gbanyan/usher-exploring
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs(06): create phase plan

Browse Source
This commit is contained in:
gbanyan
2026-02-12 04:33:17 +08:00
parent ca2b715d8e
commit 844295c681
4 changed files with 570 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

158
.planning/phases/06-validation/06-01-PLAN.md Normal file
View File

@@ -0,0 +1,158 @@
---
phase: 06-validation
plan: 01
type: execute
wave: 1
depends_on: []
files_modified:
- src/usher_pipeline/scoring/negative_controls.py
- src/usher_pipeline/scoring/validation.py
- src/usher_pipeline/scoring/__init__.py
autonomous: true
must_haves:
truths:
- "Housekeeping genes are compiled as a curated negative control set with source provenance"
- "Negative control validation shows housekeeping genes rank below 50th percentile median"
- "Positive control validation reports recall@k metrics at k=10%, 20%, top-100"
- "Known genes achieve >70% recall in top 10% of scored candidates"
artifacts:
- path: "src/usher_pipeline/scoring/negative_controls.py"
provides: "Housekeeping gene compilation and negative control validation"
exports: ["HOUSEKEEPING_GENES_CORE", "compile_housekeeping_genes", "validate_negative_controls"]
- path: "src/usher_pipeline/scoring/validation.py"
provides: "Enhanced positive control validation with recall@k and per-source breakdown"
exports: ["validate_known_gene_ranking", "compute_recall_at_k", "generate_validation_report"]
- path: "src/usher_pipeline/scoring/__init__.py"
provides: "Updated exports including negative control functions"
contains: "validate_negative_controls"
key_links:
- from: "src/usher_pipeline/scoring/negative_controls.py"
to: "DuckDB scored_genes table"
via: "PERCENT_RANK window function query"
pattern: "PERCENT_RANK.*ORDER BY composite_score"
- from: "src/usher_pipeline/scoring/validation.py"
to: "src/usher_pipeline/scoring/known_genes.py"
via: "compile_known_genes import"
pattern: "from usher_pipeline.scoring.known_genes import"
---
<objective>
Implement negative control validation with housekeeping genes and enhance positive control validation with recall@k metrics.
Purpose: Negative controls ensure the scoring system does not indiscriminately rank all genes high (complementing the existing positive control validation). Enhanced positive control metrics (recall@k) provide the specific ">70% in top 10%" measurement required by success criteria.
Output: Two modules -- negative_controls.py (new) and enhanced validation.py (updated) -- ready for integration into the comprehensive validation report (Plan 03).
</objective>
<execution_context>
@/Users/gbanyan/.claude/get-shit-done/workflows/execute-plan.md
@/Users/gbanyan/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/06-validation/06-RESEARCH.md
@src/usher_pipeline/scoring/validation.py
@src/usher_pipeline/scoring/known_genes.py
@src/usher_pipeline/scoring/quality_control.py
@src/usher_pipeline/scoring/__init__.py
@src/usher_pipeline/persistence/duckdb_store.py
</context>
<tasks>
<task type="auto">
<name>Task 1: Create negative control validation module with housekeeping genes</name>
<files>src/usher_pipeline/scoring/negative_controls.py</files>
<action>
Create `src/usher_pipeline/scoring/negative_controls.py` with:
1. **HOUSEKEEPING_GENES_CORE** frozenset constant containing 13 curated housekeeping genes:
RPL13A, RPL32, RPLP0, GAPDH, ACTB, B2M, HPRT1, TBP, SDHA, PGK1, PPIA, UBC, YWHAZ.
Include inline comments grouping by function (ribosomal, metabolic, transcription/reference).
2. **compile_housekeeping_genes() -> pl.DataFrame** function returning DataFrame with columns:
- gene_symbol (str)
- source (str): "literature_validated" for all
- confidence (str): "HIGH" for all
Follow the exact same pattern as `compile_known_genes()` in known_genes.py.
