segments-ref/ - canonical byte + disassembly layer (bytes exist ONCE here)¶

Single source of truth for the segment code behind the MON-call deliverable. Per-call folders (../mon-analysis/<call>/) do not embed .bin slices - they reference the locations here.

Each <seg>/ holds four files, all regenerated from the canonical .bin by ../../../../make-segment-ref.py (a full L rebuild is ~17 s): - <seg>.asm - whole-segment disassembly, load base applied, in-range symbols inserted as >>> NAME(TABLE) marker lines. A symbol marker does NOT prove a routine entry (see below). - <seg>.hex - one word/line: addr(oct) word(oct) hi lo(oct) byteoff(dec). - <seg>.symbols.txt - in-range symbols with their source table. - <seg>.meta.md - load base, word count, sha256, canonical .bin path.

69 of the 79 carved L segments have a bundle here. The 10 without one (CFT, CCT, SNA3270, COSPOOL, XFTRAD, FSASG, SEG131/133/140/141) have load_address: null in their .meta.json - the carver never resolved a load base, so they cannot be disassembled yet. They are COSMOS/application segments, not SINTRAN kernel, and host no MON worker.

What is committed to git (the rest is regenerated)¶

Ground truth = ../../segments/*.bin (committed). To avoid ~140 MB of regenerable bloat: - .hex is NOT committed - it is only a formatted view of bytes already in the .bin, and needs no special tooling to regenerate. - .asm is committed only for the 8-segment MON core (below) - .asm is the one artifact a visitor cannot reproduce without nd100-dis (not in this repo), and these 8 back every MON call. The other ~61 .asm are one make-segment-ref.py command away. - .symbols.txt + .meta.md are committed for all 69 (small, browsable).

See ../../../.gitignore.

The 8-segment MON core¶

The entire MON deliverable - all 216 implemented calls plus the whole dispatch chain - is backed by these eight segments:

Segment	Load	Role in the MON path
044-S3IDPIT	4000B	`MCTAB` / `9MCTA` @ 005620B - the real monitor-call table (216 workers)
026-S3IMPIT	32000B	dispatch: `ENT14 @072167B` + `GOTAB @071233B` (`MGOTA`) + `MFELL @072114B`; also ND-500 level-12 handlers 500..515
003-S3CP	30000B	78 MON workers (SYMBOL-1-LIST): `UECOM`/`COMSB`/`UELOG`, etc.
006-S3FS	26000B	72 MON workers (FILSYS): `RDISK`/`WDISK`/`OPFIL`/`MAGTP`/`CPUST`, etc.
025-S3IRPIT	32000B	59 MON workers (SYMBOL-2-LIST)
030-S3SM5	40000B	4 workers (N500-SYMBOLS) + ND-500 monitor 410/411/416/417
004-S3RTL	30000B	3 RT-loader workers
SINTRAN-DATA_commoncode	0	resident common code + data (NOTE: `071233B` here is NOT the GOTAB)

The MON dispatch model (byte-verified 2026-07-13)¶

MON N -> ENT14 072167B      level-14 entry (in 026-S3IMPIT / 017-S3SMPIT, NOT commoncode)
      -> X := MEM[MGOTA+N]  MGOTA = 071233B
      -> JMP ,X (072260B)   a DIRECT JUMP - no call, no "CALLPROC bridge"
      -> GOTAB[N] : 32/256 slots = resident fast handlers (incl. MON 200B XMSG)
                   224/256 slots = MFELL 072114B
      -> MFELL   : IRW 20 DP := CALLP 032201B ; MST PID/PIE   = a program-LEVEL switch
      -> MCTAB[N]  (005620B + N, in 044-S3IDPIT)
      -> worker

A MON call's worker address comes from MCTAB[N], never from GOTAB[N] (which is MFELL for 224 of 256 calls) and never from a symbol-name guess. Full derivation: ../mon-analysis/317B-ExecuteCommand/README.md; overview: SINTRAN/CARVING-HANDOFF.md section 3a.

Do not read GOTAB out of SINTRAN-DATA_commoncode. Its 071233B is not the GOTAB (slot 0 = 000000, slot 1 = 120303B; the real values are MFELL 072114B and M1 071633B). The real GOTAB is in the monitor-PIT segments, where ENT14 also lives.

Byte-offset contract (bin offset from an octal address)¶

word_index  = (octal_addr - load_base)      # subtract in OCTAL
byte_offset = word_index * 2                 # decimal, big-endian hi byte first
Read N words: dd if=<canonical.bin> bs=1 skip=<byte_offset> count=<N*2> | od -An -tx1
Or: grep '^<addr>  ' <seg>.hex  and read the byteoff column.
ND-500 (030-S3SM5): 32-bit units - use the .hex byte offset directly, NOT the *2 rule.

Validation stack (no trust required)¶

Byte-identity - <seg>.asm decodes the whole canonical .bin; sha256 in .meta.md.
Table anchors - a table is only "the" table if known slots match: GOTAB slot0=MFELL 072114B, slot1=M1 071633B, slot2=M2 071635B; MCTAB 005B->RDISK 102021B, 200B->XMSG 007516B, 144B->MAGTP 026354B.
Entry test caveat - "symbol lands on a 021xxx STD I prologue" holds only for JPL-called routines; dispatch-reached handlers save no link (UECOM 050701B starts 146141 RADD CLD SL DD).
Overlay check - the same virtual address decodes in many segments; pick the segment where 2-3 sibling symbols all land coherently, not one that merely "disassembles to something".
Second disassembler - Ghidra (Raw Binary, ND-100 big-endian 16-bit, base = load column) agrees.