Transparency

How Alfred stays honest

Our promise isn't returns — it's that every manager's rules and maths are published, and that the software provably does what the page says. Here is exactly how we hold ourselves to that, in terms you can check.

Latest code‑vs‑maths audit · 6 July 2026

Every one of our 22 managers was reviewed line‑by‑line against its published formulas. 13 matched with no deviation of any kind; 2 more differed only cosmetically (a timing convention, a log‑wording nuance — no trade or NAV impact). The remaining 7, plus one platform routine, carried 8 material divergences — and every one was fixed before publication.

We publish the result, including what we got wrong, rather than only what we got right. The full findings and every formula are in the manager mathematics document.

Since that review, further managers have been built and run backstage — every manager trades live for a period before it is released to customers, building the record that justifies releasing it. Live‑only managers carry no backtest to audit against (real‑time signal can't be replayed) and are judged on that record alone; rules‑based ones have their backtests replay‑verified before release. All will be covered by the next roster‑wide audit.

What we found & fixed

The audit is a running ledger. When a review — ours or an automated check — turns something up, we fix it and record it here, including whether any published number moved.

22 Sep 2026Corrected the benchmark series every manager is measured against, after a flaw in the daily update let some of them drift away from the real index.
Published comparisons moved, and one of them had been flattering a manager. Each day’s benchmark value was chained onto the day before; when the index had no price on the day the chain reached for — a thinner fund, an Indian market holiday, or a weekend in a series that holds Bitcoin — it anchored to the wrong day, and every later value inherited the error. Six of the nine series were affected, the worst by 11.8%. Among the managers you can hire, only Margaux Laurent’s comparison changed materially: her benchmark had been understated, flattering her by roughly 0.9 points, and her profile now shows the true, wider gap. Yusuf Mansour’s moved by less than a tenth of a point. The eight measured against the S&P 500, the Nasdaq 100 or Bitcoin were never affected, because those three series were always correct. All six have been rebuilt from source prices and now match an independent recomputation exactly. A manager’s own record is produced by entirely different code and was not affected; to confirm that, we rebuilt one manager’s record from source prices and it matched what had been published.
6 Jul 2026Removed a filing-lag look-ahead from the Forensic Accountant’s accounting-strength screen.
Live-only manager — no published backtest was affected and its fingerprint is unchanged. Future quarterly screens now apply the same 90-day point-in-time filing lag as every other fundamentals manager.
View this manager →
6 Jul 2026Hardened the NAV calculation against a holding that outlives its index (a delisting or index review).
Platform-wide safety fix. Such a case now pauses the affected manager with a clear, actionable reason instead of failing opaquely — no wrong NAV was ever published.
6 Jul 2026Locked the core safety rules — the trade guardrails, the backtest-equals-live settlement, and the metric maths — behind an automated test suite that runs on every change.
No number changed; this makes a whole class of future regressions impossible to ship unnoticed.
4 Jul 2026Fixed 8 material code-vs-spec divergences found in the first roster-wide audit — across 7 managers and one platform routine.
The affected backtests (Tunde Adeyemi, Walter Hayes, Margaux Laurent, Jared Fletcher, Priya Nair) were re-computed and replay-verified before publication; the corrected figures are the ones shown on those profiles today.

One code path

Backtest and live trading call the exact same strategy functions. There is no separate "marketing" backtest — the equity curve on every profile is produced by the same code that makes the live decisions. If a published curve can't be reproduced from the maths, the page is wrong, not the maths.

Deterministic & reproducible

Every backtest is stamped with a cryptographic hash of the exact code and constants that produced it (shown per manager below). Re‑running the engine yields identical decisions, bit for bit — we verify this on every rebuild, so a number can never quietly drift from the rules that made it.

Audited against the maths

On 6 July 2026, independent reviewers checked every manager's engine module against its specification — every parameter, threshold, window, tie‑break and refuse‑to‑trade guard — and re‑derived each formula against its cited source (Faber, Antonacci, Greenblatt, Piotroski, Connors, Frazzini–Pedersen, Ehlers, Moreira–Muir, and the NSE index methodologies). What diverged was fixed, not filed away.

Limitations disclosed, not hidden

Every profile names its own honest caveats — survivorship bias where a current‑constituents list is used, live‑only managers that can't be backtested honestly, model shorts, rupee‑denominatedreturns, point‑in‑time filing lags — and, above all, no promised returns. Losing managers are shown exactly as they are. When a manager's live drawdown overtakes the worst its backtest advertised, a nightly check publishes that on the profile — above the hire button, not beneath it.

Per-manager verification

The code fingerprint behind each manager's published backtest. Live‑only managers have no backtest by design and instead build a public record from day one.

Manager	Rules	Backtest fingerprint	Last computed
Tunde Adeyemi	Published in full	3c47119a7d	2026-10-20
Stefan Brandt	Published in full	da0346a0c9	2026-10-16
Walter Hayes	Published in full	8f32317653	2026-10-21
Omar Haddad	Published in full	a40f95a87a	2026-10-04
Isabela Moreira	Published in full	739b8ecb9e

Every decision is on the record

Beyond the maths, each manager writes a plain‑language reason for every action — and every deliberate non‑action — to a public log. You can read exactly why a trade was or wasn't made, and at what price, on any profile. Which holding a decision names is shown to signed‑in customers thirty days on, and as it happens to anyone who has hired that manager — the record is public; the recipe is not.

Browse the managers →See the live decision log →

Your broker key — common questions

Can I connect the same eToro key to two Alfred accounts?

No — one key belongs to one Alfred account, and a key already in use elsewhere is refused at connection. Two accounts steering the same brokerage book would each keep their own ledger against the same money, and neither would be right. Create a separate key (or eToro account) for each Alfred account. We store a one-way fingerprint of the key to enforce this — never the key itself in readable form.

If I disconnect a broker account and reconnect its key from a different Alfred account, what carries over?

The money carries over; the meaning of it doesn't. Your positions and cash live at the broker, so the new Alfred account will see them — but managers, delegations, approvals and value history belong to the old account and stay archived there. The holdings arrive unattributed: no manager claims them, and a fresh delegation trades on top of (or, on a virtual account, replaces) the old book. Cleanest is to sell down before moving.

Does disconnecting sell anything?

Never. Disconnecting only pauses your delegations and stops Alfred reading the account. Nothing is bought or sold without your approval, connected or not.

Backtested figures are hypothetical; past performance does not predict future results. Virtual accounts use practice money; real accounts trade real money, and you authorise every trade. Nothing here is investment advice.