3. **validate_negative_controls(store: PipelineStore, percentile_threshold: float = 0.50) -> dict** function:
- Register housekeeping genes as temporary DuckDB table `_housekeeping_genes`
- Use the same PERCENT_RANK window function pattern as `validate_known_gene_ranking()` in validation.py
- Query: join ranked_genes CTE with _housekeeping_genes on gene_symbol
- INVERTED validation logic: `validation_passed = median_percentile < percentile_threshold`
- Return dict with keys: total_expected, total_in_dataset, median_percentile, top_quartile_count, in_high_tier_count, validation_passed, housekeeping_gene_details (top 20 by percentile ASC)
- Clean up temp table after query
- Use structlog logger with info/warning levels matching validation.py patterns
4. **generate_negative_control_report(metrics: dict) -> str** function:
- Follow the exact formatting pattern from generate_validation_report() in validation.py
- Show gene table with Score, Percentile, headers
- Include interpretation text for pass/fail
Use structlog, polars, duckdb imports matching existing scoring module patterns. Import PipelineStore from usher_pipeline.persistence.duckdb_store.
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -c "from usher_pipeline.scoring.negative_controls import HOUSEKEEPING_GENES_CORE, compile_housekeeping_genes, validate_negative_controls, generate_negative_control_report; df = compile_housekeeping_genes(); print(f'Housekeeping genes: {df.height}'); assert df.height == 13; assert set(df.columns) == {'gene_symbol', 'source', 'confidence'}; print('OK')"` exits 0
</verify>
<done>
negative_controls.py exists with 13 curated housekeeping genes, compile function returns correct DataFrame structure, validate function uses PERCENT_RANK with inverted threshold logic, report function generates human-readable output.
</done>
</task>
<task type="auto">
<name>Task 2: Enhance positive control validation with recall@k metrics</name>
<files>src/usher_pipeline/scoring/validation.py, src/usher_pipeline/scoring/__init__.py</files>
<action>
**In validation.py**, add the following functions (do NOT modify existing functions, only ADD):
1. **compute_recall_at_k(store: PipelineStore, k_values: list[int] | None = None) -> dict** function:
- Default k_values: [100, 500, 1000, 2000] (absolute counts)
- Also compute recall at percentage thresholds: top 5%, 10%, 20% of scored genes
- Query scored_genes ordered by composite_score DESC (WHERE NOT NULL)
- For each k: count how many known genes (from compile_known_genes, deduplicated on gene_symbol) appear in top-k
- Recall@k = found_in_top_k / total_known_unique
- Return dict with: recalls_absolute (dict mapping k -> recall float), recalls_percentage (dict mapping pct_string -> recall float), total_known_unique (int), total_scored (int)
- Use structlog for logging results
2. **validate_positive_controls_extended(store: PipelineStore, percentile_threshold: float = 0.75) -> dict** function:
- Call existing validate_known_gene_ranking(store, percentile_threshold) to get base metrics
- Call compute_recall_at_k(store) to get recall metrics
- Add per-source breakdown: compute median percentile separately for "omim_usher" and "syscilia_scgs_v2" genes
- Per-source query: same PERCENT_RANK CTE but filter JOIN by source
- Return dict combining base metrics + recall_at_k + per_source_breakdown (dict mapping source -> {median_percentile, count, top_quartile_count})
- This is the "full" positive control validation for Phase 6
**In __init__.py**, add exports for: compute_recall_at_k, validate_positive_controls_extended, and also add imports/exports for negative_controls module: HOUSEKEEPING_GENES_CORE, compile_housekeeping_genes, validate_negative_controls, generate_negative_control_report.
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -c "from usher_pipeline.scoring import compute_recall_at_k, validate_positive_controls_extended, HOUSEKEEPING_GENES_CORE, compile_housekeeping_genes, validate_negative_controls, generate_negative_control_report; print('All imports OK')"` exits 0
</verify>
<done>
validation.py has compute_recall_at_k and validate_positive_controls_extended functions. __init__.py exports all new functions from both negative_controls.py and updated validation.py. Recall@k computes at both absolute and percentage thresholds. Per-source breakdown separates OMIM from SYSCILIA metrics.
