Disc support in SINTRAN III — answered from CARVED SYSTEM DATA¶

Full path: E:\Dev\Ronny\NDInsight\tools\boot-floppy\CARVED-DISC-SUPPORT.md

Scope rule for this document: every statement is derived from carved system data — carved segment binaries, the resident common-code image, the SINTRAN image symbol tables, and the recovered NPL source listings. No OCR'd manual was consulted, and none is cited. Where the NPL source is used it is used for naming and comments only and is always cross-checked against carved bytes (the NPL tree is a different revision from the carved images).

Every claim is tagged [VERIFIED] (exact file + address/offset + the bytes or words quoted) or [INFERRED] (reasoning shown). Where nothing was found the document says NOT FOUND.

Companion tool written for this analysis: E:\Dev\Ronny\NDInsight\tools\boot-floppy\tools\find_disc_layout_table.py

0. Executive summary¶

Question	Answer
Is there a disc-type / geometry TABLE inside the kernel?	YES — two of them, and they are linked. (1) `DISPE` + the `DTxxx` geometry records; (2) the device-name table `S3DNAM`, which names every supported disc unit, gives its size in pages, and points at a `DTxxx` record. Both byte-verified in all three carved systems.
Where are the valid disc SIZES?	In the device-name table (§2A). Every named large-disc unit carries a hard-coded 32-bit page count. Every SCSI and optical entry carries size = 0 — those sizes are interrogated from the drive.
How many disc types does it hold?	24 records in L-VSX-500 and M-VSX-500; 18 records in K-VSX-500 (the ST-506/Winchester group is absent from that generated system). Index range 0..47B (`MAXDI=47`).
Is SCSI in the table?	YES, as disc-type index 36B, record `DTSSS`, with all geometry fields zero — SCSI geometry is not tabulated, it is interrogated from the drive.
Are SCSI disc SIZES validated against a table?	NO. Capacity is taken from the device (`READ CAPACITY`) and from the disc's own control record; it is a 32-bit value. The only kernel-enforced size limit on the SCSI path is on the block/record size, not on the capacity.
`MSTYP` numeric value	NOT FOUND in any carved symbol table (K03/L07/M06) or NPL listing. The open question is not resolved; see §7.

1. THE disc-type table — `DISPE` and the `DTxxx` records [VERIFIED]¶

1.1 Record layout¶

The layout is documented in the recovered source listing E:\Dev\Ronny\NDInsight\SINTRAN\NPL-SOURCE\s3vs-4.symb, listing address 040752 (lines 19568-19582), quoted literally:

%  DISK LAYOUT:
%      SECWO, SECTR, SECSY, MAXCY, POLSY, REFOR, RESCY, DISPN
%          SECWO = WORDS   / SECTOR.
%          SECTR = SECTORS / TRACK.
%          SECSY = SECTORS / CYLINDER.
%          MAXCY = VALUE OF MAX CYLINDER.
%          POLSY = VALUE OF FIRST CYLINDER IN POOL.
%          REFOR = FORMAT TYPE:
%                     0 = TRACK OR NO REALLOCATION.
%                    10 = TRACK OR NO REALLOCATION + TEST SECTOR ADDRESS
%                    20 = SECTOR REALLOCATION.
%          RESCY = VALUE OF FIRST RESERVED CYLINDER.
%          ALTFO = ALTERNATIVE FORMAT (ADDRESS OF DTxxx OR 0)
%          DISPN = INDEX FOR THIS ENTRY(FORMAT).

and DILEZ=11 (octal) — 9 words per record (the comment lists 8 names but ALTFO sits between RESCY and DISPN, giving 9). [VERIFIED]

1.2 The table AS CARVED¶

Carved from the raw big-endian binaries, not from any listing:

Version	file	load base	table start	`DISPE`	records
K-VSX-500	`versions\K-VSX-500\segments\044-S3IDPIT.bin` (twin `053-S3SDPIT.bin`)	`4000B`	`012337B`	`012267B`	18
L-VSX-500	`versions\L-VSX-500\segments\044-S3IDPIT.bin` (twin `053-S3SDPIT.bin`)	`4000B`	`031041B`	`031371B`	24
M-VSX-500	`versions\M-VSX-500\segments\044-S3IDPIT.bin` (twin `053-S3SDPIT.bin`)	`4000B`	`035756B`	`036306B`	24

Independent confirmation of the L and M addresses: the L07 symbol table SINTRAN\NPL-SOURCE\SYMBOLS\L07\SYMBOL-2-LIST.SYMB.TXT:337 says DISPE=031371 and :338 MAXDI=000047; M06 SYMBOL-2-LIST.SYMB.TXT:461 says DISPE=036306. The carved table of 24 records is 24 × 9 = 216 words = 330B long and starts at 031041B; 031041B + 330B = 031371B — exactly DISPE. Same arithmetic holds for M (035756B + 330B = 036306B). [VERIFIED]

The carved K image is a K05 system while the only K symbol table available is K03; K03 DISPE=012221 does not match the carved image (that address is zero-filled). The K figure 012267B above was derived structurally, from the carved pointer array itself (see §1.4). [VERIFIED by structure, INFERRED as to the name DISPE]

1.3 The carved records — L-VSX-500 (L07), addresses octal¶

Produced by tools\find_disc_layout_table.py 044-S3IDPIT.bin 4000 (add 4000B to the addresses it prints; it prints file-relative addresses when the meta base is not supplied). Values octal.

addr	SECWO	SECTR	SECSY	MAXCY	POLSY	REFOR	RESCY	ALTFO	DISPN	name (from listing)
031041	1000	14	44	577	1124	0	0	0	10	DT014 — DISC-14MB
031052	1000	14	44	1134	1124	0	0	0	11	DT021 — DISC-21MB
031063	1000	11	33	1475	1462	0	0	0	12	DT023 — DISC-23MB
031074	1000	11	66	1475	1462	0	0	0	13	DT045 — DISC-45MB
031105	1000	11	33	1775	1763	0	1776	0	14	DT028 — DISC-28MB
031116	1000	11	110	1775	1763	0	1776	0	15	DT074 — DISC-74MB
031127	1000	22	22	1466	1465	10	0	0	20	DT030 — DISC-30/60/90MB
031140	1000	22	132	632	0	10	0	0	21	DT037 — DISC-38MB
031151	1000	22	132	1466	1434	10	1465	0	22	DT070 — DISC-70MB
031162	1000	22	132	1466	1465	10	0	0	23	DT075 — DISC-75MB
031173	1000	22	264	1466	1441	10	1465	031204	24	DT140 — DISC-140MB-F
031204	1000	21	252	1466	1462	20	1465	031173	25	DT135 — DISC-140MB-S
031215	1000	22	264	1466	1465	10	0	0	26	DT160 — DISC-2-75MB
031226	1000	22	526	1466	1465	10	0	0	27	DT288 — DISC-288MB-R
031237	1000	21	630	1306	1303	20	1305	031250	30	DT285 — DISC-288MB-S
031250	1000	22	660	1306	1260	10	1305	031237	31	DT300 — DISC-288MB-F
031261	1000	32	1160	1306	1303	20	1305	031272	32	DT450 — DISC-450MB-S
031272	1000	33	1210	1306	1260	10	1305	031261	33	DT460 — DISC-450MB-F
031303	1000	32	404	2300	2267	20	2277	0	17	DT310 — DISC-288MB-E
031314	1000	37	1115	1367	1363	20	1366	0	34	DT470 — DISC-450MB-N
031325	1000	43	536	1466	1460	20	1465	0	35	DT290 — DISC-288MB-N
031336	1000	0	0	0	0	0	0	0	36	DTSSS — SCSI DISK
031347	1000	36	0	0	0	0	0	0	40	DTOD1 — OPTICAL DISK (RS=1024)
031360	2000	36	0	0	0	0	0	0	41	DTOD2 — OPTICAL DISK (RS=2048)

The 24 records are [VERIFIED] from the carved bytes. The DTxxx names and the % DISC-nnnMB-x comments are [VERIFIED] from the recovered source listing s3vs-4.symb lines 19583-19613, where the numeric rows match the carved rows word for word.

