2026-07-01 to 2026-07-27
The watchlist-curator agent was called biweekly over the 2026-07-01 to 2026-07-27 window. At each rebalance it read a date-clean news pool for the preceding biweekly window (built upstream from GDELT GKG + Wayback, look-ahead-reduced) and proposed adds and removes against the active watchlist; the optimizer then ran mean-variance on the revised watchlist. Each rebalance is marked by an orange square on the curator curve in chart 2 — hover over one to see that rebalance's adds, removes, and the curator's rationale. The buy-and-hold curve below is the value of the initial portfolio (which never gets rebalanced or optimized) over time. The buy-and-hold portfolio has equal amounts of [AAPL, GOOGL, AMZN] and is held without any rebalancing across the window. Throughout the charts below: general_markets = defensive equity ETFs (broad-market / dividend / utilities / staples); cashlike = bonds + cash-equivalents + precious metals (e.g., AGG, BIL, IAU).
The exact optimizer and backtest knobs behind the charts below, read from investor_profile.md. This is the same config /review-portfolio uses with real money.
| Forward-test window | 2026-07-01 → 2026-07-27 |
| Rebalance cadence | biweekly (4 curator calls) |
| Starter watchlist | AAPL, GOOGL, AMZN |
| Initial capital | $50,000 |
| Risk aversion (λ) | 2 |
| optimizer_lookback_days | 150 |
| Concentration cap (max weight) | 100% |
| Min trade size | 10% of portfolio |
| Max watchlist size | 6 |
| news_lookback_days | 14 |
| Always-include anchors | SPY, AGG, IAU |
| Risk-free rate | 4% |
| Execution lag | 1 trading day(s) after each rebalance signal |
The 2 of 3 biweekly curator calls that made a change (no-change rebalances are hidden). The Rejections column lists each add/remove the validator dropped as invalid, as TICKER (action) — hover for the reason (see REFERENCE.md). Retries = reject-and-retry rounds fired that week (the validator fed a rejection reason back and the curator re-proposed); blank = none.
| Date | Adds | Removes | Rejections | Retries |
|---|---|---|---|---|
| 2026-07-01 | LMT | — | — | — |
| 2026-07-08 | RKLB | — | — | — |
Each add's forward price return (from the add date until the ticker is removed, else window end), bucketed by the news_evidence source (by URL domain) and by the wave keyword that surfaced the cited article. Answers which desks / search terms produced winning picks. n = number of adds.
Each add's forward price return (from the add date until the ticker is removed, else window end), bucketed by the news_evidence source (by URL domain) and by the wave keyword that surfaced the cited article. Answers which desks / search terms produced winning picks. n = number of adds.
How many adds cited each source (by URL domain) as evidence — the raw n behind plots 8 and 9.
Plot 8's forward price return, but bucketed by the article author (byline extracted from each cited evidence URL) instead of the source domain. Which reporters surfaced winning picks. n = number of adds; a co-authored article credits each byline. Adds whose evidence URL has no captured byline are omitted.
How many adds each byline is credited on (co-authors each counted; adds whose evidence URL carries no captured byline bucket into (no author)) — the raw n behind plot 13.
One row per biweekly rebalance. Click to expand the curator's overall rationale, each add/remove with its reason, the cited news_evidence links, and the wave keywords behind that rebalance's pool (GDELT GKG + Wayback, date-clean).
Rationale: Added LMT to the first free managed slot to capture the geopolitical realignment/defense rearmament wave, which is showing concrete acceleration through Lockheed's hypersonic missile unveiling and a structurally higher defense budget pressing states to compete for contracts. Filtered noise: SpaceX article described a pre-IPO private company with no tradeable ticker (discarded); robotics articles (XMAQUINA, Morgan Stanley China humanoid forecast) were industry surveys without a specific US-listed ticker with a near-term catalyst—Physical AI remains early and lacks a pure-play vehicle with demonstrated revenue inflection; AI demand justification article confirmed the current AI wave's durability but NVDA/MSFT/GOOGL already capture this. No remove warranted: all three current holdings remain core to the active AI wave with no thesis breach signaled this week.
