Skip to content

ND SCSI disk PHYSICAL-MANAGEMENT layer - creation spec

What this covers: the disk-management layer underneath the SINTRAN filesystem on an ND SCSI (ND-3201 controller) disk - the extra blocks a SCSI disk carries beyond its declared filesystem capacity: the spare / reserved top-of-disk region, the disk-parameter block, and the defect / reallocation table at the last block. It is written as a creation specification so an external tool (retrofs and its libraries) can generate byte-valid ND SCSI disk images.

Scope boundary. The four NDFS on-disk structures (directory label, object file, user file, page bitmap) are documented under ../../Filesystem/README.md and ../../Filesystem/on-disk-format/. This document is the layer below them - the physical medium the controller presents to SINTRAN. It does not re-specify the filesystem; it specifies the region map the filesystem lives inside and the controller metadata at the top of the medium.

This layer is SCSI / ND-3201-specific. SMD disks and floppies do not have it (proven below). See nd-scsi-3201.md for the Z80 firmware analysis and ../../Filesystem/code-logic/scsi-mount-geometry.md for the mount path that reads the last block.

Rule of evidence (same convention as the filesystem RE):

  • VERIFIED - proven from real disk bytes (SCSI-K.image), the carved 006-S3FS SINTRAN L bytes, the RetroCore C# target, or the ND-3201 firmware analysis, or from arithmetic that is fully determined.
  • INFERRED - strong reasoning from the bytes + architecture, not one decisive source.
  • OPEN - not decidable from the single available (defect-free) dump; the second dump or firmware read that would close it is named.

ND addresses are octal; SCSI LBAs, block counts and table fields are hex/decimal as marked. 1 block = 1024 bytes; 1 ND page = 2048 bytes = 2 blocks.

0. TL;DR

  1. A SCSI disk is bigger than its filesystem. On the real SCSI-K.image dump: raw = 129,312 blocks (64,656 ND pages), usable filesystem = 122,072 blocks (61,036 pages). The difference, 7,240 blocks (3,620 pages), is a reserved top-of-disk region. VERIFIED.

  2. READ CAPACITY reports the RAW size, not the usable size. It returns last-LBA = 129311 (= raw total - 1). The usable figure (122,072) is a filesystem / table-level number SINTRAN derives from the directory and cross-checks against the table - it is not what the controller reports. VERIFIED (two independent ways, Section 3).

  3. The last block is a defect / reallocation table. LBA 129311 (offset 0x07E47C00) holds a signed table: a 16-byte header (signature, flags, usable-block-count) followed by 12-byte extent entries {flag, physical-LBA, run-length}. VERIFIED (decoded from real bytes, Section 4).

  4. The second-to-last block is a disk-parameter block. LBA 129310 holds a device-type / vendor / revision / format-date record. VERIFIED (Section 5).

  5. SMD and floppy have no such table. SMD0.IMG last block = all zeros; floppy 250305L07-XX-01D.IMG last block = ordinary filesystem/loader data. The layer is unique to the ND-3201 SCSI controller. VERIFIED (Section 6).

  6. To create a valid image, retrofs writes the filesystem in blocks 0 .. usable-1, reserves the top region, and writes the parameter block and the defect/reallocation table at the top two blocks, with READ CAPACITY reporting the raw last-LBA (Section 8).