M-VSX-500 carries the identical 24 records at 035756B..036305B (spot check: 036253B = 1000 0 0 0 0 0 0 0 36 = DTSSS). [VERIFIED]

K-VSX-500 carries only 18 records at 012337B..012600B: DISPN 20,21,22,23,24,25,26,27,30,31,32,33,17,34,35,36,40,41. The six ST-506 / Winchester records (DISPN 10-15) are absent from the carved K05 system. [VERIFIED]

1.4 `DISPE` — the disc-type pointer array [VERIFIED]¶

L07, carved bytes at 031371B (044-S3IDPIT.bin, base 4000B):

 031371: 000000 000000 000000 000000 000000 000000 000000 000000   % types 00-07
 031401: 031041 031052 031063 031074 031105 031116 000000 031303   % types 10-17
 031411: 031127 031140 031151 031162 031173 031204 031215 031226   % types 20-27
 031421: 031237 031250 031261 031272 031314 031325 031336 000000   % types 30-37
 031431: 031347 031360 000000 000000 000000 000000 000000 000000   % types 40-47

M06 at 036306B is structurally identical (pointers 035756…036275).

K05 (carved), pointer array at 012267B:

 012267: 000000 x8                                                 % types 00-07
 012277: 000000 x7 012513                                          % types 10-17  (10-15 EMPTY, 17=DT310)
 012307: 012337 012350 012361 012372 012403 012414 012425 012436   % types 20-27
 012317: 012447 012460 012471 012502 012524 012535 012546 000000   % types 30-37
 012327: 012557 012570 000000 x6                                   % types 40-47

Reading: a disc type is supported by this generated system iff its DISPE entry is non-zero. MAXDI=47B is the maximum index. [VERIFIED]

The bound is enforced in code: IP-P2-DISK-START.NPL:52 (listing address 054024) reads

054024   IF 40>T AND DISPE(T)><0 THEN   % 37 IS MAX FORMAT NUMBER

for the SMD driver's fn 36 = READ ELEMENT IN "DISC-LAYOUT-TABLE" function, and IP-P2-SCSI-DISK.NPL:328 (057052) / IP-P2-SCSI-OPDI.NPL:196 (066064) use IF X<<="MAXDI" AND DISPE(X)><0. [VERIFIED from the recovered source; the table those lines index is byte-verified above.]

1.5 Correction to an earlier finding¶

A parallel pass over 006-S3FS concluded "no repeating disc-geometry table exists in SINTRAN L07". That conclusion is wrong: the table exists, but it lives in the DPIT segment (044-S3IDPIT / 053-S3SDPIT, load base 4000B), not in the file-system segment. The symbols SECSY, SECTR, etc. found in FILSYS-SYMBOLS are the field offsets into a record of this table (SECTR=1, SECSY=2, DISPN=10B — and indeed word 1 = SECTR, word 2 = SECSY, word 8 = DISPN in the carved records, exactly matching). [VERIFIED]

2A. The DEVICE-NAME table `S3DNAM` — the list of supported disc units and their sizes [VERIFIED]¶

This is the single most direct answer to "what are the valid disc sizes and supported disc units".

2A.1 Where it lives [VERIFIED]¶

version	file	load base	first entry
K-VSX-500	inside `segments\006-S3FS.bin` (twin `012-S3FSSV.bin`)	`26000B`	`DISC-38MB-1` at `114651B`
L-VSX-500	own segment `segments\101-S3DNAM.bin` (twin `100-S3SDNAM.bin`)	`164000B`	`DISC-38MB-1` at `164000B`
M-VSX-500	own segment `segments\101-S3DNAM.bin` (twin `100-S3SDNAM.bin`)	`164000B`	`DISC-38MB-1` at `164000B`

100-S3SDNAM.meta.json states "description": "Save of device name table". versions\L-VSX-500\inputs\list-segment.txt lists S3SDNAM 100 / S3DNAM 101; K's list-segment.txt has no such segments. The move of this table out of the file-system segment into its own segment pair is a K→L architectural change. [VERIFIED]

2A.2 Entry format — 17 words per entry, identical in K, L, M [VERIFIED]¶

Raw first entry, 101-S3DNAM.bin byte 0, address 164000B:

042111 051503 026463 034115 041055 030447 000000 000000
000000 044066 001000 002001 030346 000515 001100 000000 031140

word	value here	content
0-7	`042111…030447`	packed ASCII name `DISC-38MB-1` + `'` (047) terminator — [VERIFIED]
8:9	`000000 044066`	32-bit device size in PAGES = 18486 — [VERIFIED value, INFERRED unit]
10	`001000`	`1000B` disc, `400B` floppy, `1B` mag tape — [VERIFIED]
11	`002001`	unit / target selector (SCSI targets 1..8 = `2021`…`2027`,`2020`) — [VERIFIED]
12	`030346`	= symbol `BABST=030346` (`SYMBOLS\L07\FILSYS-SYMBOLS.SYMB.TXT`) — [VERIFIED]
13	`000515`	[NOT DECODED] - one observed value cannot establish a label; section 8 of this file lists words 12/13 as uncarved. To settle: tabulate word 13 across all 268 entries and show it partitions by device family, as was done for word 14 in section 2A.4
14	`001100`	LOGICAL DEVICE NUMBER — [VERIFIED, see 2A.4]
16	`031140`	pointer to the `DTxxx` geometry record — `031140B` = `DT037` in §1.3 — [VERIFIED]

Word 16 is the link to §1. Every value observed in the L07 table is one of the 24 record addresses byte-verified in §1.3 — 031041(DT014) … 031360 (DTOD2), and 031336 (DTSSS) for every one of the 112 SCSI entries. Two tables carved independently, and the pointers agree exactly. [VERIFIED]

2A.3 The valid disc sizes, decoded from the carved bytes [VERIFIED]¶

268 records parse as named entries in L07 (266 real + 2 false hits on misaligned data). Sizes are the 32-bit word 8:9, decimal pages; a SINTRAN page is 1024 words = 2048 bytes.

named family	pages	≈ bytes	`DTxxx` record(s) pointed at
`DISC-450MB-n-F` / `-N`	220584	451.8 MB	`031261` DT450, `031314` DT470
`DISC-288MB-n-R/F/E/N`	140391	287.5 MB	`031226` DT288, `031250` DT300, `031303` DT310, `031325` DT290
`DISC-225MB-n-R/F/E/N`	110292	225.9 MB	DT288 / DT300 / DT450 / DT310 / DT470 / DT290
`DISC-140MB-n-F`	69530	142.4 MB	`031204` DT135
`DISC-75MB-n`	36945	75.7 MB	`031162` DT075, `031226` DT288, `031215` DT160
`DISC-74MB-n`	36396	74.5 MB	`031116` DT074
`DISC-70MB-n(-R/F/E/N)`	34765	71.2 MB	`031151` DT070 + the -R/F/E/N variants
`DISC-45MB-n`	22032	45.1 MB	`031074` DT045
`DISC-38MB-n`	18486	37.9 MB	`031140` DT037
`DISC-28MB-n`	13648	28.0 MB	`031105` DT028
`DISC-23MB-n`	11016	22.6 MB	`031063` DT023
`DISC-21MB-n`	10728	22.0 MB	`031052` DT021
`DISC-16MB-n`	8000	16.4 MB	`031063` DT023
`DISC-30MB-n` / `-60MB-n` / `-90MB-n`	7389	15.1 MB	`031127` DT030
`DISC-14MB-n`	6912	14.2 MB	`031041` DT014
`FLOPPY-DISC-1/2`	154	0.3 MB	0
`DISC-[n-]SCSI-1…14` (112 entries)	0	—	`031336` DTSSS
`DISC-[n-]OPTICAL-1…4` (32 entries)	0	—	`031347` DTOD1
`MAG-TAPE-1…4`, `STREAMER-1/2`	0	—	0

[INFERRED] DISC-30MB / -60MB / -90MB all sharing 7389 pages and DT030 means the size field there is per surface/pack member, not the whole drive; the record DT030 is literally commented % DISC-30/60/90MB in the source listing.

