Octobus MAILBOX / MICFU message sequence - annotated live reference¶

What this is: the message-by-message record of what SINTRAN actually sends over the octobus MAILBOX (MICFU) layer during ND-500 bring-up, with each message explained, captured live from RetroCore on 2026-07-28.

Why it exists: the two existing catalogs cover DIFFERENT layers and neither answers "what does message N mean and what should we have replied":

File	Layer
`SINTRAN-OCTOBUS-MESSAGE-CATALOG.md`	ND-100 side, octobus FRAME / OMD (LMFIELD, kicks, CON5IDENT, 5OMBREAD)
`ND5800-MICROCODE-ACCP-OCTOBUS-CATALOG.md`	microcode side, B30 routines (AFLAG, polling spine, TRAP_OCBM)
this file	mailbox MICFU layer (3RMICV / CACHE / PHYSWR / PHYSRD) with operands

Without this, the raw trace is easy to misread. It was misread twice on 2026-07-28 - see "Corrections" at the end. Read this before drawing conclusions from a MICFU trace.

How to capture it¶

dotnet test Emulated.Tests --no-build --filter "FullyQualifiedName~Nd100SintranNd5000OctobusBootHarnessTests.FullFlow"

Look for ----- servicer MICFU trace (octobus-fullflow) -----. The harness logs the copy-family operands (addrA, addrB, nrbyt) via OnServicerMessage; without those the trace is not interpretable. Machine state at capture: run thread ON, SINTRAN fully initialised (N5CPU=1, NSAMSON=4, 5MBBANK=41B, 5FPMAILBOX=0x851).

The captured sequence: 56 messages¶

Count	MICFU	Name	Role
30	`0x01`	`3RMICV`	WATCHDOG ping (not a version query - see below)
1	`0x0A`	`CACHE` (12B)	cache-clear, `nrbyt=2048` (one page)
13	`0x19`	`PHYSWR` (31B)	physical write, `nrbyt=4` (one 32-bit word) each
12	`0x18`	`PHYSRD` (30B)	physical read, `nrbyt=4` each

Phase 1 - probe (msgs 1-3)¶

1  3RMICV @0x00428E30
2  3RMICV @0x0042C130
3  3RMICV @0x0042C130

Two distinct message buffers appear and stay fixed for the whole run: 0x00428E30 = the general ND-500 message buffer, 0x0042C130 = the WATCHDOG buffer.

Phase 2 - cache clear (msg 4)¶

4  CACHE addrA=0xCFCF4000 addrB=0x00216400 nrbyt=2048

addrA is not a plausible address - for CACHE these operand slots are almost certainly not address operands at all. UNVERIFIED: what CACHE's operand fields mean.

Phase 3 - WRITE a 13-word block (msgs 5-27, interleaved with the watchdog)¶

PHYSWR addrA=0x000000BC addrB=0x0000CC00 nrbyt=4
PHYSWR addrA=0x000000C0 ...
PHYSWR addrA=0x000000C4
PHYSWR addrA=0x000000B6
PHYSWR addrA=0x00000096
PHYSWR addrA=0x0000009A
PHYSWR addrA=0x0000009E
PHYSWR addrA=0x000000A2
PHYSWR addrA=0x000000A6
PHYSWR addrA=0x000000AA
PHYSWR addrA=0x000000AE
PHYSWR addrA=0x000000B2

Every write is FOUR BYTES from the SAME source buffer addrB=0x0000CC00. Targets step by 4 across 0x96..0xC4 - a contiguous block of 32-bit words, written in the order BC C0 C4 B6 96 9A 9E A2 A6 AA AE B2.

Phase 4 - READ THE SAME WORDS BACK (msgs 29+)¶

PHYSRD addrA=0x000000BC addrB=0x0000D400 nrbyt=4
PHYSRD addrA=0x000000C0 ...
PHYSRD addrA=0x000000C4
PHYSRD addrA=0x000000B6
PHYSRD addrA=0x00000096
PHYSRD addrA=0x0000009A
...

Same addresses, SAME ORDER, into a different buffer 0x0000D400. This is a write-then-read-back VERIFY of a 13-word block, not a data transfer.