1. Layout of a full ND SCSI disk image

flowchart TB
    subgraph DISK["ND SCSI disk image (raw = C x H x S blocks of 1024 B)"]
        direction TB
        FS["FILESYSTEM REGION<br/>blocks 0 .. usable-1<br/>(NDFS: page 0 label, bit file,<br/>object file, user file, data)<br/>SCSI-K: 0 .. 122071 = 61036 pages"]
        SLACK["RESERVED / SPARE SLACK<br/>blocks usable .. top_reserved-1<br/>unallocated, fill pattern<br/>SCSI-K: 122072 .. 129097 (0x43 'C' fill)"]
        POOL["SPARE-POOL EXTENTS<br/>enumerated by E0... entries in the table<br/>SCSI-K: 129098 .. 129309 (212 blocks)"]
        PARM["DISK-PARAMETER BLOCK<br/>block raw-2<br/>SCSI-K: LBA 129310 (C0... entry)"]
        TABLE["DEFECT / REALLOCATION TABLE<br/>block raw-1 (LAST block)<br/>SCSI-K: LBA 129311 (C0... entry)"]
    end
    FS --> SLACK --> POOL --> PARM --> TABLE

    classDef fs fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#1B1B1B;
    classDef slack fill:#FFF3E0,stroke:#E65100,stroke-width:2px,color:#1B1B1B;
    classDef pool fill:#E0F7FA,stroke:#00838F,stroke-width:2px,color:#1B1B1B;
    classDef meta fill:#F3E5F5,stroke:#7B1FA2,stroke-width:2px,color:#1B1B1B;
    class FS fs;
    class SLACK slack;
    class POOL pool;
    class PARM,TABLE meta;

Offset map (SCSI-K.image, VERIFIED):

Region First LBA Last LBA Blocks Byte offset (first) Content
Filesystem 0 122071 122,072 0x00000000 NDFS (label/bitmap/object/user/data). Tail 0x43 from block 122052.
Reserved slack 122072 129097 7,026 0x07736000 0x43 ('C') fill, not enumerated in the table
Spare-pool extents 129098 129309 212 0x07E12800 E0... extents (mostly zero, top run 0x43)
Parameter block 129310 129310 1 0x07E47800 disk-parameter record (C0... entry)
Defect / realloc table 129311 129311 1 0x07E47C00 the table (C0... entry)

The exact filesystem/slack boundary in the fill pattern is cosmetic: on this disk the 0x43 fill begins at block 122052 (inside the still-usable tail), so the fill does not mark the usable boundary - the usable count lives in the table header and the directory, not in the fill. VERIFIED (byte scan).

2. Raw vs usable capacity - the relationship

Definitions (all in 1024-byte blocks unless noted):

raw_blocks    = C x H x S                         (physical medium, whole ND pages => even)
last_LBA      = raw_blocks - 1                     (what READ CAPACITY reports)
usable_blocks = 2 x (directory pages-available)    (NDFS filesystem size)
reserved      = raw_blocks - usable_blocks          (top-of-disk reserve)

SCSI-K.image (VERIFIED):

Quantity Blocks Pages Hex Source
raw_blocks 129,312 64,656 0x1F920 image length 132,415,488 B / 1024; = 898*8*18

Evidence downgraded 2026-08-02. A capacity match cannot fix a C/H/S split - 129,312 also factors as 179689, 4491618 and others. The split itself comes from RetroCore SCSIHDD.cs, our own re-implementation, which is circular for a claim about the real drive. Nothing breaks today because SCSI addressing is LBA-based. To settle it: MODE SENSE page 03h/04h from the real unit, the ND-3201 firmware geometry table, or the SINTRAN disc-type table entry.

| last_LBA | 129,311 | - | 0x1F91F | READ CAPACITY; carved probe reads exactly this | | usable_blocks | 122,072 | 61,036 | 0x1DCD8 | table header word[3]; = directory pages x 2 | | reserved | 7,240 | 3,620 | 0x1C48 | raw - usable |

How usable is derived: SINTRAN reads the NDFS directory label on page 0 (pages-available in the extended-info block, see ../../Filesystem/on-disk-format/extended-info-block.md) and the carved CHDSI/GSIZE path reconciles a stored capacity against the configured disk-type geometry (scsi-mount-geometry.md Section 1). The defect-table header also records usable_blocks (word[3]) so the controller / format tool and SINTRAN agree. VERIFIED (table word[3] = directory pages x 2).

