New module bot/notify/telegram.py — httpx async, fire-and-forget via
asyncio.create_task, swallows all errors so notifications never affect
trade execution.
Three alert types:
📈/📉 TRADE ABIERTO — direction, size, edge_net (in execute())
✅/❌ GANADO/PERDIDO — approx PnL (in close_position())
🔒 LEGACY CLOSE — recovered capital + reason (in close_legacy_position())
close_position() and close_legacy_position() gain an optional question=""
param so the message shows the market name instead of market_id.
bot/main.py updated to pass question= to close_legacy_position().
Credentials (TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID) read from env vars
injected via bot-secrets k8s secret.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Verify all 3 images exist in Gitea registry via Docker API before updating manifests
- Validate YAML of modified manifests after sed (python3 yaml.safe_load)
- Notify Telegram on success/failure with job status (if: always())
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Legacy builder (DOCKER_BUILDKIT=0) cannot handle OCI image indexes from
registry-cache, causing fallback to Docker Hub which is unreachable.
BuildKit sends proper OCI Accept headers and reads buildkitd.toml to use
HTTP for both the registry-cache mirror and internal Gitea registry.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Docker in DinD cannot reach git.chemavx.xyz (Cloudflare) from within
the cluster — TCP :443 times out. Switch docker login/build/push to
gitea.gitea.svc.cluster.local:3000 (insecure, same backend storage).
k8s manifest updates still reference git.chemavx.xyz for node pulls.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Docker 24's embedded BuildKit ignores the http:// prefix in registry-mirrors
and always attempts HTTPS, breaking the local pull-through cache.
DOCKER_BUILDKIT=0 uses the legacy builder which respects the daemon mirror
config correctly. Cache still works via --cache-from + buildcache tag.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The docker-container buildkitd driver creates an isolated process that
cannot use DinD's registry-mirror config, and the cluster's registry-cache
returns 500 on BuildKit's ?ns=docker.io mirror protocol.
Plain docker build routes through the DinD daemon directly, which already
has registry-mirrors configured for docker.io pull-through cache.
Uses BUILDKIT_INLINE_CACHE=1 for layer caching between builds.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
BuildKit docker-container driver runs isolated from DinD daemon config,
so it needs its own mirror declaration to route docker.io pulls through
the cluster-local pull-through cache instead of Cloudflare CDN directly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds alpha attribution by dominant signal feature — which feat_*_lo had
the largest absolute log-odds value on each trade.
Changes:
- _dominant_feature() helper in api/main.py: picks the winning feature
from signal_components (threshold 0.0001, same as "triggered" in
/api/metrics/features)
- _enrich_trade() refactored to single exit point; adds dominant_feature
field to every open trade in /api/trades
- compute_attribution_from_db() in db.py: VALUES subquery finds dominant
feature per trade in SQL, then aggregates trade_count/avg_edge_net/
unrealized_pnl_est/realized_pnl/resolved_count/win_rate per group
- /api/metrics/attribution endpoint: returns attribution dict + total_attributed_trades
No schema changes, no strategy changes. Pure observability.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Eliminates the phantom 0.4% exposure overage after pod restarts.
During live trading execute() stores size_usdc in portfolio.positions and
deducts net_cost from cash — so total_value = bankroll − fees and
exposure_pct = sum(size_usdc) / (bankroll − fees).
Old initialize() stored net_cost in positions, making total_value = bankroll
and inflating exposure_pct (observed: 30.085% vs runtime 29.670%).
Fix: new get_open_position_data() returns both {market_id: size_usdc} and
total_net_cost in one query; initialize() uses size_usdc for positions and
total_net_cost for cash — identical model to execute().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds feat_fg_lo / feat_mom_lo / feat_news_lo / feat_mfld_lo / feat_btc_dom_lo
to every trade, all normalized to log-odds contribution for direct comparability.
