feat(evals): add Strength Acknowledgment onto the ela-writing taxonomy

[adnanrhussain]

Migrates Acknowledges Strength onto the evals/feedback/ela-writing/ taxonomy as Strength Acknowledgement, at evals/feedback/ela-writing/strength-acknowledgement/. Same pattern as the student-facing-text family (#159, #161, #163, #173–#177). The evaluator already had a complete contract; this moves it, renames it, and closes the gaps that family standardised on.

Evaluator

• evaluator.id: feedback.ela_writing.strength_acknowledgement
• stable_id: 99370b62-6791-4535-a77e-dda6806c635c, id_history: ["feedback.productive_coaching_writing_feedback.is_acknowledges_strength"]
• evaluator.name: Strength Acknowledgement Evaluator — drops the Feedback-Quality suffix so it matches the <X> Evaluator form used across student-facing-text
• Step id evaluate_is_acknowledges_strength → evaluate_strength_acknowledgement
• Single LLM step on gpt-5.4-2026-03-05, temperature 1, unchanged

quality_score

The output score field was named after the evaluator (acknowledges_strength_score). Every evaluator in this family had a differently-named field for the same binary judgement, so a consumer had to know which evaluator ran before it could read the result.

Student-facing-text names this after the measured dimension — complexity_score, identical across all 8 of its evaluators. This adopts the same convention: quality_score, shared across all 7 feedback evaluators. Values are unchanged (integer, 0/1).

Renamed in output_schema.json (property, required[], description), fixtures.json (expected), and the notebook. Nothing consumes these configs — no TypeScript or Python SDK reads them — so there is no downstream breakage.

Also fixed

The notebook carried two copy-paste references to evals/prompts/purpose, a leftover from the Purpose evaluator it was templated from. Both now point at this directory.

Not applicable

The grade_level string standardisation doesn't apply — this evaluator takes student_text and feedback_text, no grade input. supported_grades: ["8", "9"] is metadata only.

Verification

scripts/check.py all pass.

Spelling: "acknowledgement"

The public name uses acknowledgement; the repo used acknowledgment in 14 places. Both are valid — the former is the British variant, the latter standard US — so this is a variant choice, not a typo fix. Standardised on acknowledgement throughout, including system.txt.

Editing prompt text is deliberately out of scope for a rename elsewhere in this migration series, so calling it out: it was authorised here specifically, because leaving the prompt on the other spelling would make the evaluator internally inconsistent with its own name and folder. The pinned sha256 for system.txt is recomputed to match (d66c26d4… → 3a96d931…), and scripts/check.py verifies it.

[Copilot]

Pull request overview

This PR migrates the “Acknowledges Strength” feedback-quality evaluator into the evals/feedback/ela-writing/strength-acknowledgement/ taxonomy as Strength Acknowledgement, aligning it with the standardized, config-driven evaluator contract (schemas, fixtures, notebook) used across the evaluator set.

Changes:

• Adds the evaluator contract artifacts at the new taxonomy path (config.json, input_schema.json, output_schema.json, fixtures.json, prompts, and notebook).
• Renames the evaluator identity and step id, and standardizes the binary output field name to quality_score.
• Updates prompt content spelling to “acknowledgement” and refreshes the pinned prompt hash in config.json.

Reviewed changes

Copilot reviewed 5 out of 7 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
evals/feedback/ela-writing/strength-acknowledgement/config.json	New evaluator identity (id/stable_id/id_history/name) and step id; pins prompt hashes.
evals/feedback/ela-writing/strength-acknowledgement/input_schema.json	Defines structured inputs (student_text, feedback_text) for the evaluator contract.
evals/feedback/ela-writing/strength-acknowledgement/output_schema.json	Renames the binary output field to quality_score and updates schema metadata.
evals/feedback/ela-writing/strength-acknowledgement/fixtures.json	Updates expected outputs to use quality_score.
evals/feedback/ela-writing/strength-acknowledgement/system.txt	Updates prompt wording/spelling; retains evaluation rubric and examples.
evals/feedback/ela-writing/strength-acknowledgement/user.txt	Adds the user prompt template using {student_text} and {feedback_text} placeholders.
evals/feedback/ela-writing/strength-acknowledgement/example_notebook.ipynb	Updates notebook naming, paths, function name, and expected field reads to quality_score.

Suppressed comments (2)

evals/feedback/ela-writing/strength-acknowledgement/example_notebook.ipynb:118

• Typo in the notebook docstring: “wheter” should be “whether”. Note this same typo also appears in other evaluator notebooks, so fixing only here would leave inconsistent docs.
  evals/feedback/ela-writing/strength-acknowledgement/system.txt:78
• The prompt text still instructs the model to output an overall answer (and the calibration examples label the correct judgment as answer = ...), but this evaluator’s structured output schema uses quality_score. With parser.kind = structured_output, this terminology mismatch can confuse the model and/or reduce adherence to the schema. Update the prompt language to consistently refer to quality_score (and update the example judgments accordingly).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[aychi1]

LGTM, conditional on P0 max length.