What the accounting does and does NOT close (honest statement). The table enumerates only the top 214 blocks (2 metadata + 212 spare-pool, Section 4). It does not enumerate the other 7,026 reserved blocks (122072..129097). So reserved (7,240) is a capacity difference, while the last-block table is a compact directory of the top region, not a full bitmap of the whole spare area. The bulk 7,026 blocks are unallocated slack between the filesystem's declared usable size and the physical top-region reserve. VERIFIED (214 enumerated vs 7,240 reserved); OPEN: whether the slack is intended free spare or just device-vs-format rounding needs a second disk / firmware.

3. What READ CAPACITY reports: RAW, not usable

VERIFIED - one genuine proof, plus one that is not independent:

Evidence downgraded 2026-08-02. Proof 1 is RetroCore's SCSIHDD, our own re-implementation: it reports the raw last LBA because we wrote it to, and could not have failed whatever a real ND-3201 does. Proof 2 - SINTRAN successfully reading LBA 129311, above the usable count - IS discriminating and carries the conclusion alone. To settle it: nothing further; relabel proof 1 as "consistent with" rather than independent.

  1. Emulator/firmware model. RetroCore SCSIHDD.cs: DiskSizeInBlocks = cylinders*heads*sectors - 1 = 898*8*18 - 1 = 129311, and CommandReadCapacity() returns that value as the last LBA (block size 1024). That is raw_blocks - 1, i.e. the raw last LBA. (Cited in scsi-mount-geometry.md Section 3.)

  2. The real disk proves it by construction. On SCSI-K.image SINTRAN's mount issues a READ(6) of LBA 0x1F91F = 129311 and it returns SS_GOOD with real data - that block is the defect table. A block at LBA 129311 can only be addressed if READ CAPACITY reported a last-LBA >= 129311. If the controller clamped capacity to the usable 122,072, LBA 129311 would be out of bounds and the read would fail. It does not. Therefore the controller reports the raw size and the top region is normally addressable. VERIFIED (real bytes + carved probe).

Consequence for usable: usable_blocks is never surfaced by READ CAPACITY. It is a filesystem/table-level figure. The controller hands SINTRAN the whole raw medium; SINTRAN decides how much of it is the filesystem from the directory label (cross-checked against the table header word[3]).

Transparent defect remapping - OPEN. Whether the ND-3201 firmware silently re-routes a defective filesystem LBA to a spare block (SCSI "automatic reallocation") cannot be proven from this defect-free dump: no filesystem LBA is remapped here. The firmware analysis (nd-scsi-3201.md) documents floppy defect detection (FD1797 "diskette defect", event 0x1D03) but shows no SCSI read-path LBA translation - the reserved-block table sits on the medium and reads as host/format-tool-managed. Whether remap is automatic (firmware) or manual (format/repair tool rewrites the table + moves data) is OPEN; a dump of a disk with a known reallocated bad sector would settle it (an E0... extent whose target block holds the moved data instead of fill/zero).

4. The defect / reallocation table (LAST block) - full field spec

Mount-math cross-reference. This same last block is the SCSI control record the mount reads via the driver's function-42 connect. The header word[3] (0x0001DCD8 = 122072 = UHLIM) and the partition-table field that lands in ,B 11 are the two values the 006-S3FS geometry routine consumes - and the routine's (UHLIM/2)/divisor division is where the @ENTER-DIRECTORY error 243B originates. Full decode of the mount math, with the executed trace values and the pinpointed failure, is in scsi-control-record-and-mount-math.md. Per-field mount impact: word[3]=UHLIM feeds the geometry division; word[0] high byte = NPART=8 feeds the function-42 partition parse; the E0/C0 extent map is not used by the geometry routine.

