Scaling From One District to Thirty

District A had a collector who could bully a queue, a DIO who liked Linux, and a scheme clerk who wrote good Odia. The vendor filmed it. The cabinet asked for thirty districts by December. Nobody asked whether district A's village codes, connectivity, or champion would travel. Twenty-nine districts received a login and a WhatsApp group. Three used it. The film was still shown.

Scaling a government agent is copying a procedure, not cloning a vibe. The unit of scale is the packet: standing order, purpose tags, templates, exception duty, SDC parent, sampled error rate. If you cannot photocopy that packet, you cannot photocopy the service.

This playbook is for SIs and state programme units. It is dated 17 August 2026. It is not legal advice. One-to-thirty is a metaphor. Your state may have more districts. The method does not care.

Freeze before you multiply

Freeze the workflow identity. Same scheme, same human gate, same 'agent shall not deny' sentence. If district B wants the agent to write to the register, that is a new file, not a scale-up.

Freeze the parent. Inference in the SDC or a documented hub. District cupboards do not scale. They breed thirty egress stories.

Freeze the language templates, then allow a local annex. District B may add a refusal line. It may not invent English error strings.

Freeze the champion design. Every district names a post and a deputy before it gets a login. No named deputy, no login. This will feel harsh. It is how you avoid twenty-nine green dashboards.

What you copy to district N, and what you must not copy.

Object	Copy	Do not copy
Standing order	Yes, with district name filled	A WhatsApp forward of district A's chat
Model weights	Usually the same pinned hash	A local fine-tune 'because our English is different'
Retrieval corpus	State circulars plus district annex	Whatever PDFs a clerk uploaded at 1am
Exception KPI	Yes, same definition	A local right to hide the queue
Write credentials	Still none	A collector's personal login 'just here'

Cohorts, not a flood

Move in cohorts of five or six, mixed by connectivity and language, not the thirty friendliest collectors. Each cohort has a two-week dry run with the link pulled for a day, a sample of fifty files, and a transfer tabletop. Fail a cohort in public inside the programme unit. Do not fail it in a newspaper.

NIC district officers are hosts and translators, not unpaid SIs. Send them the one-page ask. Give them the runbook. Do not surprise them with a collector's order to 'enable AI tomorrow'.

Two rollouts after the same cabinet target

Objections you will hear — and what to do with them

The cabinet number is thirty

Give them twenty-four real and six honest delays, or give them thirty films. You will not give them both. Write the number you can reconstruct.

Districts will revolt if they wait

Give waiting districts the paper procedure and a date. A date is respect. A login without a deputy is flattery.

We will use the first district as the model for everyone

Only the packet. Not the collector's personality, not the coastal connectivity, not the language mix.

Central hosting will not feel local

Local is templates, circulars and the deputy. Hosting is a parent. Confusing the two is how you put GPUs in cupboards.

A four-week playbook to leave district one

1. Week 1: write the photocopiable packet. If a paragraph needs district A's hero, rewrite it.
2. Week 2: pick cohort one — mixed difficulty, not friends. Name deputies.
3. Week 3: dry run, pulled link, fifty files, transfer tabletop.
4. Week 4: go-live only the districts that passed. Tell the cabinet the true count. Schedule cohort two.

File note you can paste

Subject: Conditions for extending the agentic workflow beyond the pilot district.

Extension to further districts requires a photocopiable packet (standing order, purpose tags, templates, exception KPI, SDC parent), a named deputy in each new district, a passed dry run including a pulled-link day, and no new write path. Model weights remain pinned unless the governance body approves a change. Public translation APIs remain forbidden. Cabinet numbers will report live, reconstructable districts, not issued logins.

This note is an internal aid. It is not legal advice.

Prcept AI will not film district A as if it were the state. If you want thirty, bring thirty deputies. We will bring one runtime you can see from the SDC. That is the only scale we know how to defend.

What the weekly count should look like

Report four numbers, not one. Districts with a named deputy. Districts that passed the pulled-link test. Districts whose exception queue was opened this week. Districts whose packet export worked without the original champion. Logins issued is not a number you should show a cabinet. It is a vanity metric with an incident attached.

When a district fails two weeks running on the queue, pull the login. That will feel political. It is how you stop a green map from lying. Offer the paper procedure and a date to return. A pulled login is cheaper than a statewide pause after a newspaper.

Keep cohort notes short enough that a new SI can read last month. Scale programmes die when the knowledge is in one WhatsApp. The packet was the product in district one. It is still the product in district thirty.

This article is informational field guidance for Indian public institutions, not legal, programme-management or procurement advice. Confirm against your state's district manuals, DPDP, MCC if a poll is on, and counsel before you file it.

How to sequence this in a state, not a slide

“Scaling From One District to Thirty” is a department problem. A P4 System Integrator should name the legacy system, the officer who owns the file, and the citizen charter clock before buying “scaling government AI deployment”.

One clever district is not a state programme. Scale only what you can reconstruct: the packet, the language, the human gate, and the SDC parent. Thirty copies of a WhatsApp approval is thirty incidents. Do not invent league tables of states. Read tenders and policies. Election Model Code of Conduct can freeze a rollout. NIC is a partner, not a villain. SDC readiness is GPU, power, ops and egress — not a logo.

• Audit the legacy store first.
• Keep mutation and money as officer actions.
• Map SLAs to the citizen charter.
• Budget change requests after go-live.

Close this loop before the next CAB

Put “Scaling From One District to Thirty” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P4 System Integrator, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “scaling government AI deployment” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

What must be true before you file this

If “Scaling From One District to Thirty” is only a heading, it will not survive a file inspection. A P4 System Integrator should be able to attach one artefact that proves “scaling government AI deployment”: a log export, a clause, a scored row, a dated notice, or a refusal rule.

Write three dated sentences: what was decided, who owns it, and when it will be re-checked. Unsigned sentences are souvenirs. Dated sentences are controls.

• Name the owner of “scaling government AI deployment” inside the institution.
• Attach one artefact a stranger can open next year.
• Revisit when the model, the notice, or the SI changes.
• Do not treat a vendor slide as evidence.

One more artefact before you close the file

Add a one-page owner map: who runs this after the vendor leaves, who can stop it, and where the logs live. If those three names are missing, the project is still a demo.

Date the page. File it next to the contract. That is the difference between a blog you read and a control you can audit.

What the next file must contain

“Scaling From One District to Thirty” earns a line in the noting only if a P4 System Integrator can attach proof of “scaling government AI deployment.” A heading is not proof. A vendor slide is not proof. A workshop photograph is not proof.

Write three dated sentences: what was decided, who owns it after the next posting order, and when it will be re-checked. If you cannot write the three sentences, you are not ready to buy, to sell, or to go live.

Leave unsourced percentages out of the note. DPDP is not a blanket localisation statute. The November 2025 AI governance text is guidance, not an Act. CERT-In’s 28 April 2022 directions still set specified incident and log clocks. A PAC, when lawful, lives in GFR Rule 166.

• Name the designation that owns “scaling government AI deployment.”
• Attach one artefact a stranger can open next year.
• Record the instrument you are actually using.
• Revisit when the model, the SI, the notice or the posting changes.