</done>
</task>
</tasks>
<verification>
- `python -c "from usher_pipeline.scoring.negative_controls import HOUSEKEEPING_GENES_CORE; assert len(HOUSEKEEPING_GENES_CORE) == 13"` -- housekeeping genes compiled
- `python -c "from usher_pipeline.scoring import validate_negative_controls, compute_recall_at_k, validate_positive_controls_extended"` -- all functions importable
- `python -c "from usher_pipeline.scoring.negative_controls import compile_housekeeping_genes; df = compile_housekeeping_genes(); assert 'gene_symbol' in df.columns and 'source' in df.columns"` -- DataFrame structure correct
</verification>
<success_criteria>
- negative_controls.py creates housekeeping gene set and validates they rank low (inverted threshold)
- validation.py compute_recall_at_k measures recall at multiple k values including percentage-based thresholds
- validate_positive_controls_extended combines percentile + recall + per-source metrics
- All new functions exported from scoring.__init__
</success_criteria>
<output>
After completion, create `.planning/phases/06-validation/06-01-SUMMARY.md`
</output>

195
.planning/phases/06-validation/06-02-PLAN.md Normal file
View File

@@ -0,0 +1,195 @@
---
phase: 06-validation
plan: 02
type: execute
wave: 1
depends_on: []
files_modified:
- src/usher_pipeline/scoring/sensitivity.py
autonomous: true
must_haves:
truths:
- "Sensitivity analysis perturbs each weight by +-5% and +-10% and measures rank stability"
- "Spearman rank correlation is computed for top-100 genes between baseline and perturbed configurations"
- "Weight perturbation renormalizes remaining weights to maintain sum=1.0 constraint"
- "Rank stability assessment classifies each perturbation as stable (rho>=0.85) or unstable"
artifacts:
- path: "src/usher_pipeline/scoring/sensitivity.py"
provides: "Parameter sweep sensitivity analysis with Spearman correlation"
exports: ["perturb_weight", "run_sensitivity_analysis", "summarize_sensitivity"]
key_links:
- from: "src/usher_pipeline/scoring/sensitivity.py"
to: "src/usher_pipeline/scoring/integration.py"
via: "compute_composite_scores import"
pattern: "from usher_pipeline.scoring.integration import compute_composite_scores"
- from: "src/usher_pipeline/scoring/sensitivity.py"
to: "scipy.stats"
via: "spearmanr import"
pattern: "from scipy.stats import spearmanr"
- from: "src/usher_pipeline/scoring/sensitivity.py"
to: "src/usher_pipeline/config/schema.py"
via: "ScoringWeights import"
pattern: "from usher_pipeline.config.schema import ScoringWeights"
---
<objective>
Implement sensitivity analysis module for parameter sweep validation of scoring weights.
Purpose: Demonstrates that top candidate rankings are robust to reasonable weight perturbations (+-5-10%), satisfying success criterion 3 (rank stability). This is the core of the "are our results defensible?" validation.
Output: sensitivity.py module with weight perturbation, Spearman correlation analysis, and stability classification.
</objective>
<execution_context>
@/Users/gbanyan/.claude/get-shit-done/workflows/execute-plan.md
@/Users/gbanyan/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/06-validation/06-RESEARCH.md
@src/usher_pipeline/scoring/integration.py
@src/usher_pipeline/config/schema.py
@src/usher_pipeline/persistence/duckdb_store.py
</context>
<tasks>
<task type="auto">
<name>Task 1: Create sensitivity analysis module with weight perturbation and rank correlation</name>
<files>src/usher_pipeline/scoring/sensitivity.py</files>
<action>
Create `src/usher_pipeline/scoring/sensitivity.py` with:
1. **EVIDENCE_LAYERS** list constant: ["gnomad", "expression", "annotation", "localization", "animal_model", "literature"]
2. **DEFAULT_DELTAS** list constant: [-0.10, -0.05, 0.05, 0.10]
3. **STABILITY_THRESHOLD** float constant: 0.85 (Spearman rho threshold for "stable")
4. **perturb_weight(baseline: ScoringWeights, layer: str, delta: float) -> ScoringWeights** function:
- Get baseline weights as dict via baseline.model_dump()
- Apply perturbation: w_dict[layer] = max(0.0, min(1.0, w_dict[layer] + delta))
- Renormalize ALL weights so they sum to 1.0: divide each by total
- Return new ScoringWeights instance
- Raise ValueError if layer not in EVIDENCE_LAYERS
5. **run_sensitivity_analysis(store: PipelineStore, baseline_weights: ScoringWeights, deltas: list[float] | None = None, top_n: int = 100) -> dict** function:
- Default deltas to DEFAULT_DELTAS if None
- Compute baseline scores via compute_composite_scores(store, baseline_weights)
- Sort by composite_score DESC, take top_n genes as baseline ranking
- For each layer in EVIDENCE_LAYERS, for each delta in deltas:
- Create perturbed weights via perturb_weight()
- Compute perturbed scores via compute_composite_scores(store, perturbed_weights)
- Sort by composite_score DESC, take top_n genes
- Inner join baseline and perturbed on gene_symbol to get paired scores
- If fewer than 10 overlapping genes, log warning and record rho=None
- Otherwise compute spearmanr() on paired composite_score columns
- Record: layer, delta, perturbed_weights (as dict), spearman_rho, spearman_pval, overlap_count (how many of top_n genes appear in both), top_n
- Return dict with keys: baseline_weights (dict), results (list of per-perturbation dicts), top_n, total_perturbations
- Use structlog for progress logging (log each perturbation result)
IMPORTANT: The compute_composite_scores function re-queries the DB each time. This is by design -- different weights produce different composite_score values from the same underlying evidence layer scores.
