# Education and admissions

Admissions traffic is seasonal, repetitive and multilingual, which is the combination Connect handles best: the same prospectus answer, in the family's language, at two in the morning in the week the forms open. Admissions decisions are the opposite kind of work, and nothing here should be routed through software — eligibility, offers, fees and appeals stay with people.

- **Status:** Available
- **Audience:** both
- **Channels:** email, phone, whatsapp
- **In the app:** #/inbox, #/knowledge, #/autonomy-audit
- **Last verified:** 2026-09-10
- **Canonical:** https://connectbyjbrh.com/docs/use-cases/education-and-admissions/

## The season, and what happens when it arrives

The failure mode a school should ask about is not slowness, it is loss. When the daily allowance is spent, work is **held rather than dropped** and the read cursor deliberately does not advance — advancing it on a refusal is how a mail agent loses messages permanently. Needs You shows the refusal, so a week that overruns the plan looks like a queue rather than a silence.

1. Load the prospectus, the fee schedule, the open-day dates and the entry criteria as Knowledge before the season, not during it.
   - Result: Answers are grounded in your own documents, and anything not in them is refused rather than improvised.
2. Set autonomy per channel for the season — `autonomous` on general enquiries, `ask_before_send` on anything that reaches the admissions mailbox.
   - Result: Volume is absorbed where it is safe to absorb and reviewed where it is not. The narrowest scope wins, so one applicant family can be handled by a person without changing the channel.
3. Watch the allowance on the Plan and Usage screen in the first week.
   - Result: You learn the real shape of your season's volume against the plan, with the refusals visible rather than inferred from a gap.
4. Review the decision log at the end of the first fortnight.
   - Result: You can read what was said to whom, under which rule — including refusals, because a refusal is a decision and is recorded as one.

## Families who do not write in one language

Language handling is not a translation layer bolted on. On calls there is language detection, a known-language setting for a returning caller, a regional speaking-style layer, and handling for code-mixing — the ordinary case where a parent moves between two languages inside one sentence. In writing, the reply is composed in the language the enquiry arrived in.

- A returning caller whose language is already known does not have to be identified again, because it is on the relationship rather than guessed each time.
- The same question in two languages is answered from the same Knowledge, so the two answers agree in substance. That is a property of grounding, not a claim about translation quality.
- Anything Knowledge does not support is refused in whichever language it was asked in. The boundary does not move with the language.

## The decisions that must not be automated

**Eligibility and offers** — Whether a child meets criteria, and whether a place is offered, are judgements with consequences and an appeals process attached. Connect can say what the published criteria are; it cannot apply them to a named applicant, and the autonomy setting on the admissions endpoint should make that structurally impossible rather than merely discouraged.
**Fees, bursaries and discounts** — `safe_sales` refuses commercial terms Knowledge does not support. A published fee is a fact and can be quoted; a bursary that depends on circumstances is not, and improvising one is a promise the school then has to honour or retract.
**Anything about a specific child** — Attendance, behaviour, progress, safeguarding. These are not enquiry traffic and should not arrive on an automated channel; where they do, route them to a person by scoping the contact.
**Ranking or comparing applicants** — Not a capability, not a claim, and not something to build with Knowledge and a clever instruction. There is no defensible audit story for it.

> **Careful** Consistency is not fairness. Grounding in one set of documents makes the same question get the same answer, and the decision log makes what was said reviewable. Neither of those is a guarantee about outcomes, and no certification is held for any of this.

## Channel realities for a school

| Channel | Status | The catch |
|---|---|---|
| Email | Available | The workhorse. Gmail, Microsoft Graph or IMAP/SMTP; bounded by the plan's daily allowance. |
| Phone | Available | A 3.3 s median reply floor on the best measured call, no completed transfer, no recording on the live carrier. |
| WhatsApp | Available | How many families actually prefer to write. Templates and session windows are provider-dependent. |
| SMS | Foundation | The live carrier carries none. In India, DLT entity, header and template registration with the operators is required before any of it. |

## Where the money goes

Written channels, in volume, and calls, in intensity. A thousand emails in an admissions week is one model call each and sits inside a plan's allowance as a predictable number. Two hundred phone conversations is audio tokens at four times the text input rate with the session re-billed for its context every turn, and is the line item worth watching. Moving the repeated questions — dates, criteria, uniform, transport — into published Knowledge answers reduces both, because a family who gets a complete written answer does not ring.

## Questions

### What happens if we get more enquiries than our plan allows?

The excess is held rather than dropped, and the mail read cursor does not advance past unprocessed messages — that is deliberate, because advancing on a refusal is how messages are lost for good. The refusals appear in Needs You. You see a queue and a decision to make, not a gap you find out about in September.

### Can it tell a parent whether their child will get a place?

No, and it should be configured so that the question never reaches an automated reply. Published criteria can be quoted from Knowledge; applying them to a named child is a decision with an appeals process behind it. Set the admissions endpoint to `ask_before_send` so a person sees anything arriving there.

### Can we send admissions reminders by SMS?

Not on the live carrier, which carries no SMS. In India there is a second requirement regardless of carrier: DLT registration with the telecom operators, covering the entity, the sender header and each template. That is a process with the operators rather than a setting anybody can switch on here.

## Related

- [Serving customers in several languages](https://connectbyjbrh.com/docs/use-cases/multilingual-customers/)
- [Knowledge sources](https://connectbyjbrh.com/docs/knowledge/sources/)
- [The daily email allowance](https://connectbyjbrh.com/docs/email/daily-allowance/)
- [The decision log](https://connectbyjbrh.com/docs/autonomy/audit-trail/)
- [Detecting the caller's language](https://connectbyjbrh.com/docs/phone/language-detection/)
- [DLT registration in India](https://connectbyjbrh.com/docs/sms/dlt/)

## What this page is based on

- docs-source/sources/CHANNELS.md §1 — the allowance and the cursor
- docs-source/sources/GENERAL.md §5 — autonomy scopes and the audit trail
- docs-source/sources/PHONE.md §12 — the latency floor and carrier limits
- Connect capability registry (docs-source/facts.py)