THE DECISIVE FACT: every SCSI and every optical entry has size 0. That is the byte-level statement that SINTRAN does not hold a table of valid SCSI disc sizes — the size is obtained from the drive and from the disc's control record (§4.3, §4.4). [VERIFIED]

2A.4 Word 14 is the logical device number [VERIFIED]¶

The values observed in word 14 are exactly the logical device numbers held in the MDISCS attribute arrays of §2:

word-14 value	device family	matching `MDISCS` entry
`001100`	all SMD / large-disc names	`BBDIS = (BIGDI, 1100, …, 1540)` — SMD
`002210`…`002225`	`DISC-*-SCSI-1` … `SCSI-14` (controller 1..14)	`SCDIS = (SCDI1, 2210, …, 144300)` — SCSI, +1 per controller
`002232`…`002235`	`DISC-*-OPTICAL-1` … `-4`	continues the SCSI logical-device block
`002206`/`002207`	`STREAMER-1/2`	—
`001145`/`001156`	`FLOPPY-DISC-1/2`	—
`000560`, `001111`, `001231`, `001224`	`MAG-TAPE-*`	`1224` also appears as `WWDIS`'s ST-506 logical device no.

This resolves what word 14 is: it is not the IOX hardware device number (500/1540/144300) but the SINTRAN logical device number, and its base values come from MDISCS. [VERIFIED — the 1100 / 2210 coincidence with the byte-verified MDISCS arrays is exact.]

2A.5 SCSI addressing capacity of the name table [VERIFIED]¶

101-S3DNAM.bin names 14 SCSI controllers × 8 targets = 112 SCSI disc units (167770B DISC-SCSI-1 … 173527B DISC-8-SCSI-14) and 4 optical controllers × 8 targets = 32 units (173550B … 174547B). K's copy inside 006-S3FS has the same SCSI block starting at 165556B. Unused slots after 174652B carry the literal name 0000000000'. [VERIFIED]

Note the distinction: 112 names exist, but the number of SCSI unit datafields actually generated into a given system is 9 (K), 10 (L), 4 (M) — see §4.5.

2. The main-swap-disc attribute table `MDISCS` [VERIFIED]¶

Carved from the resident common code, not a segment: versions\L-VSX-500\resident\SINTRAN-DATA_commoncode.bin (load base 0).

 041445: 000000 000000 000000 041510 041510 041510 041510 041510
 041455: 041510 000000 041515 041515 041515 041515 041515 041515
 041465: 041515 041515 041515 041515 041515 041515 041515 041515
 041475: 041515 041522 000000 000000 000000 000000 000000 000000
 041505: 000000 000000 000000 032770 001224 101160 076732 000500
 041515: 031550 001100 077603 074436 001540 036442 002210 102020
 041525: 102020 144300 000000 000000 000000 000000 000000 000000

Decoded (array base MDISCS = 041440B, one word per disc-type index):

index range	pointer	attribute array (5 words)	meaning
00-07	0	—	no swap device
10-15	`041510`	`032770, 001224, 101160, 076732, 000500`	ST-506 / Winchester
16	0	—	—
17-35	`041515`	`031550, 001100, 077603, 074436, 001540`	SMD
36	`041522`	`036442, 002210, 102020, 102020, 144300`	SCSI
37-47	0	—	—

Cross-verification that this really is MDISCS: word 0 of each attribute array is a datafield address, and all three match the L07 symbol table exactly:

word	value	L07 `SYMBOL-2-LIST.SYMB.TXT`
`041510`	`032770`	`WIGDI=032770` (line 362)
`041515`	`031550`	`BIGDI=031550` (line 342)
`041522`	`036442`	`SCDI1=036442` (line 375)

[VERIFIED — three independent hits.]

Element meaning, from the source header at s3vs-4.symb/PH-P2-OPPSTART.NPL listing address 041352 (quoted literally):

% MAIN-SWAPPING-DISC DATA  (XXDIS)
% EACH ENTRY CONSIST OF THE FOLLOWING 4 LOCATIONS:
%       0: ADDRESS OF DATAFIELD
%       1: LOGICAL DEVICE NUMBER
%       2: ENTRY POINT OF DRIVER USED BY BOOTS-STRAP, ("SWAP DRIVER")
%       3: ENTRY POINT OF START-UP DRIVER
%       4: HARDW. DEV NO FOR SWAPPE CONT.
INTEGER ARRAY WWDIS:=(WIGDI,1224,ZWDIS,WIDIS,   500);  % ST-506
INTEGER ARRAY BBDIS:=(BIGDI,1100,ZBDIS,BDISK,  1540);  % SMD
INTEGER ARRAY SCDIS:=(SCDI1,2210,SCSWD,SCSWD,144300);  % SCSI

So, byte-verified in the carved image:

class	logical device no.	hardware device no. (IOX)
ST-506 / Winchester	1224B	500B
SMD (large disc)	1100B	1540B
SCSI	2210B	144300B

This confirms the previously-verified )9BYTT DEVNO mapping (500 / 1540 / 144300) from the inside of the running system, and adds the logical device numbers, which the generation streams do not carry. [VERIFIED]

M-VSX-500 has the same three arrays at 041367B / 041375B / 041403B but with 6-word entries (one extra field): 037175,1224,101434,744,77132,500 / 036465,1100,100007,1425,74636,1540 / 043060,2210,102400,1111,102400,144300. The datafield words again match the M06 symbol table (WIGDI=037175, BIGDI=036465, SCDI1=043060). [VERIFIED]

K-VSX-500: NOT FOUND. No 5- or 6-word array of this shape exists in the carved K resident common code or its segments (searched for word 2210B followed within 6 words by 144300B, and for 1100B followed by 1540B — no hits). The K resident carve may not cover this address range. To settle it, SINTRAN-DATA_commoncode.bin would have to be re-carved for the K05 image over the full resident extent.

2.1 The valid swap-device-type range [VERIFIED from source, table byte-verified]¶

PH-P2-OPPSTART.NPL:722 (listing 044525):

044525   IF SWTYP<<7 OR>>36 THEN CALL ERRFATAL FI  % ILLEGAL MAIN-SWAP-DEVICE TYPE
044535   IF A=7 THEN                               % MAIN SWAP-DEVICE IS FIXED PHOENIX DISK
044540      20=:SWTYP; 40000=:DSKTYPE; 100000=:XXSWTYPE

A system disc type outside 7..36B is a fatal error at cold start. Type 7 is remapped to type 20B with the "fixed" bit 100000B — this is the carved-system counterpart of the generation stream's FR=100000 fixed-disc bit. [VERIFIED as source text; the MDISCS table it indexes is byte-verified above.]