For the Spearman correlation, join baseline_top_n and perturbed_top_n DataFrames on gene_symbol (inner join). Use the composite_score from each as the paired values. This measures whether the relative ordering of shared top genes is preserved.
6. **summarize_sensitivity(analysis_result: dict) -> dict** function:
- From the results list, compute:
- min_rho, max_rho, mean_rho across all perturbations (excluding None values)
- count of stable perturbations (rho >= STABILITY_THRESHOLD)
- count of unstable perturbations (rho < STABILITY_THRESHOLD)
- most_sensitive_layer: layer with lowest mean rho across its perturbations
- most_robust_layer: layer with highest mean rho across its perturbations
- overall_stable: bool = all non-None rhos >= STABILITY_THRESHOLD
- Return dict with: min_rho, max_rho, mean_rho, stable_count, unstable_count, total_perturbations, overall_stable, most_sensitive_layer, most_robust_layer
7. **generate_sensitivity_report(analysis_result: dict, summary: dict) -> str** function:
- Follow the formatting pattern from generate_validation_report() in validation.py
- Show table: Layer | Delta | Spearman rho | p-value | Stable?
- Show summary: overall stability verdict, most/least sensitive layers
- Include interpretation text
Use structlog, polars, scipy.stats.spearmanr imports. Import compute_composite_scores from usher_pipeline.scoring.integration, ScoringWeights from usher_pipeline.config.schema, PipelineStore from usher_pipeline.persistence.duckdb_store.
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -c "
from usher_pipeline.scoring.sensitivity import perturb_weight, EVIDENCE_LAYERS, STABILITY_THRESHOLD, DEFAULT_DELTAS
from usher_pipeline.config.schema import ScoringWeights
# Test weight perturbation
w = ScoringWeights()
p = perturb_weight(w, 'gnomad', 0.10)
p.validate_sum() # Must not raise
print(f'Original gnomad: {w.gnomad}, Perturbed: {p.gnomad:.4f}')
assert p.gnomad > w.gnomad, 'Perturbed weight should be higher'
# Test renormalization
total = p.gnomad + p.expression + p.annotation + p.localization + p.animal_model + p.literature
assert abs(total - 1.0) < 1e-6, f'Weights must sum to 1.0, got {total}'
# Test edge: perturb to near-zero
p_low = perturb_weight(w, 'gnomad', -0.25)
p_low.validate_sum()
assert p_low.gnomad >= 0.0, 'Weight must not go negative'
print('All perturb_weight tests passed')
"` exits 0
</verify>
<done>
sensitivity.py exists with perturb_weight (renormalizing), run_sensitivity_analysis (computing Spearman rho for top-N genes across all layer/delta combinations), summarize_sensitivity (stability classification), and generate_sensitivity_report (formatted output). Weights always renormalize to sum=1.0 after perturbation.
</done>
</task>
<task type="auto">
<name>Task 2: Export sensitivity module from scoring package</name>
<files>src/usher_pipeline/scoring/__init__.py</files>
<action>
Update `src/usher_pipeline/scoring/__init__.py` to add imports and exports for the sensitivity module:
Add imports:
```python
from usher_pipeline.scoring.sensitivity import (
perturb_weight,
run_sensitivity_analysis,
summarize_sensitivity,
generate_sensitivity_report,
EVIDENCE_LAYERS,
STABILITY_THRESHOLD,
)
```
Add to __all__ list: "perturb_weight", "run_sensitivity_analysis", "summarize_sensitivity", "generate_sensitivity_report", "EVIDENCE_LAYERS", "STABILITY_THRESHOLD"
NOTE: Plan 01 may have already updated __init__.py to add negative_controls exports. If so, ADD the sensitivity imports alongside those -- do not remove them. Read the file first to check current state.