Real bytes, SCSI-K.image LBA 129311, byte offset 0x07E47C00 (32-bit big-endian words):

07e47c00: 0800 54d9 8000 0000 0000 0000 0001 dcd8   <- header (16 bytes)
07e47c10: c000 0000 0001 f91f 0000 0001            entry 0
          c000 0000 0001 f91e 0000 0001            entry 1  (07e47c1c)
07e47c28: e000 0000 0001 f91d 0000 0001            entry 2
07e47c34: e000 0000 0001 f909 0000 0014            entry 3
07e47c40: e000 0000 0001 f8f5 0000 0014            entry 4
07e47c4c: 0000 0000 0000 0000 0000 0000            entry 5  (null slot)
07e47c58: e000 0000 0001 f84a 0000 00ab            entry 6
07e47c64: 00 00 ... (zero fill to end of block)             terminator

4.1 Header (16 bytes, 4 x 32-bit BE) - VERIFIED layout, partial semantics

Off Word Value (SCSI-K) Meaning Grade
0x00 word[0] 0x080054D9 signature / magic (or per-disk checksum) VERIFIED value / OPEN meaning
0x04 word[1] 0x80000000 flags - bit 31 = table valid/present VERIFIED value / INFERRED meaning
0x08 word[2] 0x00000000 reserved (or high word of a 64-bit usable count = 0) VERIFIED value / INFERRED meaning
0x0C word[3] 0x0001DCD8 = 122072 usable block count VERIFIED (= directory pages x 2)
  • word[3] is proven = usable blocks: 0x1DCD8 = 122072 = 61036 pages x 2, matching the directory's declared filesystem size. This is the load-bearing field a reader trusts.
  • word[0] 0x080054D9 is a fixed-looking signature. It is not a plain 32-bit sum of the block (sum of words[1..] = 0x800DB350) nor a plain XOR (XOR of words[1..] = 0x8001DCD8). So it is either a magic constant or a checksum with an algorithm not derivable from one block. OPEN - a second disk dump distinguishes "constant" (same value) from "checksum" (different value); until then a writer should treat it as a copied constant (Section 8).
  • word[1] 0x80000000: only bit 31 set - a "valid" marker echoed in every live entry's flag (all set bit 31). INFERRED.

4.2 Entry format (12 bytes, 3 x 32-bit BE) - VERIFIED layout

+0  flag/type   (32-bit)   observed 0xC0000000 or 0xE0000000, 0 = empty slot
+4  physical LBA (32-bit)  the block/extent start, in raw LBA space
+8  run length  (32-bit)   number of consecutive blocks

Entries begin at header end (0x10) and are packed at 12-byte stride. VERIFIED (the stride and 3-field split reproduce every non-zero region in the block exactly).

4.3 Decoded entries (SCSI-K) - VERIFIED

# Off flag LBA (dec) run Covers Block content Note
0 0x10 C0000000 129311 1 129311 the table itself controller metadata
1 0x1C C0000000 129310 1 129310 parameter block controller metadata
2 0x28 E0000000 129309 1 129309 zero spare-pool extent
3 0x34 E0000000 129289 20 129289-129308 zero spare-pool extent
4 0x40 E0000000 129269 20 129269-129288 zero spare-pool extent
5 0x4C 00000000 - - - - null slot (skipped, not terminator)
6 0x58 E0000000 129098 171 129098-129268 0x43 fill spare-pool extent
- 0x64+ 00000000 - - - zero to end terminator fill

The six live extents cover 129098 .. 129311 contiguously = 214 blocks (2 x C0 metadata + 212 x E0 spare-pool). VERIFIED (they tile with no gap or overlap).