3. How SINTRAN is installed / generated onto a LARGE DISC¶

Two distinct mechanisms exist, and they must not be conflated.

3.1 Off-line: the `)9BYTT` MACM generation stream¶

Already verified in this project and documented in E:\Dev\Ronny\NDInsight\tools\boot-floppy\device-geometry.md. Not repeated here.

3.2 On-line: what the carved kernel itself does — `SINTR` cold start¶

PH-P2-OPPSTART.NPL (SINTR), listing addresses 044525-045101, is the routine that lays a SINTRAN system out on the system disc at cold start. The sequence, quoted:

044525  FILL1:  IF SWTYP<<7 OR>>36 THEN CALL ERRFATAL FI  % find swap device
044552          X:= MDISCS(SWTYPE)=:LDTA                  % ADDR OF DISC-ATTRIBUTES
044565          IF A=0 OR A >< T THEN CALL ERRFATAL FI    % ILLEGAL LOGICAL DEV.NO.
044601          LDTA.SWDDRIVER=:"CTRNSF"                  % DISC-DRIVER ADDR
044606          A=:SWPDRIVER                              % ADDR OF SWAP-DRIVER
044612          A=:MASSNO(0)                              % LOGICAL DEV.NO. OF MAIN-SWAP-DEVICE
044614          X:=DISPE(SWTYP); T:=X.S0; A:=2000=:D:=0; *RDIV ST   % SECTORS/TRACK
044623          IF D><0 THEN CALL ERRFATAL FI
044626          A=:ABLPAGE(0)=:CDABLPAGE
044631          A*200+XXSWTYPE=:BLST(1)                   % SET START OF SEGFIL0 IN SECTORS.
044637          DISPE(SWTYP)=:HTABL(0)                    % INITIAL DISC-LAYOUT TABLE
044663          IF "CTRNSF"="BDISK" OR ="SCSWD" THEN
044672             T:=42; X:=1; A:=0=:D                   % READ FORMAT NECESSARY
044676  DIBUS:    CALL CTRNSF; CALL ERRFATAL; GO DIBUS
044701          FI
044701          X:=HTABL(0); ...  =:CDIFELEMENT           % DISC LAYOUT ELEMENT ADDRESS
...
044757  % COPY FROM SAVE TO IMAGE
044764             X:="LSECO"; CALL FAR CRWDISC           % COPY EXT. COMMON
044773             X:="LSDPT"; CALL FAR CRWDISC           % COPY DPIT
045002             X:="LSRPT"; CALL FAR CRWDISC           % COPY RPIT
045011             X:="LSMPT"; CALL FAR CRWDISC           % COPY MPIT
045020             X:="LSIPT"; CALL FAR CRWDISC           % COPY IPIT
045027             X:="LSSGT"; CALL FAR CRWDISC           % COPY SEGMENT TABLE
045036             X:="LFILS"; CALL FAR CRWDISC           % COPY FILE SYSTEM
045044             X:=20;      CALL FAR CRWDISC           % COPY RT-LOADER PROGRAM SEGMENT

Reading [VERIFIED as source text; every table it consults is byte-verified in §1 and §2]:

The generated constant SWTYP selects both the driver set (MDISCS) and the geometry (DISPE). This is the single point where a system's disc type enters the running kernel.
CDABLPAGE = SECWO / 2000B — sectors per SINTRAN page — is computed from the table's SECWO field. RDIV must divide exactly (IF D><0 THEN ERRFATAL), so SECWO must be an exact divisor of 2000B (1024 words = one page). Every tabulated entry has SECWO = 1000B (2 sectors/page) except DTOD2, which has 2000B (1 sector/page). [VERIFIED against the carved records in §1.3]
BLST(1) = CDABLPAGE*200B + XXSWTYPE — the start of SEGFIL0 in sectors; XXSWTYPE is the fixed-disc bit 100000B. This is the run-time twin of the generation stream's <page>@<shift> FR address construction.
For SMD (BDISK) and SCSI (SCSWD) the layout is not taken on faith: the kernel issues driver function 42 = READ FORMAT against the actual drive and installs the answer as HTABL(0), the live disc-layout element.
Installation proper is a straight page-range copy from the disc's SAVE area to the IMAGE/segment area, one CRWDISC call per system component (common, DPIT, RPIT, MPIT, IPIT, segment table, file system, RT loader). Addresses are page * CDABLPAGE + XXSWTYPE, i.e. derived entirely from the disc-layout table entry — see CRWDISC at listing 044427.

Where the format number comes from on an SMD disc — IP-P2-DISK-START.NPL listing 054620:

054620  FIN:   0=:DIFTCOUNT;
054621         IF CTRG/\77=42 THEN                 % READ FORMAT
054626            CTRG SHZ -6/\7=:X                % X=UNIT NUMBER
054632            X:=HTABL(X); T:=X.DISPN; AD:=PARDF.MEMAD; *DEPO

i.e. fn 42 returns DISPN — the index into DISPE — for that unit; and fn 36 (054014-054052) copies the whole DILEZ-word layout record out to the caller. [VERIFIED as source text]

4. How SINTRAN is installed onto SCSI (K onward)¶

4.1 SCSI is disc type 36B and carries NO tabulated geometry [VERIFIED]¶

The carved DTSSS record (L07 031336B, M06 036253B, K05 012546B) is

 1000  0  0  0  0  0  0  0  36

— SECWO = 1000B (512 words = 1024 bytes per sector) and every other geometry field is zero. There is no cylinder/head/sector description for SCSI anywhere in the kernel. [VERIFIED]

[INFERRED] Consequence: the only disc-size information SINTRAN has for a SCSI disc is what it reads off the device at run time. §4.2 and §4.3 show exactly what it reads and what it rejects.

4.2 The SCSI init sequence, carved¶

Segment 065-S3SIPIT.bin (twin 066-S3IIPIT.bin), load base 32000B; whole-segment disassembly versions\L-VSX-500\re\segments-ref\065-S3SIPIT\065-S3SIPIT.asm. INQUI at 062613B (L07 SYMBOL-2-LIST, +376B revision offset, already established in re\kernel-carving\SCSI-DISKLAYER-COMPLETE\README.md).

INQUIRY, 8-byte allocation length [VERIFIED — full CDB build, re-carved 2026-08-02]:

The whole sequence, from 065-S3SIPIT.asm:

062617  050442   LDT ,B 42       ; T := SMBP1
062620  054443   LDX ,B 43       ; X := SMBP2  -> phys addr of the command buffer
062621  044073   LDA 73          ; A := (062621+73) = 062714
062622  143304   STATX           ; -> command word 0
062623  143315   STZTX 1         ; -> command word 1 = 0
062624  170404   SAA 4
062625  156411   SHA ZIN 11      ; A := 4<<9 = 04000B = 0x0800
062626  143324   STATX 2         ; -> command word 2
062627  143335   STZTX 3
062630  143345   STZTX 4
062631  143355   STZTX 5
062714  011000   (data word)     ; 0x1200

giving CDB words w0=0x1200, w1=0, w2=0x0800, w3..w5=0.

The byte lane is proved, not assumed. Word 0 is 0x1200 and the INQUIRY opcode 0x12 must be CDB byte 0, so byte 0 is the high byte of each word. Confirmed independently by REQUEST SENSE in the same driver, which ORs the LUN into the low byte of word 0:

063351  170407   SAA 7
063352  070400   AND ,B 0        ; A := 7 /\ SUDLU   (the LUN)
063353  156405   SHA ZIN 5       ; LUN << 5
063354  074076   ORA 76          ; \/ 1400B = 0x0300  (opcode 0x03 in the HIGH byte)
063355  143304   STATX           ; -> command word 0

LUN<<5 lands in bits 5-7 of the low byte = SCSI byte 1 bits 5-7, exactly where the LUN belongs. So word 2 = 0x0800 puts byte 4 = 0x08, byte 5 = 0x00.

