Results & the public registry
This page is the technical record: live numbers, the registered success criteria, and the tamper-evident registry of everything we have ever tried — including what failed.
The pre-registered forward record went live on 20 July 2026: from that date, timestamped stock-vs-benchmark predictions are graded forward on market prices. No capital is deployed and no orders are placed — this is a forward-graded prediction record, not an executed portfolio. The first scored outcomes mature about one month after inception.
Live: the benchmark to beat
A sleeve-matched passive blend, frozen at registration on July 18, 2026: 22.5% QQQ, 27.5% QUAL, 17.5% AVUV, 12.5% XBI, 10% VOO, 10% BIL. Beating an easier benchmark is a parlor trick; this one matches our own allocation, so any edge must come from selection, not risk-taking. Secondary criterion: beat VOO outright. Both required.
| Series | 1-day | Since Jul 20, 2026 |
|---|---|---|
| Autopilot blend | -0.60% | +2.67% |
| VOO | -0.52% | +3.26% |
| SPY | -0.53% | +3.22% |
| QQQ | -0.09% | +3.23% |
| IWM | -0.42% | +0.81% |
| XBI | -1.15% | +7.33% |
Total-return basis (dividend-adjusted, Alpaca IEX feed; closes can differ slightly from official prints). Official measurement is computed in the lab's repository, not on this page. Daily numbers are descriptive — the registered criterion is evaluated once, at window close. No interim verdicts.
AI pick index vs benchmark
Index structure. Three views, one dedupe rule. The Viveka Unique Program Index holds 87 distinct names, equal-weight, each counted once. It splits into two independently-graded sub-indices: Sub-index 1 (S&P 500, 61 names) and Sub-index 2 (VTI total-market, 51 names), each scored against its own sleeve benchmarks. 25 names (29%) appear in both sub-indices but are counted once in the Unique Index — never double-weighted.
| View | 1-day | Since inception | Names |
|---|---|---|---|
| Viveka Unique AI Index | -1.76% | +3.74% | 59 |
| Viveka Unique Rules Index | -2.15% | +3.88% | 53 |
| Sub-index 1 · S&P 500 — AI | -2.33% | +2.47% | 28 |
| Sub-index 1 · S&P 500 — Rules | -2.18% | +6.54% | 28 |
| Sub-index 2 · VTI — AI | -1.36% | +4.76% | 42 |
| Sub-index 2 · VTI — Rules | -2.06% | +4.30% | 42 |
| Benchmark: Autopilot blend | +2.67% |
Marginal contribution (the tested edge):
- Program (Unique AI - Unique Rules): -0.15%
- S&P sub-index (AI - Rules): -4.07%
- VTI sub-index (AI - Rules): +0.46%
Numbers only - the win/loss verdict is judged ONCE at the 12-month window close (registered criterion). No interim verdicts.
The public registry
Finance has no equivalent of a clinical-trials registry — strategies get announced only after they work. Ours is public: every strategy and success criterion is SHA-256-hashed and timestamped before it runs, every variant ever tried is counted, and reported statistics are deflated for the number of attempts. Failures stay on the books permanently.
| Registered object | Track | UTC timestamp | SHA-256 | Status |
|---|---|---|---|---|
| Trimmed Index V1 | Track A | 2026-07-18T20:51:01Z | 5be0a84e3b289ac3… | unrunnable |
| Trimmed Index V2 | Track A | 2026-07-18T20:54:32Z | ee1415e98c65328f… | KILLED by pre-registered criteria |
| Registered object | — | 2026-07-21T16:11:03Z | 3fc6c3203127bc0b… | registered |
| Registered object | — | 2026-07-21T17:05:26Z | cc0c08d71d930796… | registered |
| Registered object | — | 2026-07-21T18:06:20Z | 6a1dee2af0fef5c1… | registered |
| Registered object | — | 2026-07-21T18:06:20Z | 06abf6e64c2a5e9f… | registered |
| Track B success criteria v1 | Track B | 2026-07-18T22:19:19Z | 8d9a7cc25a30c01e… | ACTIVE |
| Registered object | — | 2026-07-19T06:18:40Z | fc9243bc1424e269… | registered |
| Registered object | — | 2026-07-19T13:24:27Z | 22bbeca09475a132… | registered |
| Registered object | — | 2026-07-19T13:28:58Z | 5c19da3590fc8b32… | registered |
| Registered object | — | 2026-07-19T13:42:06Z | 68e3ac385b099936… | registered |
| Registered object | — | 2026-07-19T15:28:31Z | 0c0926f635ae3bb8… | registered |
| Registered object | — | 2026-07-19T17:16:23Z | 5989f441f9e3dd5c… | registered |
| Registered object | — | 2026-07-19T20:33:56Z | caf074328c9aee2d… | registered |
| Registered object | — | 2026-07-19T20:33:56Z | 73af06953138ff22… | registered |
| Registered object | — | 2026-07-19T20:56:13Z | 1f1567d11b3ef208… | registered |
| Registered object | — | 2026-07-19T22:24:58Z | bc27511e264abf5d… | registered |
| Registered object | — | 2026-07-19T22:24:58Z | 590646a96604c42e… | registered |
| Registered object | — | 2026-07-20T00:51:34Z | 0746f03803a0e2e2… | registered |
| Registered object | — | 2026-07-20T02:08:05Z | 4933178ea3cd53e6… | registered |
| Registered object | — | 2026-07-20T02:08:05Z | 08981dd59446b021… | registered |
| Registered object | — | 2026-07-20T10:52:06Z | c686858ce357dff4… | registered |
| Registered object | — | 2026-07-20T10:52:06Z | 089d144c45054fcb… | registered |
| Registered object | — | 2026-07-21T01:50:12Z | a0140616dca59982… | registered |
| Registered object | — | 2026-07-21T12:15:10Z | 20b0d02e0dcafdaf… | registered |
| Registered object | — | 2026-07-21T21:12:21Z | 82bab8e9588447ac… | registered |
| Registered object | — | 2026-07-21T21:25:52Z | fcfa357617932daf… | registered |
Each hash is reproducible from the frozen document text. The cumulative trial count feeds the deflated-Sharpe penalty on every statistic we report.
Exhibit A — our first kill. "Trimmed Index V2" (Track A) was evaluated against its own pre-registered criteria and killed: post-tax 6.46%/yr vs 8.64% for the S&P 500 (top federal bracket + MA state tax). The underlying quality signal was real; after-tax implementation destroyed it. Taxes were ~142× larger than trading costs. That finding is worth more than a lucky backtest.
Signal evidence base
Signal families are chosen from the strongest replicated academic evidence (profitability, conservative investment, momentum — in investable, non-microcap form), with published premia haircut ~50% for post-publication decay and short-horizon signals rejected as cost-killed by default. Where the literature has no trustworthy answer — AI-based analysis, industry-specific signals, adaptive learning loops — we treat our own forward results as the experiment that fills the gap.
How we got here, and the rules we can't break: About the lab →