Karta "Dokoncene hry" (%) nahradena dvomi poctami dokoncenych hier
podla toho, ci na niektorom sedadle sedel bot (username "bot:...").
Zaroven oprava serializacie avg_game_duration_minutes -- Postgres
vracia z func.avg Decimal, ktory json.dumps nevie serializovat (500
na /api/admin/stats).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Bot doteraz vedel zahodit prvu kartu novej kopky uz po BOT_MOVE_DELAY_SECONDS
(0.8s), zatial co frontend zmetaciu animaciu predoslej kopky prehraval
1650ms -- karta tak "vyletela" uprostred zmetania. _run_bot_turns teraz pri
vedeni novej kopky caka TRICK_SWEEP_SECONDS (1.7s), zhodne s SETTLE_MS+
COLLECT_MS v GameTable.tsx.
_run_bot_turns tiez posielal player_cards PRED game_status (opacne ako
human play_card handler) -- pri prechode do noveho kola tak klient dostal
novu ruku skor, nez vedel, ze zacalo nove kolo, a fixne ju hned zobrazil.
Poradie broadcastov je teraz zhodne s human handlerom.
GameTable/Hand: ruka noveho kola sa zobrazi az po dobehnuti zmetacej
animacie poslednej kopky predchadzajuceho kola (displayedHand + freeze
efekt), a Hand.tsx rezervuje fixny priestor pre karty aj ked je ruka
docasne prazdna, aby layout neposkakoval.
tests/test_bots.py: nulovanie TRICK_SWEEP_SECONDS v setUpClass, aby testy
zostali rychle.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Boti su sedadla bez socketu: ucty bot:<kind>-<n> v tabulke players
(nehijacknutelne, recyklovane medzi hrami), handlery add_bot/remove_bot
(len hostitel, pred startom) a tahova slucka _run_bot_turns s pauzou
BOT_MOVE_DELAY_SECONDS a MC/inferenciou v executori. Druhy: heuristic,
random, neural (pure-Python siet; bez suboru vah jasna chyba a pri
restore fallback na heuristiku). Botie sedadla preziju restart servera.
Oprava po ceste: emit kariet hracovi so sid=None (offline sedadlo po
restore) broadcastoval jeho karty vsetkym klientom -- preskakuje sa.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
py -m rl.export vyexportuje checkpoint do rl/weights/neural-bot.json
(bit-exact float32, 1.3 MB) a rl/pure_net.py ho hra bez torch/numpy
(stdlib forward pass, ~16 ms/tah). Testy parity: logity aj akcie sa
zhoduju s torch, identicke trajektorie celych kol. Natrenovany model:
6.7-6.9 b/kolo proti vsetkym baseline-om (heuristika prekonana).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
BridzikNet (trup + guess/play/value hlavy s maskovanim), self-play
generator so zdielanou sietou na 4 sedadlach a opponent mixingom
(random/heuristicke sedadla pre robustnost), vlastny clipped-PPO
so skalovanim odmien a lr/entropy annealom. Torch je len trenovacia
zavislost na hoste (requirements-rl.txt); checkpointy a logy su
gitignorovane. Spustenie: py -m rl.train.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Vlastne __eq__ rusilo zdedeny __hash__, takze karty ani enumy sa nedali
pouzit ako kluce dictov a v setoch. Hash doplneny konzistentne s __eq__.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Ucet bez potvrdeneho kodu (auth_token is NULL a totp_last_step == 0)
uz neblokuje meno: opakovany register_account vyda novy secret (stary
QR prestane platit) a login vyhodi RegistrationIncomplete, na ktoru
server odpovie novym QR -- klient sa prepne na registracny tab.
Admin statistiky vykazuju nedokoncene registracie osobitne, Hraci
celkom pocita len potvrdene ucty. Novy event register v analytike.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Na /auth sa da dostat len plnym loadom stranky (vsetky interne
redirecty nan su REPLACE a beacon ich skipuje) a kazdy plny load uz
posiela event landing. Kazdy riadok /auth by tak mal dvojicku z toho
isteho loadu. Skip na frontende aj v _SKIPPED_PATHS (pokryje aj stare
cache-ovane verzie frontendu). Stare /auth riadky v DB ostavaju.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Navsteva zacinajuca na / (preklik z FB/Messengeru, natukanie domeny)
doteraz nezanechala v page_views ziadny zaznam: / je v skip liste
beaconu a gate redirect na /auth je REPLACE navigacia, takze sa skipli
oba hopy a referrer sa stratil. trackLanding() v main.tsx posle raz za
kazdy plny load stranky pomenovany event landing s document.referrer.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pageview beacony uz neposielaju player_id z klienta (nedovereny vstup na
neautentifikovanom endpointe) -- prihlasenie sa eviduje server-side ako
event "login" (aj z registracie), s player_id skutocneho uctu. Admin
dashboard dostal prepinac scope vsetci/prihlaseni, ktory konzistentne
pocita grafy aj tabulky (kazdy login samostatne, navstevnicky den pre
anonymnu navstevnost). PageView.country teraz uklada cely anglicky nazov
krajiny namiesto ISO kodu (potrebna zmena schemy). Pridane sledovanie
klikov na "Pravidla hry" aj z GameList.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Per-username lockout (5 failed TOTP attempts / 5 min) stops account-targeted
brute force regardless of source IP. Player.totp_secret is now Fernet-
encrypted (ENCRYPTION_KEY env, db/crypto.py) instead of stored in plaintext,
and auth_token is stored as a SHA-256 hash rather than the raw session token.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Records pageviews (path, referrer, browser/OS/device, IP, GeoIP country) via
a POST /api/track beacon into a new PageView table, and exposes aggregated
daily/breakdown stats behind a token-gated GET /api/admin/stats endpoint with
brute-force lockout. Frontend gets a /admin dashboard (charts + breakdown
tables) built on recharts, with a switchable per-day pageviews chart.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Remove config.py, an unused Flask SECRET_KEY leftover from before the
legacy HTTP backend was replaced by the Socket.IO/ASGI server.
- Move tests.py / tests_history.py / test_socket.py into a tests/
package as test_engine.py / test_history.py / test_socket.py, and
update CLAUDE.md's documented commands to match.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
timandClaude Opus 4.8