Original Bugs & Oddities

The night-inn owner publishes ten controller bytes and ten callback/action cells even though $URBAN01 card 0 contains only nine source rows. The tenth control is permanently zero, so the generic controller cannot return ordinal 9. Nevertheless, its latent action is coherent: it presents card 0 in mode 0, runs the zero-hour calendar/event service, and publishes state 0004h, the ordinary $CITYE00 family. It contains no audio gateway.

This may be harmless fixed-width template residue, an abandoned tenth choice, or an accidentally omitted resource row. The rewrite keeps it absent. The latent destination must not be exposed unless a distinct original publisher is proved or an explicit enhanced-mode policy is adopted separately from faithful behavior.

$HANSE01.MSG contains one rowless card describing the nighttime Hanseatic League office. Nevertheless, the state-0047h owner publishes ten selectable rows and later card 6 from the thirteen-card $HANSE00.MSG corpus. The resource basename copied at owner entry, exact controller publication, and extracted MSG corpus all agree; this is not a missing $HANSE01 extraction.

The arrangement may be intentional day/night resource reuse rather than a defect. It is recorded because conflating owner and card-document identity causes unresolved-token failures and misleading row bindings. Clean code keeps $HANSE01.MSG as the owner resource and $HANSE00.MSG as the displayed card document. Tests must reject any future attempt to treat $HANSE01.MSG as the selectable multi-card document.

The complete ordinary owner writes row 0’s publication state to zero and never enables it. Nevertheless, its callback and local action form a coherent lane: signed selector parity chooses card 1 or 2, the Arms Outfitter opens, two hours pass, state returns to $MILCR00, and the existing audio cue is retained. A controlled forced-publication experiment confirmed both latent branches without establishing natural reachability.

This may be deliberately cut content, an unfinished alternate publisher, or an accidental omission. The rewrite keeps it hidden. Proving a distinct original publisher through the unresolved alternate dispatch would explain the apparent anomaly; absent that proof, neither the complete latent code nor the runtime experiment authorizes exposing it.

The owner permanently leaves row 4 unpublished despite substantive prose in the resource. Its callback returns the ordinary no-check sentinel, but its local action is only a one-byte RETF and produces no gameplay, presentation, calendar, or audio effect. Neighboring rows 5 through 8 are explicit literal placeholders and are catalogued separately rather than treated as completed content.

This may be an intentional stub or an abandoned feature rather than a defect. The rewrite keeps it hidden; a future reclassification requires either a distinct original publisher/action or external design evidence that explains the stub without overriding the byte-proven faithful behavior.

The complete owner permanently suppresses both literal placeholder rows, but each retains a callback and a substantive action. The two action bodies are byte-identical: they present council card 0, invoke the zero-hour helper, and publish state 0004h / ordinary $CITYE00. Row 7 additionally returns a selected-character threshold pair from its callback even though the action does not consume that argument. No audio gateway occurs.

This could be compiler/template residue, deliberately abandoned exits, or an unfinished pair of choices. The rewrite keeps both rows hidden and does not convert their latent destination into a useful route. A distinct original publisher would be required to explain the apparent anomaly and reopen their classification.

The council owner leaves row 5’s publication state at zero, so the generic controller cannot select it. This is despite a complete callback/action path that would negotiate guard training, create five kind-0028h training events, display success/refusal cards 12/13, create a kind-0041h cooldown, and return to the day/night city square.

The latent action also formats a dynamic displayed price, while the five training records contain one fixed 50-pfennig fee and four fixed 25-pfennig fees. Because the parent makes the route unreachable, controlled normal runtime cannot establish whether the hidden publication and price/fee mismatch are abandoned content, an accidental omission, or an unfinished feature.

The generic extended-key dispatcher maps BIOS words 5E00h..6200h to party slots 0..4, writes the chosen leader to byte DS:907Bh, and repaints the party panel. That is consistent with the quick-reference card’s “define leader” description.

The nested merchant controller intercepts the same keys. It validates the chosen active party slot, writes DS:907Bh, and repaints only the party panel. It does not call the full merchant redraw at proc_0032_078E, does not update the merchant transaction target at DS:884Eh, and does not recompose the displayed barterer text, inventory lists, or prices. Execution then falls through into ovl_0032:160F..1763, the same block used to switch Purchase and Sell, which flips the active list and redraws the two controls with their enabled/disabled colors.

The Goods Merchant Purchase arm checks affordability, calculates its payment price, and calls proc_0029_5090 to subtract that price from the three-word party purse before calling proc_0031_4A04 to insert the item. The inventory owner first tries to merge the item into an exact code/type/quality stack. If no compatible stack accepts the additional quantity, it may append a new six-byte record only while the inventory count is below 40h.