The 6-byte CDB is 12 00 00 00 08 00 — allocation length 8. 0x0800 was never a length; it is SINBL(4 words) x 2 bytes x 256 (into the high byte lane), and SHZ 11 (<<9) applies both scalings at once.

Independently corroborated by the DMA byte count, which is a literal 8:

063367  000000   (data) SUEBC high word = 0
063370  000010   (data) SUEBC low  word = 8      ; SINBL@1 = 4<<1
063376  024371   LDD -7          ; load the double (0, 8)
063377  020403   STD ,B 3        ; =: SUIBC (byte count) = 8

Symbols: SINBL=000004, SINBS=000070, SMBP1=000042, SMBP2=000043 (L07 SYMBOL-1-LIST.SYMB.TXT).

Why the previous citation was not enough. This block used to quote only 062624/062625 and jump from "A = 04000B" to "allocation length 8". Those two words show only that A becomes 2048; the step to a field value of 8 is a byte-lane claim that was never demonstrated. The conclusion happened to be right, but nothing shown could have detected it being wrong — and SCSI-DEVICE-STRINGS.md section 5.3 leans on it to argue that SINTRAN structurally cannot hold SCSI vendor/product strings. That conclusion now rests on two independent constants (CDB allocation length and DMA byte count, both 8) rather than on an assumed lane.

Device-type dispatch [VERIFIED bytes, SCSI meanings INFERRED]:

addr	word	test	outcome
`062656`	`156570`	`A := SUTYP >> 8` (INQUIRY byte 0)	—
`062657`	`131007`	`JAZ` type == 0	→ `062666`
`062660-062662`	`171003 142065 124004`	type == 3	→ `062666`
`062663-062665`	`171004 140065 124006`	type == 4	→ `062666` else `062673`
`062666-062670`	`044423 174265 004423`	set `SUTYP` bit 6 (`5SCDA`, "direct access")	—
`062671`	`044033` → `062724` = `022400`	CDB word `0x2500` = READ CAPACITY(10)
`062673-062675`	`171001 140065 124003`	type == 1	→ `062676`
`062676`	`044027` → `062725` = `002400`	CDB word `0x0500` = READ BLOCK LIMITS
`062700-062704`	`171177 140065 124004 171007`	type == `177B` (`0x7F`)	`T := 7` error

Named by the recovered NPL (IP-P2-SCSI-DISK.NPL:1240-1256, listing 062260):

062260   IF A SHZ -10=0 OR =3 OR =4 THEN
062270      SUTYP BONE 5SCDA=:SUTYP           % DIRECT ACCESS DEVICE
062273      22400                             % READ CAPACITY
062275   ELSE IF A=1 THEN
062300      2400                              % READ BLOCK SIZE
062302   ELSE IF A=177 THEN
062305      T:=NOLUN; GO FAR ERREX            % NO SUCH LUN

So the SCSI disc layer accepts INQUIRY peripheral device types 0, 3 and 4 as direct-access, type 1 as sequential (tape), and treats 177B as "no such LUN". Anything else falls through to the shift-instruction default. Separately, SCSDISK's device-type gate at 057552-057556B (LDA ,B 23 / SHA ZIN SHR 10 / JAZ / SAT 1 / JMP I 54) rejects any nonzero device type with error TYPER = 1 on the disk path. [VERIFIED]

4.3 THE hard SCSI limit: block/record size, not capacity [VERIFIED]¶

Carved (065-S3SIPIT.asm, addresses octal):

addr	word	meaning
`062757`	`173470`	`AAX 70` — point at the DMA response buffer
`062760`	`143322`	`LDDTX 20` — load the second double of the READ CAPACITY data = BLOCK LENGTH (bytes 4-7)
`062761`	`140005`	`SKP IF DA EQL 0` — high 16 bits must be zero
`062762`	`125110` → `063072`=`063120`	error exit
`062763-062764`	`146115 004441`	`SURSZ (,B 41) := low 16 bits`
`063023`	`171412`	`SAX 12` — X := 10 decimal (2^10 = 1024)
`063036-063041`	`044441 171001 143056 125031`	`SURSZ <= 1` → error exit
`063042-063046`	`175005 124004 173777 156577 124374`	`while (A & 1)==0 { X--; A >>= 1 }`
`063047-063051`	`171001 140065 125021`	`A != 1` after shifting → error exit (not a power of two)
`063052`	`146671`	`D := D - X` = log2(record size)
`063053-063055`	`143007 124002 146107`	if `X < 0` then `X := 0`
`063120-063121`	`171002 125054`	the common error exit: `T := 2`, jump away

Named by the recovered NPL (IP-P2-SCSI-DISK.NPL, listing 062354-062522):

062354   IF SUTYP BIT 5SCDA THEN
062357      T:=SMBP1; X:=SMBP2+SINBS; *LDDTX 20
062363      IF A><0 GO FAR RSZER              % TO BIG
062365      A:=D=:SURSZ                       % DIRECT ACCESS DEVICE SIZE
...
062422   IF SUTYP BIT 5SCDA THEN
062425      X:=12                             % RECORD SIZE 1024 BYTES
...
062434   IF X=:D>1 THEN
062440      IF SURSZ<=1 GO FAR RSZER
062444      DO WHILE A NBIT 0 ; X-1; A SHZ -1 ; OD
062451      IF A><1 GO FAR RSZER              % NOT POWER OF 2
062454   FI
062454   D-X; IF X<0 THEN X:=0 FI             % BYTE AND RECORD SHIFT
...
062522   RSZER: T:=ILRCS; GO FAR ERREX        % ILLEGAL RECORD SIZE

Answer to "any hardcoded size or addressing limit":

Block/record size — YES, a hard limit. The READ CAPACITY block length must (a) fit in 16 bits (IF A><0 GO FAR RSZER % TO BIG), (b) be > 1, and (c) be an exact power of two. Otherwise the unit is rejected with ILRCS ("illegal record size"). SINTRAN's own record is 2^12B = 2^10 = 1024 bytes and the shift 12B - log2(size) converts between them; a size larger than 1024 clamps the shift to 0 at 063053. [VERIFIED]
Capacity — NO limit, and no table. The capacity is a 32-bit double (UHLIM, FILSYS-SYMBOLS L07: UHLIM=177776 = field ,B -2; stored with a double store 057742 020776 STD ,B -2). The bounds check in SCSDISK (057352-057362) is a 32-bit COMPD of ABPA2 + ABP32 against UHLIM, with BADPA (T=5) on overflow. [VERIFIED]
LBA addressing — 32-bit, with automatic 6→10-byte CDB promotion. Carved in CACOB: 064102 171037 SAT 37 / 064103 143456 SKP IF DT MLST SA / 064105 174354 BSET ONE 150 DL, and the buffer-store fork at 064110 175354 BSKP ONE 150 DL → 064112 143316 STDTX (4-byte LBA) versus 064117 145454 RORA SA DL (packed 6-byte CDB). Named by NPL 063504 IF A>>37 THEN L BONE 15 FI % 10 BYTE FORMAT NECESSARY and 063426 IF A><0 OR D>>377 OR D=0 THEN L BONE 15 % 10 BYTES FORMAT NECESSARY. [VERIFIED]