- fg / mom / btc_dom: raw probability-delta × 2 → log-odds
- news / mfld: already log-odds (LOGODDS_WEIGHT already applied), no scaling
- btc_dom tracked separately in bayesian.py instead of bundled in total_adj
- reasoning string updated to fg_lo= / mom_lo= notation for self-documentation
Schema: 5 new DOUBLE PRECISION columns + 2 partial indexes
Stack: TradingSignal → Order → Trade → save_trade all carry feat fields
Startup: backfill_feature_columns() recovers fg/mom/news/mfld from old
reasoning strings (×2 applied to fg/mom); btc_dom_lo stays NULL for legacy
API: /api/metrics/features — triggered/material split per feature with
two-level thresholds (0.05 for fg/mom/btc_dom, 0.10 for news/mfld)
API: /api/trades/legacy — exposes pre-Phase-1 trades (edge_net IS NULL)
API: _enrich_trade backward-compat: reads DB columns first, falls back to
reasoning regex with unit conversion for pre-Phase-6 trades
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
schema.sql
trades: + close_pnl, resolution (market outcome storage)
metrics_daily: + unrealized_pnl_est, realized_pnl, open/closed/resolved_count
db.py
close_paper_position(): accepts resolution; computes close_pnl in SQL
BUY_YES: (resolution − entry_price) × shares
BUY_NO: ((1 − resolution) − entry_price) × shares
save_daily_metrics(): persists new columns
compute_metrics_from_db(): single DB query for all metrics; no in-memory state
tracker.py — complete rewrite (stateless)
Removed self._trades, self._daily_returns, compute_metrics(), _compute_sharpe(),
check_promotion_thresholds(), _empty_metrics()
update_daily_summary() now reads compute_metrics_from_db() every cycle
Safe across pod restarts: always reflects full DB history
paper.py
close_position(): passes resolution to close_paper_position()
api/main.py /api/summary
Added unrealized_pnl_est (estimated, open trades) and realized_pnl (exact,
closed+resolved) as separate fields alongside total_pnl
win_rate: null if < 5 resolved trades (was proxy on entry_price < 0.5)
calibration_score: Brier-based, null if < 10 resolved trades
resolved_count exposed as field
Each field annotated with: exact/estimated, source, null conditions
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Kaniko has its own HTTP client and ignores the dind --registry-mirror flag.
Passing --registry-mirror=mirror.gcr.io directly to kaniko lets it pull
python:3.11-slim (and other Docker Hub base images) via Google's mirror
instead of the Cloudflare CDN endpoint (172.64.66.1) which is unreachable
from the runner's dind network.
Also adds mirror to dind deployment args (--registry-mirror=https://mirror.gcr.io)
so node:16-bullseye (the job runner container) is also fetched via mirror.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
total_pnl now uses edge_net × net_cost instead of (0.5 - entry_price) × shares.
The old formula overestimated BUY_NO trades at low entry prices by 3–10× because
buying at price 0.158 yields 3164 shares — any exit-at-0.5 assumption produced
$1072 PnL on $500 deployed. edge_net × net_cost is bounded by net_cost per trade
and uses the model's own signal, giving $122 for the same position.
calibration_score is now None (null in API) instead of 1 - 2×|avg_edge|. That
formula was not a real calibration: it requires knowing market resolutions
(YES=1/NO=0) which we do not store yet. Returning null is more honest than
returning 0.0 or a meaningless proxy. Fix 3 will compute it from closed trades.
check_promotion_thresholds updated to handle None calibration (null → not ready).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- db: update_family_key() persists corrected family slugs for open trades
- db: get_recently_closed_inverted() returns markets closed for inversion
within N hours; used as reentry guard in the trading loop
- db: get_recent_trades() accepts status=open|closed|None and adds a
computed "status" field to every row
- bot/main.py: legacy scan now computes family_key from stored question
alone (dummy Market) when a position's market is no longer active —
fixes NULL family_key on legacy trades like Ken Paxton (562186)
- bot/main.py: legacy scan (Step 2.5) persists corrected family_keys in
DB so family conflict guards work correctly on next restart
- bot/main.py: positions with NULL edge_net and no live market are tagged
legacy_incomplete instead of OK; counted separately in scan summary
- bot/main.py: reentry_guard blocks re-entering any market closed for
inversion bug within 24h; logs reentry_guard_triggered per skip
- api/main.py: /api/trades now accepts ?status=open|closed|all (default
open) and includes status_filter in response
DB fix (applied directly): 629558 family_key politics-2026 →
ohio-gubernatorial-2026; 562186 family_key NULL → texas-republican-2026
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds run_legacy_scan() that executes once at startup before the trading loop:
1. Re-keys every open DB position using the current market_family_key()
2. Groups by new family key; KEEP = highest edge_net, CLOSE_RECOMMENDED = sibling
3. Manifold re-query for positions whose family key changed; if corrected
probability contradicts the trade direction → CLOSE_RECOMMENDED
4. Logs full report (KEEP / REVIEW / CLOSE_RECOMMENDED) before any closures
5. In paper mode: auto-closes all CLOSE_RECOMMENDED positions
For the existing Ohio bug:
- Democrats win Ohio governor (629557): CLOSE_RECOMMENDED
family changed ohio-democrat-2026 → ohio-gubernatorial-2026
Manifold re-query confirms prob=0.05 contradicts BUY_YES (inversion bug)
$X returned to cash at break-even
- Republicans win Ohio governor (629558): KEEP
higher edge_net (0.349 > 0.247)
Infrastructure:
- schema.sql: closed_at TIMESTAMPTZ, close_reason TEXT on trades
- db.py: all open-position queries filter WHERE closed_at IS NULL
+ close_paper_position(market_id, reason)
- paper.py: close_legacy_position(market_id, reason) → float
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Republicans (plural) previously didn't match _PARTY_RE because the pattern
was r"\bRepublican\b" (no optional s). Added Republicans? for symmetry with
Democrats?. The general-election family fix already handles this case via
etype_m, but the plural match is needed for the party-only fallback branch.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
FASE 1 — market_family_key() general election fix
General elections now group by office, not by party, so complementary
markets ("Republicans win Ohio governor" / "Democrats win Ohio governor")
share the same family key (ohio-gubernatorial-2026). The second market
is blocked by the occupied_families check rather than traded as independent.