At count 40h, a failed merge returns the unchanged count without inserting the purchased item. The Purchase caller does not test that return for failure and does not refund the earlier purse subtraction; it stores the returned count and redraws the held-item list through proc_0032_1770. The resulting observable behavior is therefore:

The original action calls proc_0029_1326(0005h) and displays $URBAN00 card 1 when the result is nonzero, otherwise card 7. Selector 0005h returns nonzero only when every active party record owns a nonzero-quantity item whose definition word at +20h intersects mount flag 2000h. Therefore:

Both cards say that boarding is included in the inn charge and introduce the Stablemaster’s available horses and mules. Card 7 alone adds that the Stablemaster asks whether any of the party’s mounts are for sale. This is counterintuitive: the extra sale inquiry appears even when nobody owns a mount, while it is omitted when every active member owns one. The same shared merchant controller follows either card and allows selling in both cases.

The owner accepts either an indexed current location in DS:907Eh or the sentinel FFFFh, which selects the coordinate pair at DS:E3D0h:E3D2h. After the shared value-0004h prefix and the location-specific value-0002h search fail to choose a final card, the two modes test a due value-000Ch event with opposite fallback results:

Thus, with no due value-0002h, value-0004h, or value-000Ch event, an indexed location displays $AFFAI00 card 3 while the coordinate fallback displays no final card. With only value-000Ch due, the indexed route displays the card-17 prefix and no final card, while the coordinate route displays card 17 followed by card 3. Exact bytes, Ghidra, and independent Reko output agree on this asymmetry. Raw owner tests preserve both sides, including negative cases that prevent the coordinate arm from inheriting the intuitive indexed fallback.

The owner searches for kind 0008h with field1A=DS:E896h and field28=0003h. It stores the returned slot but never compares it with the FFFFh not-found sentinel. Instead it immediately multiplies the slot by four, indexes the event-pointer table, reads event field +00h, and passes that word to proc_0029_0DB4.

If every legal state-0090h publisher has already created this exact record, the omission is a harmless original invariant. If state 0090h can be reached without it, the FFFFh index wraps and reads an unrelated pointer cell. The same unchecked pattern begins the paired state-0091h proc_0080_038C completion owner for its required kind-0024h record.

The action treats high local reputation as a guaranteed-recruitment tier. If fewer than five party slots are occupied, it shows card 4 and constructs the Schulz companion. If all five slots are occupied, it instead shows card 6, whose prose says that the Schulz drops one florin into the party’s hand.

That full-party branch jumps directly to the common cooldown suffix and never calls proc_0029_5040; the purse is unchanged. The same card 6 is also used by the low-reputation fame-weighted success branch, but that separate branch does call proc_0029_5040(DS:906Bh,0001h,0000h,0000h) and therefore really adds one florin. Result-card identity is consequently insufficient to infer the money effect.

The $RAUBI00 offer allows the player to accept or decline a temporary Schulz companion for origin 0086h, or a Hanse guard/knight for origin 0047h. Both actions return through the same inherited origin value. Decline only writes DS:E48Ah=DS:E7D8h; it does not mark the quest event with a distinction later read by $RAUBI01.

At completion, $RAUBI01 selects card 1 unconditionally for origin 0086h. That card says the Schulz departs from the party. For origin 0047h, the owner draws RNG(2) and selects card 6 on a nonzero result or card 7 on zero. Card 6 says the Hanse friend leaves and card 7 does not. The completion owner does not inspect party membership, does not inspect an acceptance flag, and contains no party/roster-removal call.

The owner reads the zero-based current month from DS:00E4h. It selects frozen-water card 1 and disables the two swimming variants when the signed month is at least 10 or at most 4: November through December and January through May. Only June through October selects ordinary card 0. The predicate does not read the current hour.

Card 1 explicitly describes chunks of ice in the water, making its use through all of May geographically and seasonally surprising. This may be an intentional broad cold-water season rather than a defect, so the catalogue does not classify it as a confirmed bug. The faithful projection preserves the exact month boundary; it must not substitute a modernized climate rule.

At night, proc_0049_0000 skips the copy that would replace the active MSG name with $BUSIN00. The inherited $SIDES01.MSG card-0 text therefore remains visible over XNMAIN.PIC. The same owner then publishes the ten state-0014h callback and action cells. Those handlers are not the proc_0046_1132 $SIDES01 handlers: a displayed row is acted upon according to its coincident numeric ordinal in the state-0014h table.

The mismatch can make a label describe one destination while its action goes somewhere else. It appears to be a missing night-resource replacement rather than intentional content, but runtime confirmation of the exact presentation and several selections is still required before classifying it as a confirmed defect. The rewrite must not use the displayed document name as dispatch authority or silently substitute $BUSIN00 text.

