Add comp cache, unit tests, and tiered evaluation framework#202
Open
akivanc88 wants to merge 3 commits intosantifer:mainfrom
Open
Add comp cache, unit tests, and tiered evaluation framework#202akivanc88 wants to merge 3 commits intosantifer:mainfrom
akivanc88 wants to merge 3 commits intosantifer:mainfrom
Conversation
- Tiered evaluation: Stage 0 pre-screen gates full A-F evaluation behind
preliminary score (0.4×archetype fit + 0.6×CV overlap). Offers scoring
<3.0 skip immediately with a SKIP TSV — no WebSearch, no STAR stories.
Verbosity tiers: 3.0-3.9→A+B only, 4.0-4.4→A-E, ≥4.5→full A-G.
- Comp research caching: new comp-cache.mjs (no new deps) caches salary
data from WebSearch with 60-day TTL. Modes check cache before querying
Glassdoor/Levels.fyi/Blind, saving 3 WebSearch calls per repeated role.
- JD summarization: long JDs (>1500 words) are summarized before the
pre-screen to reduce input tokens for the cheap filtering step.
- Batch context pre-building: batch-runner.sh reads cv.md + profile.yml
once before spawning workers and inlines them via {{CONTEXT_PRELOADED}},
eliminating per-worker Read tool calls for static files.
- Test infrastructure: 106 tests across 4 files using Node built-in
node:test (no new deps). Refactored merge-tracker, verify-pipeline, and
normalize-statuses to export pure functions under a main guard.
https://claude.ai/code/session_01PQ1cMmnoQPQTeR1cZYkq4e
This commit translates all mode instruction files and related systems from Spanish to English while preserving all functionality: Mode files translated: - modes/oferta.md → English with preserved technical terminology - modes/auto-pipeline.md → Translated pipeline stages and instructions - modes/_shared.md → Translated archetype definitions and guidelines - modes/ofertas.md → Translated comparison matrix terminology - modes/scan.md → Translated portal scanning instructions - modes/tracker.md → Translated tracker status explanations - modes/apply.md → Translated application assistant mode - batch/batch-prompt.md → Translated batch worker prompt and section labels Code changes: - Updated CANONICAL_STATES arrays in merge-tracker.mjs, verify-pipeline.mjs, and normalize-statuses.mjs from Spanish to English values - Updated aliases to support both Spanish and English input, normalizing to English output - Updated all test fixtures and test assertions to use English status values - All 106 tests passing with new English-based canonical statuses Status values (Spanish → English): - Evaluada → Evaluated - Aplicado → Applied - Respondido → Responded - Entrevista → Interview - Oferta → Offer - Rechazado → Rejected - Descartado → Discarded - NO APLICAR → SKIP Backward compatibility maintained: system accepts Spanish input (for existing trackers) and normalizes to English output. https://claude.ai/code/session_01PQ1cMmnoQPQTeR1cZYkq4e
- normalize-statuses.mjs: rechazada? only matched 'rechazada', not 'rechazado' — changed to rechazad[ao] to cover both forms - test: monitor correctly maps to SKIP (not Evaluated) per states.yml https://claude.ai/code/session_01PQ1cMmnoQPQTeR1cZYkq4e
akivanc88
force-pushed
the
claude/review-project-claims-I2L61
branch
from
b322dd0 to
069875a
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
This PR introduces three major improvements to the career-ops pipeline:
Compensation Research Cache (
comp-cache.mjs): A local YAML-based cache for salary data (p25/p50/p75) with 60-day TTL, eliminating repeated WebSearch calls for the same role/stage/location combinations. Includes CLI for lookup, save, list, and purge operations.Comprehensive Unit Tests: Added test suites for all major modules (
comp-cache.test.mjs,verify-pipeline.test.mjs,merge-tracker.test.mjs,normalize-statuses.test.mjs) covering pure functions and edge cases. Tests use Node's built-innode:testframework with no external dependencies.Tiered Evaluation Framework: Introduced Stage 0 pre-screen (cheap, ~1K tokens) that runs before full evaluation blocks. Pre-screen scores alignment to North Star archetypes and overlap with must-haves, allowing early rejection of misaligned roles without expensive full analysis.
Additional improvements:
verify-pipeline.mjsto export pure validation functions for testabilitybatch-runner.shto inject cv.md + profile.yml once, reducing tool calls.gitignoreto excludedata/comp-cache.yml(local-only, personal salary data)Related issue
N/A (foundational infrastructure improvements)
Type of change
Checklist
Test Plan
Run the new test suite:
npm testAll tests pass, covering:
The cache CLI can be tested manually:
https://claude.ai/code/session_01PQ1cMmnoQPQTeR1cZYkq4e