The hard part isn't writing forty remarks. It's saying the same true thing forty different ways.

The engine — honest about the mechanism

Many education-tech vendors call this "AI report card writing." It is mostly not AI. SchoolDeck calls it what it is: a rule-based template engine with variable substitution. Here is what actually happens.

1. Library: 500+ sentence templates, each authored by Indian academic editors and reviewed by school Academic Councils. Each template carries metadata — tier (Mastery / Proficient / Developing / Foundational-Support), NEP HPC domain (academic / co-scholastic / socio-emotional), subject applicability, and any token slots like [Name] [He/She] [Subject] [Topic] [Medium].
2. Selection: The teacher picks 2-4 attributes that match her actual observation of the student — e.g. "strong foundational understanding" + "consistent homework" + "growing peer collaboration." Each attribute maps to a pool of compatible templates.
3. Assembly: The engine picks one template per attribute, places them in sandwich order (positive opener → observation → forward action), substitutes the tokens with values from the SIS profile (student name, pronouns from /features/students/) and from the marks entry (subject name, grade tier).
4. Edit: The teacher reads the Smart Draft, edits anything that doesn't reflect what she actually observed, and adds the one specific anecdote the engine couldn't have known.

An optional final step — LLM grammar polish — is available, opt-in per teacher, off by default. We don't call this AI because the engine never generates sentences from scratch; it composes from a curated library. That distinction matters because parents and inspectors increasingly ask whether teacher remarks were written by a machine. The honest answer for SchoolDeck is: the structure was assembled from human-written templates, then the teacher edited it.

The sandwich method — positive, observation, forward

Even the best teachers struggle with feedback for a student who is genuinely behind. Pure praise misleads the parent; pure criticism demotivates the child. Every reputable Indian education editor recommends the sandwich pattern: a positive observation first, the constructive note in the middle, a forward-looking action at the end.

SchoolDeck's engine assembles drafts in this order by default:

• Opener: Names a strength the teacher actually selected. "[Name] is an energetic presence during [Subject] discussions" — never empty flattery.
• Observation: States the specific area for growth. "Foundational understanding of [Topic] would benefit from further consolidation" — clear but not harsh.
• Forward: Offers a path the student and parent can act on. "With consistent guided practice in the next term, [he/she] will see steady progress."

The teacher can override the structure — some prefer to lead with the observation and end with the strength. The engine just gives a sensible default, never a forced one.

Variables, pronouns, and the end of copy-paste embarrassment

The single most common report-card embarrassment in Indian schools is a pronoun error introduced by copy-pasting between students. "He has been a delightful student — Anjali." Once printed and shared with the parent, the moment is impossible to recover.

The narration engine eliminates this category of error entirely. Templates carry token placeholders — [Name], [He/She], [his/her], [him/her], [Subject], [Topic], [Medium]. At the moment of generation, the engine reads the student's profile from /features/students/ and the marks entry context, and substitutes every token. The teacher never sees a Word-document with a wrong-pronoun sentence. The published report card is mathematically incapable of carrying the error.

Other variables the engine uses: term name (Term 1 / Term 2 / Annual), grade tier from the school's grading scale, subject grade (B+ / A2 / 8), club or sport the student participated in this term, and any behaviour tags (perseverance, leadership, art initiative) logged during the term by subject teachers. None of these are inferred — they all come from real records the school already keeps in /features/examinations/ and /features/staff-attendance/.

NEP 2020 HPC + PARAKH — every domain has its own template library

The NEP 2020 Holistic Progress Card requires schools to narrate three different stories about every child each term — academic, co-scholastic, and socio-emotional. The PARAKH framework (NCERT) breaks the socio-emotional story into five specific domains. That is potentially eight to ten paragraphs per child per term.

SchoolDeck's library is structured by these domains, not just by subject:

• Academic (per subject): Mastery / Proficient / Developing / Foundational-Support tier templates per subject — Mathematics, Science, English, Hindi, Social Studies, Sanskrit, regional language, third language.
• Co-scholastic per NCF-SE 2023: Physical Education, Visual Arts, Performing Arts (Music + Dance + Drama), Work Education, Health and Wellbeing.
• Socio-emotional per PARAKH: Self-awareness, social awareness, self-management, relationship skills, responsible decision-making.

The class teacher's overall paragraph at the end of the report card pulls together threads from each domain — a structured holistic narration rather than a 200-word ramble.

School-specific tone and the banned-word list

An IB-curriculum school speaks differently from a State Board school. The tone of a residential boarding school's report cards is more formal than a day school. Forcing every school onto an off-the-shelf comment library produces remarks that feel synthetic.

SchoolDeck supports two layers of customisation:

• Custom Institutional Comment Bank: The Academic Council uploads the school's own pre-approved phrasing — sometimes a literal phrasebook the senior English HOD curated over years. These templates appear alongside (or instead of) the standard library.
• Banned-word list: Common entries — "fails", "lazy", "naughty", "weak", "careless", "stupid". The engine filters every template against this list. Any template containing a banned word is hidden from the teacher's selection altogether.

The result: a Class 12 English teacher, a Class 4 craft teacher and a Class 6 mathematics teacher across the school all sound like they work for the same institution — because in narration tone, they actually do.

Draft → teacher edit → principal review → lock → parent app

Narration sits in a clear workflow with audit trail at every step.

1. Draft: Teacher selects attributes, engine assembles, teacher edits. Saved as draft in /features/audit-logs/.
2. Teacher save: Teacher marks the remark complete. It enters the principal review queue.
3. Principal review: Principal sees all remarks for a class together — can accept, edit, or send back to the teacher with a comment ("please add a specific example of his improvement in Term 2").
4. HPC lock: Once all remarks for the class are reviewed, the principal locks the entire class's HPC. This triggers the report card PDF generation via /solutions/report-cards-academic-records/.
5. Parent app release: The PDFs release to the parent app for the entire class simultaneously — typically as one "Term Results Now Available" notification. No parent sees results earlier than another.

Narration ≠ Examination ≠ Report Card PDF generation

The SchoolDeck report-card cluster has three feature/solution pages. Knowing the boundary helps schools evaluate them correctly.

• This page · /features/report-card-narration/ — Owns the teacher remark text. Templates, sandwich method, NEP HPC + PARAKH domains, school tone, principal review of remarks. The narrative content per student per subject per domain.
• /features/examinations/ — Owns the exam scheme and grading engine. CBSE 80+20 split (Periodic Test best-of-3 + Notebook + Subject Enrichment), Class 10 Two Board Exams from 2026, 9-point grading, FA/SA traditional pattern. The numbers and the calculation logic.
• /solutions/report-cards-academic-records/ — Owns the final report card PDF layout. NEP HPC card design, CBSE / ICSE / State Board templates, school logo, signatures, parent app delivery. The deliverable artefact and the buyer outcome story.
• /features/student-performance-ai/ — Owns at-risk student early-warning. Identifies students whose performance trajectory predicts they may fall behind, using attendance + assignment + assessment trend data. Distinct from narration — predictive analytics, not text generation.

Each page targets a distinct query intent. This page is for schools that have already decided they need report card automation and want to understand how the remark text is produced. The solutions page is for the buyer earlier in the journey. The examinations page is for the parallel concern about grading.

Hand-typing remarks vs SchoolDeck narration

Practical differences for a class teacher with 40 students under NEP HPC requirements.