_miss(searcher_with_tests):
+ """Неизвестный символ → []."""
+ s, adapter, _ = searcher_with_tests
+ assert adapter.get_tests_for_symbol("nonexistent_xyz", "D:/Project/src/core/paths.py") == []