Altrix research reports
Self-contained interactive reports. Each page settles in the browser; no server, no network.
Portfolio composer
| version | members | basis | built |
|---|
| current | 23 | 69,763 legs / 20,667 events | 2026-09-06 |
| 23m-20260906 | 23 | 69,763 legs / 20,667 events | 2026-09-06 |
Experiment 264 — hold longer split 29/31
Does holding 2–4× longer than the live 180 min pay? Sweep of 617 books over 68 chains,
with the deployed t20/t26 replayed at frozen weights as the control.
| report | what it is | scope | built |
|---|
| 264-sweep-report | horizon sweep + pre-registered gates P1–P6 | 617/617 books, 68 chains | 2026-09-07 |
| 264-audit-report | audit of the “longer timeout would have paid” claim on the live ledger | 13,770 closed legs | 2026-09-07 |
Verdict: REGIME_LOADING — longer horizons book more PnL, but 0 of 268 arms keep
half their edge once the three biggest tokens are dropped. Nothing is promoted to deploy.
Experiment 300g — concentration caps and dumb gates tier-1
Can capping how much the book stacks into one token cut the episode
drawdowns the exit side could not remove? Caps by dollars, by entry count, on a calendar day, on a
rolling window, plus dumb gates on a token’s own unrealised PnL. 53 cells per trader, re-admitted
from the frozen post-val books and asserted against the screen to 7.3e-12 USD.
| report | what it is | scope | built |
|---|
| current | PnL curves, drawdown panels and metric tables for every cap, window and gate cell | 53 cells × 2 traders, 108 curves | 2026-09-09 |
| 20260909 | immutable copy of the same build | 53 cells × 2 traders, 108 curves | 2026-09-09 |
Verdict: TIER1_SCREEN — 2 of 53 cells clear the pre-registered
gates on both traders, and both are dumb uPnL gates; no structural cap and no rolling-window cell passes.
Exit clusters are not re-settled, so every number understates a real cap and nothing here is
executable.
334 Arc stable re-selection · 3 Oct validated cut
27 Sep–3 Oct 10:45 UTC is in sample; 16–26 Sep is the pre-window holdout. Post-window replay is pending validation. Frozen rosters, not live trading changes.
current always serves the latest build. Versioned files are immutable once published.
Earlier builds remain reachable at their own deployment subdomains.
334 Arc stable re-selection · 4 Oct holdout revision
335 RH stable re-selection · 27 Sep–3 Oct
Six frozen selection approaches, with and without additional RH v1 bundle admission. Post-window complete entries cover at most one hour.
330j Arc composer · 4 Oct validated cut
Five frozen research pages through 4 Oct 15:00 UTC, using 334’s FINAL-free tape. P1/P2 and Chrome QA passed. Live eight-wallet reference remains at 3 Oct 10:45. Known exhaustion-twin tie-order instability is disclosed on each page.
330 RH composer · 4 Oct forward pages
Three frozen RH morning pages, from separately full-GET-verified B2 releases.
330k RH composer · 6 Oct 12:00 UTC forward page
Frozen 330k presets extended to the 6 Oct 12:00 UTC validated RH tape, with new entry windows 4 Oct 06→5 Oct 14:05 and 5 Oct 14:05→6 Oct 12 (paper period). Immutable B2 release.
021 Solana organic filter vs 027a detectors · 29 Oct
15 tokens inside Misha's InsightX-decided set (19,376 safe-v2 tokens from Sep 27) that our wallet-based ladder/bot detector flags (from age ≥24h) but his 3h ladder fingerprint and InsightX (clustered >95%) allow. Charts and gmgn links per token.
Ladder detectors on young Solana tokens: ours vs 027a at 1h / 3h · 29 Oct
All 19,376 tokens in Misha's InsightX-decided set judged at +1h and +3h after pool creation: crash rates per detector, who catches what, his 40 control ladders, cost on M0_V2-pen0, suggested approach, and per-token charts with each detector's decision (incl. the 49 3h ours-only tokens).
021 Solana organic filter · 50 good tokens, hour-by-hour verdict · 2 Nov
Hourly causal verdicts Oct 8–24: green band = live verdict GOOD for that hour (decided from trades before it). 1,452 tokens eligible, 308 with ≥1 good hour; 50 shown with charts, gmgn links and snapshot metrics.
Entries by token age: Solana models vs RH/Arc live · 2 Nov
Share of entries, PnL, win rate and size by time since the traded pool was created (<1h … >30d) for SOL 235 baseline100, SOL 148 g8x6u4-h300, SOL 190 M0_V2-pen0, RH 135 control backtest, RH live September and Arc live Sep 16–30.