Primaries still keep the party (texas-republican-2026) because each party
runs its own separate primary race.
FASE 2 — Manifold party inversion guard
_detect_party() identifies the winning side in both the Polymarket question
and the matched Manifold title. If they are confirmed opposites (republican
vs democrat), the probability is inverted (1 - prob) before use.
Full audit log per query:
poly_question / manifold_title / manifold_url / match_score /
prob_raw / inverted / prob_final
Root cause of Ohio Manifold:0.95 on both sides: both queries matched the
same Manifold market ("Republicans win Ohio governor" prob=0.95). For the
"Democrats win" query the inversion now produces prob_final=0.05 instead of
blindly applying 0.95 to the wrong direction.
FASE 4 — startup contradiction scan
get_open_position_details() added to db.py. main.py checks all open
positions at startup, warns on any family with >1 position, and recommends
keeping the one with the highest edge_net. No auto-close.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signal 5: ManifoldClient queries Manifold Markets API for a matching binary
market by keyword overlap (threshold 0.25) and applies a log-odds adjustment
proportional to the divergence from the Polymarket prior.
manifold_log_adj = (log_odds(manifold_prob) - log_odds(prior)) × 0.6
A 30pp divergence (Manifold 0.75 vs Poly 0.45) produces edge_gross ≈ 0.19,
clearing the politics far-horizon regime_min=0.12 after costs. Confidence
boosted +0.08 when Manifold match found.
Per-feature observability: every SKIP_EDGE_NET and TRADE log line now includes
fg=±X.XXX mom=±X.XXX mfld=±X.XXXX news=±X.XXXX
so the contribution of each signal to edge is auditable per market.
Files: bot/data/manifold.py (new), bot/strategy/bayesian.py, bot/main.py
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 1 — Edge neto real (paper.py, bayesian.py, risk/manager.py, db.py):
- Trade records now store edge_gross, edge_net, prior_prob, final_prob,
mid_price, spread_estimate, commission, family_key
- edge_net = edge_gross - SPREAD_ESTIMATE(0.02) - COMMISSION_RATE(0.02)
NOTE: both constants are heuristics, not exact Polymarket exchange costs
- Execution gate changed from edge_gross > MIN_EDGE to edge_net > regime_min_edge
Phase 2 — Market families (polymarket.py):
- market_family_key(market) groups related markets:
texas-republican-2026, fed-april-2026, openai-2026, etc.
- At most 1 trade per family per cycle; occupied_families propagated via main.py
- Family key logged on every TRADE and SKIP line
Phase 3 — GNews priority (news.py, bayesian.py, main.py):
- NewsClient.get_freshness() returns 1.0/0.75/0.40/0.10 by cache age
- gnews_priority(market, news) = uncertainty × volume_score × freshness
- Politics markets sorted by priority DESC before eval so best markets get
the 5-query/cycle GNews budget first
Phase 4 — Regime min-edge by category/horizon (bayesian.py):
- politics >60d → 0.12, 30-60d → 0.10, <30d → 0.08
- tech / crypto/finance → 0.10
- All thresholds applied to edge_net (not edge_gross)
Phase 5 — Observability (bayesian.py, main.py):
- Structured skip labels: SKIP_UNSUPPORTED, SKIP_NO_SIGNALS,
SKIP_PRIOR_EXTREME, SKIP_FAMILY, SKIP_GNEWS_PRIORITY, SKIP_EDGE_NET
- TRADE lines now include family_key, edge_gross, edge_net, regime_min, days
- schema.sql: 8 new cols on trades, 7 new cols on signals (via ALTER TABLE IF NOT EXISTS)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add European football leagues (La Liga, Premier League, Bundesliga, etc.)