4.4 Where the disc size actually comes from [VERIFIED]¶

READ CAPACITY gives the last block address, which the disk layer uses to locate the control record on the last block. FINEX (057655B) then:

addr	word	meaning
`057673`	`143300`	`LDATX` — control-record header
`057674-057675`	`156570 004775`	`NPART := header >> 8`, stored to `,B -3` (`NPART=177775`)
`057702-057706`	`143300 145051 173401 040347 124374`	XOR checksum loop over the record
`057707-057710`	`140001 124007`	checksum `!= 0` → `NOCRC`
`057711-057713`	`171002 143046 124004`	`NPART <= 2` → `NOCRC`
`057714-057716`	`171012 143046 124003`	`NPART > 012B` (= `NCOPA=000012`) → `NOCRC`
`057732`	`143110`	`MOVEW` — copy the partition table to the caller
`057741-057742`	`143316 020776`	`STDTX` then `STD ,B -2` → `UHLIM` (32-bit)
`057743-057744`	`170436 143304`	status `36B` to the caller
`057747`	`146106`	`T := 0` (success)

So a SCSI disc's size and partitioning are self-describing, held in a control record on the disc's own last block, protected by an XOR checksum, and the only structural constraint is 2 < NPART <= 10 partitions (NCOPA = 000012 in FILSYS-SYMBOLS). There is no kernel table of legal SCSI disc sizes. [VERIFIED]

4.5 How many SCSI units a generated system supports [VERIFIED counts, INFERRED meaning]¶

Scanning the carved DPIT segments for the SCSI hardware device number 144300B yields regularly-spaced datafield records:

version	file	unit-datafield addresses	stride	count
K-VSX-500	`044-S3IDPIT.bin`	`013242, 013470, 013716, 014144, 014372, 014620, 015046, 015274, 015522`	`226B`	9
L-VSX-500	`044-S3IDPIT.bin`	`036437, 036642, 037045, 037250, 037453, 037656, 040061, 040264, 040467, 040672`	`203B`	10
M-VSX-500	`044-S3IDPIT.bin`	`043055, 043260, 043463, 043666`	`203B`	4

Plus, in each, one further record whose neighbour word is the SCSI driver's interrupt entry (L: …144300 56621 67247 — 67247B is SCINT, the carved SCSI interrupt handler), i.e. the controller datafield as opposed to a unit datafield. [VERIFIED addresses/words; INFERRED that these are per-unit datafields and therefore that the number of configured SCSI units differs per generated system: 9 / 10 / 4.] This is a generation parameter, not a kernel-imposed maximum; no maximum-unit constant was found.

5. `@CREATE-DIRECTORY`, `ALBIT`, and the bit file¶

Carved from versions\L-VSX-500\segments\006-S3FS.bin (load base 26000B); disassembly re\segments-ref\006-S3FS\006-S3FS.asm. ALBIT = 137500B, CRDIR = 136741B (SYMBOLS\L07\FILSYS-SYMBOLS.SYMB.TXT:1314).

Bit-file default placement [VERIFIED] — 137523-137532: LDT ,B 1 / JPL I 63→GSIZE (037101B) / 156777 SAD ZIN SHR 1 / 171011 SAT 11 / 141660 RDIV ST / 141265 RMPY ST DA / 020410 STD ,B 10 ⇒ bit_start = 9 * floor(floor(pages/2)/9). The 9 is a hard literal (SAT 11), identical for every device — it is not read from the disc-layout table.
Bit-file span [VERIFIED words, INFERRED arithmetic] — 137535-137544: GSIZE then 156602 SAD ZIN 2 (32-bit shift left 2) then 172777 AAA -1 ⇒ ceil(pages / 16384) bitmap pages, i.e. 16384 bits per page = 1 bit per page over a 1024-word page.
"Place at top of disc" — a supplied 32-bit bit address of -1 (137545-137562) makes ALBIT compute pages - bitmap_pages. [VERIFIED words, INFERRED meaning]
The only size check [VERIFIED] — 137563-137575: 32-bit compare of bit_start against the declared page count; on failure 170500 SAA 100 (error code 100B) and 124123 JMP to the error return.
NO MAXIMUM DISC SIZE IS ENFORCED IN ALBIT OR CRDIR. [scope corrected 2026-08-02 - the exhaustive read covers those two routines (~370 words), not the kernel. A ceiling enforced by a caller, by GSIZE, by DEFINE-MASS-STORAGE-UNIT or on the disc-layout path would leave both looking exactly as read. Absence in two routines is not absence in SINTRAN.] An exhaustive read of ALBIT (137500B..137730B) and CRDIR (136741B..137477B) finds no comparison of the page count against any ceiling constant. [VERIFIED by exhaustive read] The only implicit ceiling is structural: the bitmap-page count is kept in the 16-bit A register (137544 004421 STA ,B 21) after a 32-bit <<2, which overflows at pages >= 2^30. [INFERRED]
@CREATE-DIRECTORY does NOT write a disc boot area [VERIFIED for the call, INFERRED for the byte range] — CRDIR reserves the device with 137046 153124 MON 124, calls ALBIT (137145, pointer word 137234 = 137500), GSIZE (137153 → 037101), WBFBU (137167 → 050565), ALPAG (137324 → 050627), then writes the directory label / master block through WDIRS at 137403 (pointer 137440 = 040221) and releases with 137417 153125 MON 125. There is no bootstrap-area write.

Correction to existing repo docs: SINTRAN\Filesystem\create-directory-placement.md (≈ lines 49, 104, 182) and create-directory.md §4 read 137710B MPY 20 as "multiply by 16 = 16 bits/word". That is wrong: MPY (120000B) is a memory-reference instruction with no immediate form; 20 is a P-relative displacement and both MPY sites (137701B MPY 27 and 137710B MPY 20) fetch the same constant word 000011B = 9 at 137730B. They belong to the bad-page relocation (± one 9-page track), not to bitmap sizing. The 16384-bits-per-page conclusion is still correct but comes from 137540 156602 SAD ZIN 2. [VERIFIED]

6. Installation artefacts that survive inside the running system¶

6.1 The command-name table of the command processor [VERIFIED]¶

versions\L-VSX-500\segments\003-S3CP.bin, load base 30000B. The '-terminated ASCII name table runs 054131B … 060230B, immediately followed by a parameter-descriptor pointer table at 060231B. Confirmed against re\segments-ref\003-S3CP\003-S3CP.asm:11217:

054131  025102      LDD I 102        ; 0x2A42 = '*','B'  -> "*BACKUP-DIRECTORY"
054132  040503                     ; 0x4143 = 'A','C'

Disc / directory / mass-storage commands, octal addresses in L:

addr	commands
`054131`	`BACKUP-DIRECTORY` `COPY-USERS-FILES` `*CREATE-VOLUME`
`054163` / `054172`	`LIST-VOLUME` / `RETRIEVE-DIRECTORY`
`054352`	`CHANGE-BIT-FILE` `CHANGE-DIRECTORY-ENTRY`
`054476` / `054543`	`CLEAR-MAIN-DIRECTORY` / `COLD-START`
`054573` / `054611`	`COPY-DEVICE` `COPY-DIRECTORY` / `COPY-FILE` `CREATE-DIRECTORY`
`054777`	`DEFINE-MASS-STORAGE-UNIT`
`055135`	`DELETE-FILE` … `DELETE-MASS-STORAGE-UNIT`
`055250`	`DEVICE-FUNCTION` `DIRECTORY-STATISTICS`
`055331` / `055337`	`DSCNT` `DUMP` / `DUMP-BIT-FILE` `DUMP-DIRECTORY-ENTRY`
`055447`	… `ENTER-DIRECTORY` `EXECUTE-IOX` `EXPAND-FILE`
`056103`	… `LIST-MASS-STORAGE-UNITS` …
`056332` / `056373`	`LOAD-BINARY` … / `MEMORY-LIMITS` `MODE`
`056533`	`PRSRV` `RECOVER` `REGENERATE-DIRECTORY`
`056653`	… `RESERVE-DEVICE-UNIT` `RESERVE-DIRECTORY`
`057055` / `057376`	`SAVE-DIRECTORY` / `SET-MAIN-DIRECTORY`
`057410`	`SET-MASS-STORAGE-SIZE` `SET-MEMORY-CONTENTS`
`060046` / `060136` / `060152`	`TEST-DIRECTORY` / `UNLOCK-DIRECTORY` / `UNRESERVE-DIRECTORY`