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -c "from usher_pipeline.scoring import perturb_weight, run_sensitivity_analysis, summarize_sensitivity, generate_sensitivity_report, EVIDENCE_LAYERS, STABILITY_THRESHOLD; print(f'EVIDENCE_LAYERS: {EVIDENCE_LAYERS}'); print(f'STABILITY_THRESHOLD: {STABILITY_THRESHOLD}'); print('All sensitivity exports OK')"` exits 0
</verify>
<done>
All sensitivity analysis functions and constants are importable from usher_pipeline.scoring. Existing exports from negative_controls (Plan 01) are preserved.
</done>
</task>
</tasks>
<verification>
- `python -c "from usher_pipeline.scoring.sensitivity import perturb_weight; from usher_pipeline.config.schema import ScoringWeights; w = ScoringWeights(); p = perturb_weight(w, 'gnomad', 0.05); p.validate_sum(); print('OK')"` -- weight perturbation works and renormalizes
- `python -c "from usher_pipeline.scoring import run_sensitivity_analysis, summarize_sensitivity, generate_sensitivity_report"` -- all exports available
- `python -c "from usher_pipeline.scoring.sensitivity import STABILITY_THRESHOLD; assert STABILITY_THRESHOLD == 0.85"` -- threshold configured
</verification>
<success_criteria>
- perturb_weight correctly perturbs one layer and renormalizes to sum=1.0
- run_sensitivity_analysis computes Spearman rho for all layer x delta combinations
- summarize_sensitivity classifies perturbations as stable/unstable
- generate_sensitivity_report produces human-readable output
- All functions exported from scoring package
</success_criteria>
<output>
After completion, create `.planning/phases/06-validation/06-02-SUMMARY.md`
</output>

211
.planning/phases/06-validation/06-03-PLAN.md Normal file
View File

@@ -0,0 +1,211 @@
---
phase: 06-validation
plan: 03
type: execute
wave: 2
depends_on: ["06-01", "06-02"]
files_modified:
- src/usher_pipeline/scoring/validation_report.py
- src/usher_pipeline/cli/validate_cmd.py
- src/usher_pipeline/cli/main.py
- tests/test_validation.py
autonomous: true
must_haves:
truths:
- "CLI validate command runs positive controls, negative controls, and sensitivity analysis in sequence"
- "Comprehensive validation report documents all three validation prongs with pass/fail verdicts"
- "Weight tuning recommendations are generated based on validation results with documented rationale"
- "Tests verify negative control logic, recall@k computation, weight perturbation, and report generation"
artifacts:
- path: "src/usher_pipeline/scoring/validation_report.py"
provides: "Comprehensive validation report combining all three validation prongs"
exports: ["generate_comprehensive_validation_report", "recommend_weight_tuning"]
- path: "src/usher_pipeline/cli/validate_cmd.py"
provides: "CLI validate command orchestrating full validation pipeline"
exports: ["validate"]
- path: "tests/test_validation.py"
provides: "Unit tests for negative controls, recall@k, sensitivity, and validation report"
key_links:
- from: "src/usher_pipeline/cli/validate_cmd.py"
to: "src/usher_pipeline/scoring/negative_controls.py"
via: "validate_negative_controls import"
pattern: "from usher_pipeline.scoring import validate_negative_controls"
- from: "src/usher_pipeline/cli/validate_cmd.py"
to: "src/usher_pipeline/scoring/sensitivity.py"
via: "run_sensitivity_analysis import"
pattern: "from usher_pipeline.scoring import run_sensitivity_analysis"
- from: "src/usher_pipeline/cli/validate_cmd.py"
to: "src/usher_pipeline/scoring/validation.py"
via: "validate_positive_controls_extended import"
pattern: "from usher_pipeline.scoring import validate_positive_controls_extended"
- from: "src/usher_pipeline/cli/main.py"
to: "src/usher_pipeline/cli/validate_cmd.py"
via: "Click group add_command"
pattern: "cli.add_command.*validate"
---
<objective>
Create comprehensive validation report generator, CLI validate command, and unit tests for all Phase 6 validation modules.
Purpose: This plan wires together the positive control, negative control, and sensitivity analysis modules (from Plans 01 and 02) into a single CLI command and comprehensive report. Tests ensure correctness with synthetic data. This completes Phase 6 by providing the user-facing validation workflow.
Output: validation_report.py, validate_cmd.py (CLI), updated main.py, and test_validation.py.