4.4 Flag-bit meanings - what is proven vs OPEN

Only two flag values occur, each tied to a role by the block content:

  • 0xC0000000 marks the controller's own metadata blocks - the table (129311) and the parameter block (129310), and only those two. VERIFIED (association).
  • 0xE0000000 marks spare-pool / reserved extents at the top of the medium. VERIFIED (association).
  • bit 31 (0x80000000) = entry valid / in use (also the header's word[1]). A 0x00000000 word is an empty/null slot. INFERRED.
  • The two values differ only in bit 29 (0x20000000): clear = controller metadata (C0), set = spare-pool (E0). The precise per-bit naming (e.g. bit 30 = "allocated", bit 29 = "spare vs private", and whether other bits encode "defective/remapped") cannot be proven from one defect-free dump - no entry carries a bad-block or remap flag here. OPEN - the firmware (nd-scsi-3201.md reserved-block handling) or a dump with a real grown defect would pin bits 28..0.

4.5 Terminator - INFERRED

The live list is not terminated by the first all-zero slot: entry 5 (0x4C) is an all-zero null slot in the middle of the list, followed by a live entry 6. So the reader must not stop at the first zero. The table ends with zero fill to the end of the 1024-byte block. INFERRED reader rule: iterate a fixed slot count ((1024 - 16) / 12 = 84 slots), skipping null (flag == 0) slots, to end of block. Whether the firmware uses a fixed slot count or a header-stored entry count is OPEN (word[2] is 0 here, consistent with "reserved" or "count high word"); a disk with more entries would show whether word[2]/another field is a live count.

4.6 Two readings of the E0 extents - INFERRED / OPEN

Because this dump is defect-free, the E0... extents are consistent with either role, and one dump cannot separate them:

  • (a) Factory spare reserve - the top 212 blocks are set aside as an alternate block pool; extents are zero/fill because none has been consumed.
  • (b) Grown-defect reallocation records - each extent is a spare block already assigned to replace a filesystem defect; consumed top-down (129309 first).

The block contents lean toward (a) (the extents are zero or 0x43 fill, not relocated live data), but this is OPEN; a dump of a disk with a known reallocated sector distinguishes them.

5. The disk-parameter block (block raw-2)

Real bytes, SCSI-K.image LBA 129310, offset 0x07E47800:

07e47800: 001e 444d 4d00 0000 0000 0000 0000 0000   ..DMM...........
07e47810: 0000 4320 3032 07c5 0811 0f33 2020 0000   ..C 02.....3  ..
07e47820: 00 ... (zero to end)

Decoded (VERIFIED bytes, INFERRED field names):

Off Bytes Value Meaning (INFERRED)
0x00 00 1E 0x001E = 30 device type word - 0x1E is exactly the "floppy-style SCSI disk" device type the ND-3201 init handler at 0x0383 serves (nd-scsi-3201.md)
0x02 44 4D 4D "DMM" vendor / media tag (ASCII)
0x12 43 20 30 32 "C 02" revision string (ASCII)
0x16 07 C5 08 11 0F 33 1989-08-17 15:51 format/manufacture timestamp: 0x07C5=1989, 08=month, 0x11=17, 0x0F=15h, 0x33=51m
0x1C 20 20 " " padding

The device-type value 0x1E cross-checks against the firmware (nd-scsi-3201.md "Floppy-Style SCSI Init Handler (0x0383) - device type 0x1E"). This is the disk-parameter / geometry-identity block: it tells the controller/host what kind of unit and format this is. VERIFIED (bytes + the 0x1E firmware cross-ref); field names INFERRED.

6. Cross-device confirmation - this layer is SCSI-only

Device Image Size Last block Verdict
SCSI (ND-3201) SCSI-K.image 129,312 blk signed defect/realloc table (0x080054D9...) has the layer
SMD SMD0.IMG 76,800 blk all zeros no table
Floppy 250305L07-XX-01D.IMG 1,232 blk (616 pages) ordinary filesystem/loader data (contains the "IF HERE TYPE ANY CHAR" loader text + code) no table
  • SMD0.IMG last block (offset 0x04AFFC00) is entirely 0x00. VERIFIED.
  • Floppy 250305L07-XX-01D.IMG last block (offset 0x00133F00) is live boot/loader content, not a table. VERIFIED.