M's table is the same, sorted, at 061770B…066022B (DEFINE-MASS-STORAGE-UNIT 062636B, LIST-MASS-STORAGE-UNITS 063753B, SET-MASS-STORAGE-SIZE 065260B). K's table is in definition order, not sorted, observed extent 047753B…053723B; K's exact start address is NOT FOUND. [VERIFIED except where marked]

The S3CP table entry that dispatches @CREATE-DIRECTORY to CRDIR=136741B (the pointer, as opposed to the name string at 054611B) was NOT FOUND.

6.2 The `DEVICE-FUNCTION` subcommand table [VERIFIED]¶

L 003-S3CP.bin, 116734B…117406B. Disc-relevant subcommands: CLEAR-DEVICE, READ-STATUS, READ-FORMAT, CLEAR-SELECTED-UNIT, RESERVE-DEVICE, RELEASE-DEVICE, SET-FLOPPY-FORMAT, FORMAT-FLOPPY, GET-CURRENT-DISC-ADDRESS, DUMP-BOOTSTRAP, SET-CURRENT-DISC-ADDRESS.

Version deltas: K has no RESERVE-DEVICE/RELEASE-DEVICE here; M adds FORMAT-TRACK at 125622B. [VERIFIED]

DUMP-BOOTSTRAP exists as a subcommand at K 153367B, L 117361B, M 125575B. [VERIFIED — the name-table entry and handler address]

~~This is the mechanism that writes the disc boot area; @CREATE-DIRECTORY does not (§5).~~ [DEMOTED TO INFERRED 2026-08-02 — and contradicted.]

A name-table entry proves the subcommand exists and where its handler starts. It says nothing about what the handler writes, or which device classes it accepts. The competing reading — that DUMP-BOOTSTRAP is floppy-only, and a hard-disc page 0 is authored by the cold-start kernel routine in PH-P2-OPPSTART.NPL (FILL2/PL011) — produces the identical name-table entry at the identical address. The cited evidence cannot separate them.

Two sources point the other way:

DISC-BOOTSTRAP.md lines 454-456.
ND-60.128.5 p.97: "Allowed only on floppy disk".

DOC-AUDIT.md (lines 42, 86-105, edit #5 at 174-176) already ruled that DISC-BOOTSTRAP.md is the better-supported document and recommended this demotion. The edit was never applied; it is applied now.

Why it matters: acting on the original sentence would point a boot-block writer at a hard disc pack.

To settle it: carve the handler at L 117361B (base 30000B) and look for a device-class parameter and a MOVNP/CRDISC page-0 write — the check FIRST-BOOT.md lines 256-260 already proposes.

6.3 The canned cold-start `ENTER-DIRECTORY` line [VERIFIED]¶

Present verbatim in all three versions, twice each:

version	file	octal addr	text
K	`003-S3COM.bin`	`067171`	`ENTER-DIRECTORY PACK-ONE DISC-75MB-1 0'`
K	`005-S3ERRS.bin`	`130740`	same
L	`003-S3CP.bin`	`074123`	same
L	`005-S3ERRS.bin`	`144744`	same
M	`003-S3CP.bin`	`102325`	same
M	`005-S3ERRS.bin`	`144744`	same

Followed in 005-S3ERRS by SC-75MB-1', SYMB', DATA', MODE', (SCRATCH)SCRATCH01', SYSTEM'. [VERIFIED bytes; INFERRED that this is the built-in default main-directory entry command plus default file types.] Note DISC-75MB-1 is exactly a name in the device-name table of §2A (164021B, 36945 pages).

6.4 The `@CREATE-DIRECTORY` / `@DEFINE-MASS-STORAGE-UNIT` dialogue [VERIFIED]¶

006-S3FS.bin (base 26000B), prompt table from 043220B:

043220  SA DIRECTORY NAME: '
043233  SA OLD DIRECTORY NAME: '
043247  SA NEW DIRECTORY NAME: '
043263  SA DEVICE NAME: ' ID DEVICE UNIT: ' ID DEVICE SUB-UNIT: ' SA FIXED(F) OR REMOVABLE(R): '
043334  SU USER NAME: ' ID NUMBER OF PAGES: ' ...
043535  IB DEVICE NUMBER (OCT): '
046135  IB OCTAL BIT FILE ADDRESS (-1 END OF DISC): ' SA MANUAL CHECK? '
046174  SA VOLUME NAME: '
153331  MASTER BLOCK'   BIT FILE'

SA DEVICE NAME: is the prompt that is matched against the §2A device-name table; ID NUMBER OF PAGES: is the operator-supplied size that reaches GSIZE and thence ALBIT (§5); IB OCTAL BIT FILE ADDRESS (-1 END OF DISC) is exactly the -1 case carved at ALBIT 137545B. [VERIFIED strings; the linkage to ALBIT is VERIFIED at the -1 site, INFERRED for the others.]

6.5 Cold-start MODE-file machinery [VERIFIED]¶

070-S3SSM.bin / 071-S3SM.bin (L): SET-COLDSTART-MODE-FILE' RESET-COLDSTART-MODE-FILE' at word offset 014063B, and at 064273B: PARAMETERS TO THE ENTER-DIRECTORY COMMAND' $WHEN ENTERING MAIN DIRECTORY: ' S COLDSTART INPUT FILE: ' S COLDSTART OUTPUT FILE: '. In K these live inside the merged 003-S3COM / 013-S3OPCSV.

6.6 Library marks — the system-generation record of each build [VERIFIED]¶

SYMBOLS\{K03,L07,M06}\LIBRARY-MARKS.SYMB.TXT are the conditional-assembly switches of each build (K 687, L 732, M 914 marks). Example raw line (L07, line 539): 176000/^8SCSI %000004.

mark	K03	L07	M06	[INFERRED] meaning
`8SCSI`, `8SCS1`, `8ZSCS`	Y	Y	Y	SCSI support / controller 1
`8SCS2`	—	Y	—	SCSI controller 2
`8SCOD`, `8SCMT`	—	Y	Y	SCSI optical disc / SCSI mag tape
`8BDIS`, `8ZBDI`	Y	Y	Y	big disc (SMD) driver
`8BDIO`, `8WDIS`, `8ZWDI`	—	Y	Y	BDIO pool / Winchester disc
`7D1U0`…`7D8U0`	only `7D1U0`	all 8	`7D1U0`,`7D2U0`	SCSI disc unit marks
`7O1U0`,`7O2U0`	—	Y	Y	optical units
`WM300`, `WM310`	Y	—	—	K only

8SCS2 existing only in L matches symbol SCSI2 existing only in L07 — this particular L machine was generated with two SCSI controllers; the K and M machines with one. [INFERRED — downgraded 2026-08-02]

A conditional-assembly mark EXISTING in a build's mark list is not the same as the mark being SET, and neither is the same as a controller COUNT. "The L build defines a second-controller mark that is not enabled", or "the mark gates a driver variant rather than an instance", produce the identical symbol-table difference. Section 4.5 also counts 10 SCSI unit datafields in L against 9 in K, which is not a 2:1 ratio.

