feat: add Organization evaluator (early access)

[slee430]

Summary

Adds the external-facing package for the Organization dimension text-complexity evaluator, mirroring the existing intertextuality package structure.

What's included

• config.json — evaluator spec (id literacy.gla.organization), preprocessing (FK score), single LLM step, prompt sha256 pins, and fixtures tolerance
• system.txt / user.txt — evaluation prompts
• input_schema.json / output_schema.json — I/O contracts
• fixtures.json — sniff-test cases drawn from unanimously human-agreed items in the annotated sniff test
• example_notebook.ipynb — runnable demo + drift check + sniff test

Verification

• Prompt sha256 hashes verified against config.json
• example_notebook.ipynb runs end to end
• Sniff test: 7/7 pass (5 exact, 2 adjacent within tolerance)

Notes

• Output schema follows the public package shape (complexity_score + reasoning + details).
• Fixtures are currently weighted toward slightly_complex; more very/exceedingly examples can be added later as expert-labeled data becomes available.

[Copilot]

Pull request overview

Adds a new external-facing evaluator package for the Organization dimension under evals/literacy/qualitative-text-complexity/organization, intended to mirror the existing qualitative text complexity evaluator package structure.

Changes:

• Added Organization evaluator prompt assets (system.txt, user.txt) and JSON I/O schemas.
• Added evaluator configuration (config.json) and sniff-test fixtures (fixtures.json).
• Added a runnable example notebook (example_notebook.ipynb) demonstrating evaluation + drift checking + fixture sniff testing.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 13 comments.

Show a summary per file

File	Description
evals/literacy/qualitative-text-complexity/organization/config.json	Defines evaluator metadata, preprocessing, LLM step, schema refs, and fixture settings (currently diverges from repo config schema).
evals/literacy/qualitative-text-complexity/organization/system.txt	System prompt defining the Organization rubric and evaluation guardrails.
evals/literacy/qualitative-text-complexity/organization/user.txt	User prompt template with {text}, {grade_level}, {fk_score} placeholders.
evals/literacy/qualitative-text-complexity/organization/input_schema.json	Input contract for text + grade_level.
evals/literacy/qualitative-text-complexity/organization/output_schema.json	Output contract for complexity_score, reasoning, and details.
evals/literacy/qualitative-text-complexity/organization/fixtures.json	Sniff-test fixture inputs + expected labels (currently uses the wrong expected key name).
evals/literacy/qualitative-text-complexity/organization/example_notebook.ipynb	Example workflow: load config/prompts, compute FK score, run structured output, and execute fixtures sniff test.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[aychi1]

Overall LGTM, there are just a few minor changes needed.

[aychi1]

Hi Seung Yeon, for this, you can cross-compare with lines 75-90 of the Intertextuality config: https://github.com/learning-commons-org/evaluators/blob/main/evals/literacy/qualitative-text-complexity/intertextuality/config.json

[slee430]

@aychi1 Thanks for the thorough review! I've addressed all the comments:

• Set temperature to 1
• Renamed the fixtures label key from complexity_level to complexity_score to match output_schema.json, and updated the notebook's sniff-test reader accordingly
• Brought config.json into full conformance with config.schema.json: switched spec_version to $schema, added supported_grades, added type to the preprocessing and step entries, renamed fixtures.sniff_test_path to fixtures.path, and removed fields the schema doesn't allow (output_binding, the extra parser fields, the unused format_instructions placeholder, and the tracing block)
• Stripped the notebook's execution artifacts so the strip-notebooks check passes

Re-ran the example notebook end to end — sniff test is now 7/7 exact. All 5 local checks pass.