So the defect / reallocation table and the parameter block are managed by the ND-3201 controller / SCSI format tooling, not by the filesystem, and appear only on SCSI units. SMD and floppy media hand SINTRAN a medium whose last block is just part of the filesystem. VERIFIED.

7. How SINTRAN and the controller use this layer

  • Controller (READ CAPACITY) reports the raw last-LBA; the whole medium including the top region is addressable (Section 3). VERIFIED.
  • SINTRAN mount reads the last block (the control-record / function-42 connect) (scsi-mount-geometry.md): a highest-addressable-block presence/size probe. On the real disk that block is the defect table; the mount's carved path (ENDIR -> CHDSI -> RXDIR -> RCBLO) then reads block 0 for the directory. VERIFIED (carved).
  • Known emulation failure: when the emulator presents the raw disk but the controller drops the completion interrupt for that last-block probe (the RSTAU/RITRG bug), the mount stalls before block 0 - see SCSI-MOUNT-FIX-PLAN.md and ../../Filesystem/code-logic/scsi-mount-geometry.md Section 4-5. The disk content is fine; the blocker is interrupt delivery + geometry match, not the last-block bytes. VERIFIED (seam) / INFERRED (cause).

8. How to CREATE a valid ND SCSI disk image (retrofs)

8.1 Steps

  1. Pick geometry C x H x S. raw_blocks = C x H x S must be even so the medium is a whole number of 2048-byte ND pages (raw_blocks = 2 x pages). Example (matches SCSI-K): 898 x 8 x 18 = 129312 blocks = 64,656 pages. last_LBA = raw_blocks - 1.

  2. Set the reported capacity. The SCSI target must answer READ CAPACITY with last_LBA = raw_blocks - 1 and block size 1024. (RAW, per Section 3 - do not report the usable count.)

  3. Choose usable_blocks (even). Must satisfy usable_blocks <= raw_blocks - reserved_top, where reserved_top is the top region you reserve (at least 2 blocks: table + parameter block; SCSI-K reserved the top 214 as metadata+spare-pool and left 7,026 as slack). A minimal image can use reserved_top = 2 and usable_blocks = raw_blocks - 2.

  4. Write the filesystem into blocks 0 .. usable_blocks-1 per ../../Filesystem/create-directory-placement.md and ../../Filesystem/create-directory.md. The directory's pages-available must equal usable_blocks / 2.

  5. Fill the reserved slack (usable_blocks .. raw_blocks-3). Cosmetic: the real SCSI disk uses 0x43 ('C'); zeros are equally acceptable (SMD uses zeros).

  6. Write the parameter block at raw_blocks-2 (Section 5): device-type word (0x001E for a floppy-style SCSI disk), optional vendor/rev/date. A minimal writer may leave it zero only if the table does not point a C0... entry at it; to faithfully reproduce an ND disk, include it and mark it C0....

  7. Write the defect / reallocation table at the last block raw_blocks-1 (Section 4), big-endian:

    • Header: word0 = 0x080054D9 (copied signature - see caveat), word1 = 0x80000000, word2 = 0x00000000, word3 = usable_blocks.
    • Entries (12 bytes each, from offset 0x10):
      • C0000000, raw_blocks-1, 1 (the table block itself)
      • C0000000, raw_blocks-2, 1 (the parameter block) - include if step 6 wrote one
      • optionally one or more E0000000, start_LBA, run extents describing the reserved spare pool
    • Zero-fill the rest of the block.