</objective>
<execution_context>
@/Users/gbanyan/.claude/get-shit-done/workflows/execute-plan.md
@/Users/gbanyan/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/06-validation/06-RESEARCH.md
# Need SUMMARYs from Plans 01 and 02 for what was actually built
@.planning/phases/06-validation/06-01-SUMMARY.md
@.planning/phases/06-validation/06-02-SUMMARY.md
@src/usher_pipeline/scoring/__init__.py
@src/usher_pipeline/scoring/negative_controls.py
@src/usher_pipeline/scoring/sensitivity.py
@src/usher_pipeline/scoring/validation.py
@src/usher_pipeline/cli/score_cmd.py
@src/usher_pipeline/cli/main.py
@tests/test_scoring.py
</context>
<tasks>
<task type="auto">
<name>Task 1: Create comprehensive validation report and CLI validate command</name>
<files>src/usher_pipeline/scoring/validation_report.py, src/usher_pipeline/cli/validate_cmd.py, src/usher_pipeline/cli/main.py</files>
<action>
**Create `src/usher_pipeline/scoring/validation_report.py`:**
1. **generate_comprehensive_validation_report(positive_metrics: dict, negative_metrics: dict, sensitivity_result: dict, sensitivity_summary: dict) -> str** function:
- Generate a multi-section Markdown report combining all three validation prongs
- Section 1: "Positive Control Validation" -- median percentile, recall@k table, per-source breakdown, pass/fail
- Section 2: "Negative Control Validation" -- median percentile, top quartile count, in-HIGH-tier count, pass/fail
- Section 3: "Sensitivity Analysis" -- Spearman rho table (layer x delta), overall stability verdict, most/least sensitive layers
- Section 4: "Overall Validation Summary" -- all-pass/partial-fail/fail verdict
- Section 5: "Weight Tuning Recommendations" -- call recommend_weight_tuning()
- Return the full Markdown string
2. **recommend_weight_tuning(positive_metrics: dict, negative_metrics: dict, sensitivity_summary: dict) -> str** function:
- Analyze validation results and suggest weight adjustments
- If positive controls pass AND negative controls pass AND sensitivity stable: "Current weights are validated. No tuning recommended."
- If positive controls fail: suggest increasing weights for layers where known genes score highly
- If negative controls fail (housekeeping genes ranking too high): suggest examining which layers boost housekeeping genes
- If sensitivity unstable: identify most sensitive layer and suggest reducing its weight
- Document rationale for each recommendation
- CRITICAL: Note that any tuning is "post-validation" and flag circular validation risk per research pitfall
- Return formatted recommendation text
3. **save_validation_report(report_text: str, output_path: Path) -> None**: Write report to file
**Create `src/usher_pipeline/cli/validate_cmd.py`:**
Follow the established CLI pattern from score_cmd.py (config load, store init, checkpoint, steps, summary, cleanup):
1. Click command `validate` with options:
- `--force`: Re-run even if validation checkpoint exists
- `--skip-sensitivity`: Skip sensitivity analysis (faster iteration)
- `--output-dir`: Output directory for validation report (default: {data_dir}/validation)
- `--top-n`: Top N genes for sensitivity analysis (default: 100)
2. Pipeline steps:
- Step 1: Load configuration and initialize store
- Step 2: Check scored_genes checkpoint exists (error if not -- must run `score` first)
- Step 3: Run positive control validation (validate_positive_controls_extended)
- Step 4: Run negative control validation (validate_negative_controls)
- Step 5: Run sensitivity analysis (unless --skip-sensitivity) -- run_sensitivity_analysis + summarize_sensitivity
- Step 6: Generate comprehensive validation report (generate_comprehensive_validation_report)
- Step 7: Save report to output_dir/validation_report.md and provenance sidecar
3. Use click.echo with styled output matching score_cmd.py patterns (green for success, yellow for warnings, red for errors, bold for step headers)
4. Provenance tracking: record_step for each validation phase with metrics
5. Final summary: display overall pass/fail, recall@top-10%, housekeeping median percentile, sensitivity stability
**Update `src/usher_pipeline/cli/main.py`:**
- Import validate from validate_cmd
- Add: `cli.add_command(validate)`
- Follow the existing pattern used for score and report commands
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -c "from usher_pipeline.cli.validate_cmd import validate; print(f'Command name: {validate.name}'); print('OK')"` exits 0 AND `cd /Users/gbanyan/Project/usher-exploring && python -c "from usher_pipeline.cli.main import cli; commands = list(cli.commands.keys()); print(f'CLI commands: {commands}'); assert 'validate' in commands; print('OK')"` exits 0
</verify>
<done>
validation_report.py generates comprehensive multi-section Markdown report with weight tuning recommendations. validate_cmd.py provides CLI command running all three validation prongs. main.py registers validate as a CLI subcommand. All follow established patterns from score_cmd.py.
