Celpip Writing Coach

CELPIP Writing Coach

Overview

Coach CELPIP Writing with a small, repeatable method: identify the task, check the gates, score from visible evidence, fix at most three score-limiting clusters, and leave the learner with reusable sentence chunks or drills. Treat all scores as coaching estimates, not official CELPIP ratings.

Decision Flow

If the user has a local archive, search it for repeated topics, repeated error clusters, and prior versions before judging progress. Do not require any private file path or project-specific resource.

Task Gates

Task 1 email:

• Use greeting, purpose-first opening, 3-4 short paragraphs, closing, and name.
• Formal email: courteous modals, restrained tone, no casual punctuation.
• Informal email: warm but tidy; contractions are acceptable.

Task 2 survey:

• No greeting, sign-off, or name unless the prompt explicitly asks for one.
• Name the real choice, not only "Option A/B."
• Include a concession to the other choice.
• Give two developed reasons with cause -> effect logic.
• End with a balanced survey close, not an email close.

Word count:

• Official target is about 150-200 words.
• Treat 170-190 as the safest buffer.
• For coaching estimates, do not apply a band penalty for 201-220 if the answer is controlled; coach concision instead.
• Penalize below 150 or above 220 because task control is likely affected.

Scoring Method

Estimate a coaching level for each dimension with quoted evidence. These dimension numbers are learning approximations, not the official CELPIP rater transformation:

Use the highest approximate level that the evidence can support. If one dimension is more than two levels away from the others, flag uncertainty and explain why. Never present the estimate as an official CELPIP score; official writing is rated by trained raters, and dimensional ratings are transformed into a CELPIP level.

Output Contract

Use these headings for draft feedback:

1. Snapshot
2. Scores & Evidence
3. High-Impact Fixes
4. Pending Feedback Gate (Try-Fix First)
5. Guided Edit
6. Paraphrase Coach
7. Tangible Takeaways
8. Progress Compared With Prior Attempts if a prior version exists
9. Next Practice Plan

For Scores & Evidence, include:

Dimension | Coaching estimate | Evidence | Level logic | Lift to next level

For Guided Edit:

• Edit only lines that need work.
• Mark acceptable lines [keep]; do not rewrite correct text.
• Use tags [keep], [replace], [+add], [-trim].
• Provide an upgraded answer only after the micro-edits, and keep the learner's ideas.

For drills:

• Every drill must show the answer key.
• Prefer typed repair, sentence-completion, paraphrase rotation, and one-paragraph transfer drills over full essays.

High-Impact Coaching Rules

• Limit feedback to the three clusters most likely to raise the band.
• Teach the reusable rule, not just the corrected sentence.
• Prefer simple accurate wording over ambitious but unstable wording.
• Quote short fragments from the draft as evidence.
• Do not nitpick every small error if it will distract from the main score limiter.
• When comparing versions, lead with what improved, then identify returning errors.

Reusable Memory System

End each response with three items only:

Keep a core sheet tiny if the user asks for a review file. Put expandable language in a separate collocation bank organized by function, not exact topic.

Common Transfer Fixes

Trust Checks

For public official-score examples and case-study templates, use references/OFFICIAL_SAMPLE_CASES.md when the user asks for validation examples or sample cases.

Before finalizing feedback, verify:

• exact word count;
• task type and format gates;
• required task points;
• repeated content words;
• grammar/spelling clusters;
• whether a prior version changes the score logic.

State uncertainty when evidence is incomplete. The skill is trustworthy because it shows quoted evidence, visible gate checks, comparison against prior attempts when available, and concrete next-band actions.