8.2 Minimal defect-free table (all spare, no defects)

A brand-new, defect-free disk has no remapped-defect entries - only the controller-metadata C0... entries (and, to mirror ND, an E0... extent for the spare pool). Example for raw_blocks = N, usable = U:

offset  bytes (big-endian)                     meaning
0x00    08 00 54 D9                            signature (copied constant)
0x04    80 00 00 00                            flags: table valid
0x08    00 00 00 00                            reserved
0x0C    <U as 32-bit BE>                       usable block count
0x10    C0 00 00 00  <N-1 BE>  00 00 00 01     table block (self)
0x1C    C0 00 00 00  <N-2 BE>  00 00 00 01     parameter block
0x28    E0 00 00 00  <U   BE>  <(N-2-U) BE>    spare pool: everything between
                                               filesystem end and metadata
0x34..  00 ...                                 zero fill to end of block

That single E0 extent (step: start = usable_blocks, run = raw_blocks-2 - usable_blocks) declares the whole gap between the filesystem and the two metadata blocks as spare. SINTRAN reads word3 = U and is satisfied; the last block is readable (the mount probe passes).

8.3 What is minimally required vs faithful

  • Minimum for SINTRAN to mount (per scsi-mount-geometry.md Section 5): a page-aligned raw capacity, READ CAPACITY = raw last-LBA, and a readable last block. There is no required "label" content at the last block for the mount probe to pass - the probe checks presence/readability, and the size check reconciles capacity, not a signature.
  • Faithful ND SCSI reproduction (what tools that read the table expect): write the full header (real word3 = usable) + the C0.../E0... extents + the parameter block, as in 8.2. retrofs should do the faithful version so the image round-trips through any ND-3201-aware tool.

8.4 Open items a writer must be aware of

  • Signature word0 - copy 0x080054D9 as a constant. OPEN whether it is a fixed magic or a per-disk checksum; if a second dump shows a different word0, it is a checksum and its algorithm must be recovered before images with arbitrary geometry are byte-perfect. (word[3]=usable is the field readers actually depend on, and it is fully specified.)
  • Flag low bits (bits 28..0 of the C0/E0 flags) - unused/zero on this disk; set them zero. Their meaning is OPEN (Section 4.4).
  • Entry-count vs fixed-slot terminator - use zero-fill after the last entry and skip null slots on read (Section 4.5). Whether the firmware honours a stored count is OPEN; zero-fill is safe either way.

9. VERIFIED / INFERRED / OPEN summary

Claim Verdict
SCSI disk raw 129,312 blk vs usable 122,072 blk; reserved 7,240 blk VERIFIED (bytes + arithmetic)
READ CAPACITY reports RAW last-LBA (129311), not usable VERIFIED (RetroCore + real disk addresses LBA 129311)
Last block = defect/reallocation table; header 16 B + 12 B entries {flag,LBA,run} VERIFIED (decoded bytes)
Header word[3] = usable block count (122072 = pages x 2) VERIFIED
Header word[0] 0x080054D9 = signature (not plain sum/xor) VERIFIED value / OPEN meaning
Entry flags: C0... = controller metadata blocks, E0... = spare-pool extents VERIFIED (association)
Exact per-bit flag semantics; defect/remap bits OPEN (defect-free dump)
Null slot mid-list; terminator = zero-fill, skip null slots VERIFIED (bytes) / INFERRED (reader rule)
Block raw-2 = disk-parameter block (dev type 0x1E, vendor/rev/date) VERIFIED (bytes + firmware 0x1E cross-ref)
SMD & floppy have no such table (SCSI/ND-3201-specific) VERIFIED (both last blocks)
Firmware transparent LBA remap on read OPEN (no remap present to observe)
E0 extents = factory spare reserve vs grown-defect records OPEN (defect-free dump)

Provenance: real dumps SCSI-K.image (last block 0x07E47C00, parameter block 0x07E47800), SMD0.IMG, 250305L07-XX-01D.IMG; carved 006-S3FS SINTRAN L bytes and RetroCore SCSIHDD.cs via ../../Filesystem/code-logic/scsi-mount-geometry.md; ND-3201 firmware analysis nd-scsi-3201.md; floppy/streamer controller manual ND-11.021.1 EN-Floppy and Streamer Controller 3106 3112.md.

See also