SNPP — how to show a 1995 protocol to a 2026 jury
RFC 1861 (October 1995) describes Simple Network Paging Protocol level 3: a two-way transaction with pre-seeded multiple-choice replies, a read acknowledgement, an expiry and a message tag. That is, almost word for word, the feature list of a modern "human in the loop" API — thirty-one years earlier.
This document is not the transcript. The canonical, byte-for-byte transcript is owned by the gateway package and replayed by its test:
packages/snpp/test/fixtures/session.txt— the transcript, one line per frame,S:for the server,C:for the client.packages/snpp/test/session.test.ts— replays it against an in-process TCP pair and fails on a single differing byte..kiro/steering/rfc1861.md— the commands and reply codes we implement, extracted from the RFC.
If this page and the fixture ever disagree, the fixture wins and this page is the bug. What follows is only the shooting script: what to type, in which order, and what to say while it happens.
The 25 seconds that carry the point
Open two windows side by side: a terminal on the left, the pager on the right.
nc 127.0.0.1 4440
Then type the sequence below. Type slowly — the pager needs a beat to light up, and the jury needs to read the reply codes.
| What you type | What appears | What you say |
|---|---|---|
2WAY |
250 OK, Beginning 2-Way Transaction |
"A two-way transaction. 1995." |
PAGE lea |
850 Two-Way Unit Online and Available |
"The pager is online. 950 would have meant queued." |
ACKR 1 |
250 Read Acknowledgment Enabled |
"I want to know when it was read." |
MESS 310 fw PUSH main>origin 3 commits +120-14 #07c2 |
250 Message OK |
"One line. That is the whole page." |
RTYP MULTICHOICE |
250 Reply Type Accepted |
"Multiple choice — seeded by the sender." |
MCRE 0001 Approve |
250 Response Added to Transaction |
"Option A." |
MCRE 0002 Deny |
250 Response Added to Transaction |
"Option B." |
SEND |
861 4821 7730 Delivered, Awaiting Reply (MCR) |
"Fired. The tag is 4821." |
| (tap ✔ on the pager) | — | "The human approves." |
MSTA 4821 7730 |
888 3 … 0001 MCR Reply Received |
"And the reply comes back as a code, not as a chat message." |
Close with KTAG 4821 7730 then QUIT, and let the 421 closing line land.
The unit side — a pager of 1995, in a terminal
bip-pager is the other half of RFC 1861: a unit, the device that
receives. It connects on the same TCP port as the terminals, registers itself
with its capcode, and draws what it receives on a 4 × 20 LCD in the terminal.
The keys answer; the answer travels back as an MCResponse.
Two windows, side by side — the terminal of 1995 on the left, the pager on the right:
# left — the terminal
nc 127.0.0.1 4440
# right — the pager
bip-pager --host 127.0.0.1 --port 4440 --capcode lea \
--central http://127.0.0.1:8787 --code ABC123
The terminal types 2WAY, PAGE lea, MESS …, RTYP MULTICHOICE,
MCRE 0001 Approve, MCRE 0002 Deny, SEND — and the pager lights up:
U: UNIT lea
G: 250 OK, Unit Registered
G: PAGE pg_7f3a9c 3f786a68…
G: MESS 310 fw PUSH main>origin 3 commits +120-14 #07c2
G: MCRE 0001 Approve
G: MCRE 0002 Deny
G: SEND 4821 7730
The human presses a (or b) on the pager: the answer goes back as
MCR 0001, and the terminal reads 888 … MCR Reply Received. The pager also
signs the decision at the Central with its ECDSA P-256 device key, so the
audit trail, the attention budget and the web PWA all see the same verdict.
| Key | Effect |
|---|---|
a b c d (or 1…4) |
answer A, B, C, D → MCR 0001…MCR 0004 |
q |
hang up → QUIT |
Without --code the emulator still receives and answers; it simply cannot
sign a decision at the Central, and says so on screen.
The wire format of the channel — the commands, the frames and the rules — is
written down in .kiro/steering/rfc1861.md
and in packages/snpp/docs/SNPP.md.
Why this is the strongest 25 seconds of the video
Every notification product today sends a paragraph and expects a conversation. This protocol, designed for a device with four rows of twenty characters, already had: a bounded message, a closed set of replies, a read receipt, an expiry and a correlation tag. The pager did not "inspire" Beeper — Beeper is that protocol, with a JSON API behind it. Showing a jury a live TCP session is the difference between claiming a lineage and demonstrating it.
What not to do
- Do not demo it through a wrapper. A
curlto a REST endpoint proves nothing; the point is that the wire speaks 1995. - Do not skip
MSTA. The reply code is the punchline: a decision that comes back as888 … 0001. - Do not fake it with a pre-recorded terminal. Record the real
nc; if it breaks, keep the break — it is more convincing than a perfect take. - Do not expose the port to make the demo easier. It binds
127.0.0.1:4440andLOGInis required outside the loopback; a demo is not a reason to open a plaintext protocol to the Internet.