To settle it: the LIBRARY-MARKS line format carries a value (176000/^8SCSI %000004) — read 8SCS2's value and compare against a mark known to be off; or count SCSI CONTROLLER datafields (those adjacent to SCINT) rather than unit datafields.

6.7 Version deltas that matter for disc support [VERIFIED]¶

Device-name table moved out of 006-S3FS into its own segment pair 100/101-S3*DNAM at K→L.
SCSI is already fully present in K: DISC-SCSI-* names at 165556B in K's 006-S3FS, marks 8SCSI/8SCS1/8ZSCS, symbols SCSI1/SCSID, DTSSS disc-layout record at 012546B, DISPE[36B] non-zero.
K's carved DISPE has no ST-506/Winchester types (indices 10-15 zero); L and M have all six. Marks 8WDIS/8ZWDI are absent from K03 accordingly.
DTOD1/DTOD2 (optical) are absent from K03, present in L07/M06 — matching the carved DISPE[40B]/DISPE[41B] being zero in K and non-zero in L/M.
RESERVE-DEVICE/RELEASE-DEVICE added to DEVICE-FUNCTION in L; FORMAT-TRACK added in M.
Command-name table becomes alphabetically sorted from L onward.

6.8 Things explicitly NOT FOUND in the strings sweep¶

No error text about a disc being too large. The nearest are DIRECTORY INDEX TOO LARGE / OBJECT INDEX TOO LARGE (L 014-S3ERRP.bin, word offset 012423B).
No WINCHESTER string anywhere in any of the three systems.
No standalone SCSI / SMD / ND-3201 device-type word — the only SCSI text in the images is inside the DISC-n-SCSI-m device names.
MODE is an ordinary command (L 056373B), not an install-only :MODE construct.
M-VSX-500 has no list-segment.txt / list-rt-programs.txt in inputs\.
No RT program with a disc/SCSI name in either K's or L's list-rt-programs.txt; the mass-storage-adjacent ones are 1SWAP, 5SWAP, RTDIL, RTRFA, DIMWD (L only).

7. `MSTYP` — the open question is NOT resolved¶

The cross-reference asked whether MSTYP's value could be recovered from the carved symbol tables.

NOT FOUND. Exhaustive grep of

SINTRAN\NPL-SOURCE\SYMBOLS\K03\*.SYMB.TXT
SINTRAN\NPL-SOURCE\SYMBOLS\L07\*.SYMB.TXT
SINTRAN\NPL-SOURCE\SYMBOLS\M06\*.SYMB.TXT

for MSTYP, and for every symbol beginning MST, returns only MSTAT, MSTEN, MSTMP, MSTOR, MSTPN, MSTRM, MSTRN, MSTS, MSTTY, MSTUS — no MSTYP. The only near-hit anywhere is CMSTY (L07 SYMBOL-2-LIST.SYMB.TXT:2960 CMSTY=120462, M06 :2862 CMSTY=121200), which the NPL listing (s3vs-4.symb:38028 SUBR CSTYP,CMSTYP) shows is an unrelated routine. [VERIFIED negative]

Since MAC symbol significance is the last 5 characters and MSTYP is exactly 5, truncation cannot be hiding it. [INFERRED]

The documented Drum=0 / NCR=1 / CDC=2 / Large disc=3 numbering is therefore neither verified nor refuted by the carve. What the carve does show is that the running kernel does not use a value of that shape at all: it uses SWTYP in the range 7..36B indexing MDISCS and DISPE (§2, §2.1). Those two numbering schemes are not the same and should not be equated without further evidence. [INFERRED]

8. What is still uncarved¶

Gap	What would have to be carved
K-VSX-500 `MDISCS` / `WWDIS` / `BBDIS` / `SCDIS`	Re-carve the resident common code for the K05 image over the full resident extent (`SINTRAN:DATA` pages 1..63 of the K05 SMD image) into `versions\K-VSX-500\resident\SINTRAN-DATA_commoncode.bin`; the present K carve does not contain the array.
`SINTR` cold-start code as carved bytes	`PH-P2-OPPSTART` lives on the start segment; segment `0002 S3IMAGE` (load base 0, `madr` page 0, 63 pages) is carved but has no `segments-ref` disassembly. Produce `segments-ref\002-S3IMAGE` to byte-verify §3.2 rather than relying on the NPL listing.
`@CREATE-DIRECTORY` command-table dispatch pointer	`003-S3CP` (load base `30000B`). The name table is now located (`054131B`-`060230B`, §6.1) and the parameter-descriptor pointer table starts at `060231B`; the routine-address table that maps `CREATE-DIRECTORY` → `CRDIR 136741B` still has to be walked out from `060231B`.
K-VSX-500 command-name table start	K's `003-S3COM` table is unsorted and the run found begins mid-word at `047753B`; a few words earlier is the true start. Disassemble `K-VSX-500\segments\003-S3COM.bin` around `047740B` to pin it.
Meaning of device-name-table words 12/13 (`BABST=030346`, `000515`)	Not decoded. They are constant per device class.

No source image was modified; every read in this analysis was read-only.

9. Provenance¶

Carved binaries: E:\Dev\Ronny\NDInsight\tools\sintran-segment-carver\versions\{K,L,M}-VSX-500\segments\*.bin, …\resident\*.bin
Disassemblies: …\versions\L-VSX-500\re\segments-ref\065-S3SIPIT\065-S3SIPIT.asm, …\006-S3FS\006-S3FS.asm
Prior byte-verified carves relied on: …\versions\L-VSX-500\re\kernel-carving\SCSI-DISKLAYER-COMPLETE\README.md, …\SCSI-DRIVER-COMPLETE\, …\SMD-DRIVER-BASELINE\README.md
Symbol tables: E:\Dev\Ronny\NDInsight\SINTRAN\NPL-SOURCE\SYMBOLS\{K03,L07,M06}\*.SYMB.TXT
Recovered source listings (naming/comments only, always cross-checked): E:\Dev\Ronny\NDInsight\SINTRAN\NPL-SOURCE\s3vs-4.symb, …\NPL\PH-P2-OPPSTART.NPL, …\NPL\IP-P2-DISK-START.NPL, …\NPL\IP-P2-SCSI-DISK.NPL
Device-name table: …\versions\L-VSX-500\segments\101-S3DNAM.bin (+ 100-S3SDNAM.bin), …\M-VSX-500\segments\101-S3DNAM.bin, …\K-VSX-500\segments\006-S3FS.bin
Command tables / strings: …\segments\003-S3CP.bin (K: 003-S3COM.bin), 005-S3ERRS.bin, 006-S3FS.bin, 014-S3ERRP.bin, 070-S3SSM.bin / 071-S3SM.bin, 037-S3RTD.bin
Generation switches: E:\Dev\Ronny\NDInsight\SINTRAN\NPL-SOURCE\SYMBOLS\{K03,L07,M06}\LIBRARY-MARKS.SYMB.TXT
Live listings: …\versions\{K,L}-VSX-500\inputs\list-segment.txt, list-rt-programs.txt
Tool: E:\Dev\Ronny\NDInsight\tools\boot-floppy\tools\find_disc_layout_table.py

No file under D:\ND\, no .img/.image, and no carved binary was modified. Nothing was written outside E:\Dev\Ronny\NDInsight\tools\boot-floppy\ and the session scratchpad. INSTALL-PROCEDURE.md was not touched.