Phase 5 - nothing but watchdog¶

After the read-back the trace contains ONLY 3RMICV pings. SINTRAN issues no further work.

What each MICFU means here¶

3RMICV (0x01) is the WATCHDOG, not a version query. MP-P2-N500.NPL:1209 (LCLTSB) stamps MICFU=3RMICV into the WATCHDOG buffer and arms a timer (LTTMR=:TMR). The timeout check is RP-P2-N500.NPL:127642: when the timer fires, if the watchdog message's N5STA is not ANSWER, SINTRAN checks X5BRK and otherwise raises N5TIMOUT ("ACCP was terminated; Microprogram has stopped") and calls RSTARTALL. Do not read a burst of 3RMICV as a version negotiation - it is a heartbeat, and its count is just elapsed time.
PHYSWR/PHYSRD are the copy family. Nd500MicrocodeServicer.PerformOctobusBlockCopy: addrA = ND-500/target side (SC3), addrB = buffer side (SC7), nrbyt = byte count (SC4), both resolved as host.Nd500AddressBase + addr. WR copies B->A, RD copies A->B.

Emulator behaviour on this sequence (2026-07-28)¶

Every message IS serviced and answered - N5STA transitions 1 (MSGN5) -> 2 (WAITING) -> 3 (ANSWER) are observed live, and 914 station->ND-100 frames were read by SINTRAN. The sequence completes. Then SINTRAN waits forever and the ND-5000 stays PC=0 stopMode=WAIT: no activation (3START, MICFU 0x13) is ever sent, so the swapper is never started.

[STALE - CORRECTED 2026-09-01, ND5000UC standoff 189.] The half of this sentence about 3START is NO LONGER TRUE on the octobus lane. Decoding a place-domain run in order (not counting it) gives:
CACHE(12B) -> PHYSWR(31B) x13 over 0x96..0xC4 -> 3START(23B) -> 3RMICV -> 3MONCO(24B)
-> 3RMICV x~108 to the end of the run
So 3START IS sent, immediately after the 13-word block below, the swapper runs, and it takes one monitor call - genuinely forwarded, not canned (posted=2 seen=1 taken=1, so the Seen > Taken gap is absent). What survives unchanged is the "waits forever with nothing but watchdogs" half.

Why this correction is a banner rather than an edit: the true half and the false half sit in ONE sentence, so the accurate "watchdogs forever" lends its credibility to the inaccurate "never sends 3START". A reader today would hunt for a missing 3START, find it present, and lose time deciding which half to disbelieve. The observation was correct when written; the lane moved.

Also answered below: 0x96..0xC4 is written immediately before 3START - it is the start-swapper verify block, and the 13 writes complete and pass.

Current state: the swapper starts, runs, parks after one monitor call, and a 3SWMESS node sits unconsumed at SWPPING(6) while 5ACTSWAPPER has run twice. See standoff sections 185-189 in ND5000UC: docs/OCTOBUS-SWAPPER-STANDOFF-2026-08-28.md

OPEN - the next question, and do NOT guess it¶

What is ND-500 physical 0x96..0xC4? 13 consecutive 32-bit words that SINTRAN writes and immediately verifies during start-swapper. Identifying this block is the next step, because whatever SINTRAN expects to happen after a successful verify is not happening.

Also UNVERIFIED: the meaning of CACHE's operand fields; whether addrA on the copy family should resolve to ND-500 LOCAL memory rather than through Nd500AddressBase into the MPM window (a self-consistent round-trip would hide the difference, so the passing verify does NOT prove the target is right).

Corrections - misreadings this file exists to prevent¶

"Only 4 messages were serviced." WRONG - an artifact of reading a truncated trace view. 56 messages were serviced. Count the real trace section, do not eyeball a window.
"The PHYSWR addresses cycle - it is a livelock." WRONG - they ADVANCE by 4 across 0x96..0xC4. A 16-line window straddled the write and read-back phases and looked like a loop. The only repetition is the watchdog.
"MicroVersion=0x2E9A is being rejected." Not supported by any evidence here - 3RMICV is the watchdog and its payload is not implicated.

Raw ordered capture: regenerate with the command above; the run used for this file is run_ops.txt in the session scratchpad (transient - re-capture rather than relying on it).