How the track record is counted.

Repeat alerts on the same name within the same day are not counted twice. The first signal in a group is the measured entry; later matching signals become supporting context for the same recommendation.

This rule keeps a noisy cluster of related events from inflating the public count, and it prevents a single fresh insider filing from changing the entry price after the recommendation has already been timestamped.

key = (ticker, direction, DATE(issued_at AT TIME ZONE ‘UTC’))

Each walk-forward checkpoint records the recommendation's return alongside three benchmark returns over the same window, so the same data answers three different questions:

• SPY— the broad S&P 500 baseline. This is the primary published benchmark.
• QQQ— the Nasdaq-100 view. Surfaced where the checkpoint has it (newer calls always do; older ones backfilled as data permits).
• Sector ETF— an SPDR sector ETF mapped from the ticker's sector at issuance (XLK, XLF, XLE, XLV, XLI, XLP, XLY, XLB, XLU, XLRE, XLC). Strips broad sector beta out of the “alpha” number; a tech name beating SPY only because tech rallied looks honest under this comparison.

The headline /track-record page shows all three side by side so a visitor can pick the comparison that matters to them.

One large outlier — a single recommendation that ran +30% in a week — can lift the mean of the whole record without saying anything about whether the rest of the calls worked. The median is the value that survives a small outlier set: half the recommendations did better, half did worse.

Median is the primary headline number on each performance card. Mean, hit rate, and benchmark return are shown as secondary stats next to it. The mean is published for completeness, not as the marketing claim.

Each signal's canonical payload (ticker, direction, signal type, breakdown, why-now, target range, horizon, confidence, issued price, issued-at, benchmark) is hashed using a deterministic SHA-256 scheme at publish time. The hash is unique to that exact published payload.

The hash is submitted to OpenTimestamps. When the calendar batches into a Bitcoin block, the proof attests that the exact payload existed before that block, and therefore before its results. The block height becomes part of the receipt; anyone can re-verify it offline against the Bitcoin chain.

Direct /signal/:id proof pages expose the canonical hash, the issued price, the OpenTimestamps URL, the Bitcoin block height (once confirmed), the walk-forward checkpoint schedule, and the gate status. The page resolves whether the signal is PUBLIC, WATCH, or INTERNAL.

A signal becomes a public recommendation only when every criterion below is satisfied at publish time. The gate is re-evaluated at backfill, so historical signals that no longer satisfy a criterion get demoted automatically.

Direction

LONG only. SHORT public recommendations are disabled until walk-forward evidence at D15 and D30 supports promoting them.

Conviction score

Score ≥ 14 (upper-WEAK band and above). Calibrated 2026-05-28 from a 90-day sweep against the gated cohort.

Confidence band

Never LOW. A LOW classification is always demoted regardless of score.

Issued price

≥ $5 at issue. Sub-$5 names dominate the variance tail in both directions and are kept internal.

Ticker type

No warrants, units, or rights. Detected via ticker-suffix patterns (standalone W / WS / RT, dot-separated .U / .R / -W, etc.) and AlphaVantage AssetType when available.

Coverage

For LONG, AlphaVantage OVERVIEW must exist for the ticker. Tickers we cannot fundamentally cover are kept internal.

Suppress cohorts

Source-combination + sector cohorts with confirmed D7+D15 underperformance against SPY, QQQ, and sector ETF are demoted. Active rules are listed in the gate service and revisited on cadence.

When any criterion fails, the signal is still written and still has its proof receipt. It just does not appear in public-headline counts or in /signals. WATCH-tier signals appear on per-ticker pages as context.