CareerSession gets search_profile, suggest_search_profile and dispatch_search_profile (owner and
modal-state checks, revision only when the stored profile changed). The console replaces the
OpenSearch placeholder with three commands and documents them; with apply the matching listings
are reported with the original's "Spieler ... ist interessant!" message. search_profile_valid
separates a refused profile from a switched-off one.
Part of #79 (the dialog in the graphical client is still open)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
New game/search_profile.* holds the 18-byte search profile of a human manager (manager record
+0x1f6): the dialog's closing rule, the suggestion generated from the squad (original sub_18F06),
the position strengths (sub_19A38) and the listing match (sub_190B5). search_profile_news in
month.cpp now uses the same matching rule. Compared with the Diamond 3.7 routines in the
emulator harness on 10 savegames (306 runs, 406 matched listings, no difference; a deliberately
changed profile is detected).
Part of #79 (the dialog in the graphical client and the console commands follow)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Strength of the new player is the youth-team strength setting, the academy is
skipped when the squad's youth average is more than 25 above it, up to two rounds
need W(0, 3) == 0, and the retirement/retire rules use the 3.7 slot range. The
age test was inverted: only a player older than the roll W(t, t + 3) is recycled
(found by comparing with the 3.7 programme in an emulator, 192 runs on whole
savegames, no difference). Removes the old strength-band deviation entries.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
kFacilityValue (entries 1, 2, 5), the per-visitor facility income table and the
season-ticket price of the first division. Checked against the 3.7 programme in
an emulator on whole savegames: weekly statistics now match without difference.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Credit limit 8 million, fee base 10000 per 10000 of value, DFB fee 150 %,
championship bonus 700000, club-popularity premiums 400000/850000/1000000/850000,
periphery prices, advertising base 0x87, start capital step 3 million, the
broke limit -4 million, the training-injury chance, retirement after an injury
of more than 20 matchdays (1 in 40), the contract-talk limit and the side
rhythm (human: no balance term, computers: at least 48). Values read from the
3.7 programme; the earlier build the engine was derived from used other numbers.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Injury chance 1 in 31 (was 1 in 39), no-booking roll W(0,3), straight red 1 in 34
(was 1 in 28), the injury type thresholds (match injuries now types 3-30), the
duration table, 1/4 longer for a player injured before (was 1/8) and the club
doctor's cut (/8, capped only for the heavy types 20-30). Values read from the
3.7 programme, measured against the earlier build the engine was derived from.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Every source file in ui, app, game, formats, cli, audio and launch that had none
now starts with a short English comment: what it implements, what state it owns
and how it relates to its neighbours (request/acknowledge flow, pairs of files,
random generator use). Comment lines only; the full build and the runnable
tests are unchanged.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The original computes ((base / 100) * den) / 100 with 64-bit-mantissa x87
steps and one final store to a double. The host long double is only 64 bits
wide on MSVC and Apple silicon, so the annuity differed by 1 for roughly 0.2 %
of realistic loans (1-60 million) there. The chain is now evaluated with
integers, identical on every platform. Verified against the original routine
in an emulator: 5684 term/principal combinations, no difference.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Use 64-bit products in actual monthly fund/property settlement, monthly and annual previews, and weekly dividend offers. Retain per-unit rounding, recipient policy and final native fields. Test representable high-value settlements against their forecasts. Full 32-bit money storage overflow remains unresolved.
Refs: #46
UndoReschedule of the original clears bit 0x4000 of the calendar day
(high byte & 0xbf) when no replay of another club stays on it; the
engine kept the bit, so the day and its neighbour were no longer offered
as targets after a move was taken back. The test checks the same target
days after the undo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The six texts the engine posts from the game image (contract warning,
Masters, indoor entry bonus, league title bonus, friendly share) were
set only by the career sessions. week_check and the other standalone
tools left them empty, so the indoor entry bonus mail held only the
venue and the replays failed from the first public commit on.
install_message_texts(exe) now sets them for both sessions and in the
test executable loader.
Classic regression: the ten turns of #14 (game 1: 18, 21, 65, 88, 109,
201, 247; 2P: 24, 29, 34) pass again; no other turn changed against the
0.3 M0 baseline. The divergence the issue saw in the youth teams block
(bd4bc, since 310e3e2) is among the tolerated block differences.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>