to _SPORTS_EXCLUSIONS so those markets are filtered before category detection
- Reorder _detect_category: check tech before crypto/finance so company-specific
markets (OpenAI IPO, NVIDIA, Apple) resolve to "tech" instead of "crypto/finance"
- Widen resolution horizon default from 60 to 90 days to surface more
markets in the 0.08–0.92 uncertainty zone
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Reconstruct dashboard from compiled container: App.jsx, main.jsx, index.css
- nginx.conf with SPA routing and /api proxy to api:8000
- Multi-stage Dockerfile: node:20-alpine build + nginx:alpine serve
- Add third kaniko build step in ci.yml for chemavx/polymarket-bot-dashboard
- Update k8s manifest sed to patch deployment-dashboard.yaml image on each push
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add _SPORTS_EXCLUSIONS list checked first in _detect_category so NBA/NFL/
MLB/NHL/tennis/golf/UFC/boxing/wrestling/tournament markets never bleed into
politics or events categories. Also removes 'super bowl' from _EVENTS_KEYWORDS
since it's now covered by the sports exclusion.
Keywords excluded: nba, nfl, mlb, nhl, basketball, football, baseball, hockey,
soccer, mvp, rookie of the, championship, super bowl, world series, playoffs,
playoff, tournament, tennis, golf, ufc, boxing, wrestler, wrestling,
slam dunk, home run, touchdown.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Markets where Polymarket consensus is near-certain leave no room for our
signals to generate MIN_EDGE=0.10 — evaluating them wastes GNews quota and
produces noise. Filter them out early with a clear log reason.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
vaderSentiment==3.3.2 added to requirements.txt.
_score_headlines now:
- scores each article (title + description) with VADER compound ∈ [-1, +1]
- filters out articles with |compound| ≤ 0.05 (no clear signal)
- weights remaining articles by recency (GNews newest-first, rank 0 → highest weight)
- returns weighted mean clamped to [-1, +1]
Removes the custom keyword sets (_POSITIVE/_NEGATIVE) and the set-based
bag-of-words algorithm that capped scores at ~±0.5 in practice.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- CACHE_TTL: 4h → 6h (≤36 req/day with ≤9 politics markets)
- GNews only called for is_politics markets (BTC/F&G cover crypto/macro)
- MAX_NEWS_QUERIES_PER_CYCLE=5: BayesianStrategy.reset_cycle() called each
iteration; counter increments only on actual API call (cache hits free)
- 2s asyncio.sleep in news.py finally block after each real HTTP request
- main.py sorts markets: politics first by end_date ascending, so soonest-
resolving markets consume the 5-query budget before others
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Drop the 'from' date filter — it's a paid GNews feature, causes 403 on free tier
- Add User-Agent header to httpx client; urllib default passes, httpx default blocked
- Log actual HTTP status code for every request (INFO) and response body on non-200
- Cache neutral result on 400/401/403/429 to avoid hammering the quota
- Remove unused _iso_days_ago() helper and 'days' param from get_sentiment()
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
bot/data/news.py (new):
- NewsClient with in-memory cache (TTL=4h) to stay within 100 req/day limit
- _build_query(): strips dates, punctuation and stopwords from market question
- _score_headlines(): keyword-based pos/neg vote per article, averaged ∈ [-1, +1]
- Degrades to 0.0 on missing key, 403 quota, or network error
bot/strategy/bayesian.py:
- BayesianStrategy(news=NewsClient) — optional, backwards compatible
- Signal 4: GNews sentiment applied as direct log-odds shift (weight=1.5)
so a ±1.0 sentiment score moves a 50% prior to 82%/18%
- +0.10 confidence boost when news signal is present
- NEWS_LOGODDS_WEIGHT constant documented at module level
bot/main.py:
- Instantiate NewsClient, pass to BayesianStrategy, close in finally block
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Every market now emits an INFO line:
TRADE/SKIP <question> | cat=... | prior=... | est=... | edge=... | conf=... | dir=... | signals=... [| reason=...]
Unsupported-category and no-external-signals early exits also log at INFO
so the full evaluation funnel is visible without changing log level.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- polymarket.py: add keyword lists for politics (election, trump, ukraine…),
tech (AI, OpenAI, Apple, nvidia…), and events (super bowl, oscar, spacex…);
introduce _detect_category() so all four categories flow through a single
code path; filter already-expired markets (end_dt < now) in addition to
the existing future-cutoff filter; log per-category counts at startup
- bayesian.py: extend is_any_supported to include is_politics / is_tech /
is_events; use BTC as a risk-sentiment proxy for non-crypto categories
(halved weight to reflect weaker correlation); cap confidence_cap at 0.65
for macro/politics/tech/events; MIN_EDGE stays at 0.10
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
chemavx
chemavx