Rationale: Filled the single free managed slot with RKLB to establish early exposure to the rockets & spacecraft wave, which the profile identifies as a key 'next wave' but which was absent from the current watchlist. Filtered significant noise: QUBT (quantum computing microcap with acquisition hype but no clear commercial traction), private companies (SpaceX, Anduril, Gritt, AI2 Robotics, Apptronik, UMA), foreign/defense contractors without US listings (PGZ, Rheinmetall), and non-investable entities (governments, agencies). Also filtered robotics names (Boston Dynamics/Hyundai, Tesla Optimus competitors) because the wave is earlier-stage with no clear public pure-play and the profile's 'current wave' is still AI, where we already hold NVDA/MSFT/GOOGL. Defense names (LMT, ATACMS production) are already held via LMT. Nuclear SMR news (Aalo criticality, Centrus contract) is promising but RKLB offers cleaner wave exposure with more immediate catalysts.
Rationale: Weekly cadence demands high conviction for any churn; no clearly superior rising-wave vehicle emerged to displace current holdings. Filtered noise: (1) Private/unlisted names—SpaceX (SPCX is a tracking stock with thin float, not a direct listing; no pure-play), Gritt, Apptronik, AI2 Robotics, UMA, Hadron Energy pre-merger SPAC; (2) Foreign-only or thin names—ispace (Japan), Digantara (India), Hyundai/Boston Dynamics (private), FANCO (private), Steady Energy (private), PGZ (Poland state), Rheinmetall (foreign); (3) Non-wave or excluded sectors—solar/wind excluded per profile, analog chips (general AI supply chain, no wave edge), real estate/REITs (demographics play but no immediate catalyst), Samsung/Meta/Google/WhiteFiber (big tech, no wave-specific exposure); (4) Hype without ticker—quantum computing executive orders, World Cup robot publicity stunts, Starship test flight previews (already priced, RKLB holds the space slot). Valid wave signals existed: nuclear SMR momentum (Centrus $1B+ DOE contract, Aalo criticality), quantum computing policy acceleration, robotics funding surge, defense rearmament (ATACMS co-production, Barracuda missile, counter-drone contracts). However, current watchlist already captures these themes: NVDA/MSFT/GOOGL = AI (current wave), LMT = defense/geopolitical, RKLB = rockets/spacecraft (next wave with near-term catalyst—Victus Haze demo, NASA contracts). No single add candidate showed enough edge over these five to force a swap at weekly frequency. Centrus (LEU) is the strongest nuclear SMR pure-play with concrete catalyst, but would require dropping LMT (defense) or RKLB (space); both have equally valid near-term catalysts (LMT: Germany ATACMS, Poland missile deals; RKLB: Space Force demo, Starship supply chain). Quantum pure-plays (QUBT, IONQ) remain speculative with no revenue inflection yet. Robotics lacks liquid pure-plays—Apptronik and peers are private. Demographics/healthcare REITs (SBRA, WELL) are slow-moving, no immediate wave alignment. Result: maintain current 5-ticker watchlist, let optimizer weight across the five.
Rationale: Weekly window with no clear catalyst-driven swap opportunity. Current watchlist (NVDA/MSFT/GOOGL for AI, RKLB for rockets, LMT for defense/geopolitical) covers the three active wave buckets well. Pool noise filtered: SpaceX (SPCX) is newly public but showing execution risk (aborted launch, below IPO price) rather than buildup momentum—too early and too noisy for a weekly rebalance. SMR/Oklo nuclear names lack fresh contract wins versus prior catalysts. Quantum (QUBT) is speculative with no commercial milestone this week. Robotics saw private funding rounds (Humanoid, Gritt) but no new liquid US-listed exposure with immediate catalyst. Samsung/Hyundai robotics developments are supply-side, not new investable vehicles. Demographics REITs (SBRA/WELL) are steady but not wave-accelerating. No remove triggered either: NVDA/MSFT/GOOGL remain AI infrastructure core; RKLB holds active mission momentum (Victus Haze, recent launch activity); LMT has fresh defense news (cheaper Patriot interceptor, German ATACMS production) keeping geopolitical relevance. Watchlist at 5/5 managed slots with balanced wave exposure—no forced churn warranted.