</done>
</task>
<task type="auto">
<name>Task 2: Create unit tests for all validation modules</name>
<files>tests/test_validation.py</files>
<action>
Create `tests/test_validation.py` with comprehensive tests using synthetic DuckDB data. Follow the test pattern from tests/test_scoring.py (use tmp_path fixtures, create in-memory DuckDB with synthetic data).
**Test helper: create_synthetic_scored_db(tmp_path)** function:
- Create DuckDB with gene_universe (20 genes: GENE001-GENE020)
- Create scored_genes table with composite_score and all 6 layer scores
- Design scores so that:
- MYO7A, IFT88, BBS1 (known cilia genes) get high scores (0.8-0.95)
- GAPDH, ACTB, RPL13A (housekeeping genes) get low scores (0.1-0.3)
- Other genes get mid-range scores (0.3-0.6)
- This ensures positive controls rank high and negative controls rank low in tests
**Tests to include:**
1. **test_compile_housekeeping_genes_structure**: Verify compile_housekeeping_genes() returns DataFrame with 13 genes, correct columns (gene_symbol, source, confidence), all confidence=HIGH, all source=literature_validated
2. **test_compile_housekeeping_genes_known_genes_present**: Assert GAPDH, ACTB, RPL13A, TBP are in the gene_symbol column
3. **test_validate_negative_controls_with_synthetic_data**: Use synthetic DB where housekeeping genes score low. Assert validation_passed=True, median_percentile < 0.5
4. **test_validate_negative_controls_inverted_logic**: Create a DB where housekeeping genes score HIGH (artificial scenario). Assert validation_passed=False
5. **test_compute_recall_at_k**: Use synthetic DB. Assert recall@k returns dict with recalls_absolute and recalls_percentage keys. With 3 known genes in top 5 of 20, recall@5 should be high (>0.5)
6. **test_perturb_weight_renormalizes**: Perturb gnomad by +0.10, assert weights still sum to 1.0. Perturb by -0.25 (more than weight value), assert weight >= 0.0 and sum = 1.0
7. **test_perturb_weight_invalid_layer**: perturb_weight with layer="nonexistent" should raise ValueError
8. **test_generate_comprehensive_validation_report_format**: Pass mock metrics dicts, assert report contains expected sections ("Positive Control", "Negative Control", "Sensitivity Analysis", "Weight Tuning")
9. **test_recommend_weight_tuning_all_pass**: Pass metrics indicating all validations pass. Assert response contains "No tuning recommended" or similar
All tests should use tmp_path for DuckDB isolation. Import from usher_pipeline.scoring (not internal modules directly where possible). Use PipelineStore with direct conn assignment pattern from test_scoring.py.
</action>
<verify>
Run: `cd /Users/gbanyan/Project/usher-exploring && python -m pytest tests/test_validation.py -v --tb=short` -- all tests pass
</verify>
<done>
test_validation.py contains 9+ tests covering negative controls, recall@k, weight perturbation, sensitivity analysis, and report generation. All tests pass using synthetic DuckDB data with designed score patterns ensuring known genes rank high and housekeeping genes rank low.
</done>
</task>
</tasks>
<verification>
- `python -m pytest tests/test_validation.py -v` -- all validation tests pass
- `python -c "from usher_pipeline.cli.main import cli; assert 'validate' in cli.commands"` -- CLI command registered
- `python -c "from usher_pipeline.scoring.validation_report import generate_comprehensive_validation_report, recommend_weight_tuning"` -- report functions importable
- `usher-pipeline validate --help` displays usage information with all options
</verification>
<success_criteria>
- CLI `validate` command runs positive + negative + sensitivity validations and generates comprehensive report
- Validation report includes all three prongs with pass/fail verdicts and weight tuning recommendations
- Unit tests cover negative controls, recall@k, perturbation, and report generation
- All tests pass with synthetic data
- validate command registered in main CLI
</success_criteria>
<output>
After completion, create `.planning/phases/06-validation/06-03-SUMMARY.md`
</output>