Both action outcomes call proc_0029_54D8(0001h). The ordinary outcome then calls proc_0029_563A and publishes state 0021h by day or 0022h by night. The rare outcome does not call the classifier: it always writes DS:E7D8h=0022h before publishing state 003Ch, even when its one-hour advance ends in daytime.

The retained value may be intentional history for the unresolved state-003Ch owner rather than a defect. Until that destination owner is decoded or a controlled runtime capture proves how it consumes E7D8h, the rewrite must not replace the fixed value with a day/night classification. The selected ordinary grove route is independently certified, while this rare branch remains fail-closed.

The adjustment owner computes the second RNG range as abs(upper)-abs(lower). Equal bounds therefore pass zero. The original RNG wrapper uses divisor 2 for both range 0 and range 1, so RNG(0000h) returns either 0 or 1. With lower and upper both -1, an accepted score gate therefore subtracts either one or two points. A rejected gate still consumes only its RNG(0064h) draw and changes nothing.

This may be a deliberate general RNG convention rather than a gameplay bug. Do not normalize the overnight loss to a fixed value without broader caller analysis. The reconstruction preserves the byte-proven result and RNG order.

The extracted MSG payload gives encoded rows 2 through 7 only the literal texts 2 through 7. The owner initializes all six corresponding control bytes to zero and never enables them. Latent action entries still exist and each would perform a zero-hour calendar call before publishing state 0062h, but the ordinary controller cannot select them.

This looks like unfinished reserved content, although it could be an intentional fixed-size table. The rewrite must neither expose the placeholder rows nor infer destinations from their latent actions. If a different owner or external writer is later proven to publish those controls, this entry must be updated rather than silently removed.

Selector 0001h uses RNG bound 0058h, making exactly 88 consecutive female first-name strings selectable. The same contiguous resource area continues with Suzanne and Viktoria, but neither can be selected through this owner. Selector 0002h similarly uses bound 0010h over exactly 16 family-name strings; the immediately trailing Welser entry is outside the draw.

This may be deliberate shared-table packing rather than a defect, and no claim is made that these names were intended for the two $OTHER00 substitutions. The rewrite must retain the byte-proven bounds and must not broaden the pools merely because additional valid strings are adjacent. A future caller that selects the trailing entries would revise this classification.

The 483.07 data ships nineteen consecutive $REVOLxx resources with 230 cards. Their prose forms a large, coherent revolution/counter-revolution quest covering political investigation, recruitment, arrests and rescues, timed summons, crowds, marches, sieges, relief forces, and both rebel and government victories. This is substantially more authored content than the three family resources currently encountered by the reconstructed city route.

The complete daytime entry owner initializes rows 5–8 to zero and never publishes them. Row 5 is the substantive Force-entry choice; rows 6–8 are blank placeholders. The complete nighttime owner likewise leaves rows 5–8 at zero, where they are literal numeric placeholders. Callback/action table cells exist for these zero-state rows, but their latent effects remain an explicit decode boundary. Table existence cannot make a row selectable.

After a successful signed RNG check, nonzero DS:E3D0h publishes cards 2 and 12 before the selector-0029h result is considered. The selector then always publishes card 2 again followed by card 3 when its result equals 0002h, or card 4 otherwise. The two exact nonzero-lane sequences are therefore 2,12,2,3 and 2,12,2,4; card 2 appears twice. The zero lane omits the prefix and publishes only 2,3 or 2,4.

The repeated card may be deliberate recap after the lane-specific card 12, or it may be an original presentation bug. Do not collapse, reorder, or deduplicate the sequence without evidence from another original version. In particular, card 12 depends only on nonzero DS:E3D0h, while the choice of card 3 versus card 4 depends only on the selector result; neither predicate may be substituted for the other.

After the one-hour calendar call and signed RNG check select the complementary lane, the owner calls proc_0029_3742(0002h,0000h,current-city) once. If the result is FFFFh while DS:E3D0h is zero, it immediately takes the no-card state-0079h tail. Every other case calls the same helper again with the same three words. A second result of 0001h suppresses card 5 only when DS:E3D0h is nonzero; otherwise the owner publishes card 5 and state 0074h.

The selected helper scans the same live kind-001Ch event records and writes the selected slot index to DS:E3CEh; it does not mutate the matched record. Consequently the second result is ordinarily identical to the first, and the second scan appears redundant. It may be compiler output from two separately authored selector predicates, intentional refresh of DS:E3CEh, or an unfinished design in which another operation once separated the calls. Exact bytes do not prove which explanation is correct.