Research · How the desk reasons
The reasoning is the product
The desk never hands you a number without the working behind it. This is how a call is built — the pipeline every ticker runs through, the philosophy behind each valuation, and the rules the desk holds itself to.
Research
Filings, broker notes and the news log are read and summarised per name — sources, never memory.
Value
Four independent models run: peer P/E, earnings power, Graham and DDM, each with full working.
Backtest
Any rule or signal is tested across ~19 years of PSX history before it earns a place.
Debate
The Desk Room argues both sides; a Chair issues a house view with an explicit dissent.
Score
The call is dated with a falsifiable expectation, then graded against the outcome.
How the desk values a stock
Four lenses, and the honesty to disagree
Multiple lenses, no black box
No single model decides. Four methods triangulate a fair-value band; where they disagree, the disagreement itself is the signal.
Anchored to the risk-free rate
Earnings power is judged against the prevailing bond yield, so a stock is never called cheap in a vacuum.
Evidence over narrative
A compelling story is not a reason. Every claim ties back to a filing, a price series, or a tested rule.
Unknown stays unknown
If a number isn’t in the data layer, the desk renders it as unknown rather than guessing a plausible figure.
No lookahead, ever
A signal at a given bar may use only data available at that bar’s close. Fills come from the next open — never a price the rule could not have seen.
Costs assumed per stock
Friction is estimated name by name from the spread, because a flat cost assumption flatters exactly the thin stocks that momentum rules pick.
Frozen once live
A strategy that has published a backtest cannot be edited. Changing a rule means a new version with a fresh record — so old results can never be quietly invalidated.
Even the sky gets tested
The astrology lens is held to the same bar as every rule here: written down first, permutation-tested, multiplicity-corrected, and published whatever it shows.
Sample research note
A call you can check, line by line
Every note states its expectation and a date, cites the evidence, and lists what would prove it wrong. When the window closes, it is graded — automatically, in public.
- A dated expectation, stated up front
- The evidence the call rests on
- The condition that would prove it wrong
The desk expects <name> to <stated move> within <window>, if <condition> confirms.
Price closes below <level> for <n> sessions, or the quarter misses on margins.
Structure of a real note. Live notes carry the actual name, figures and dates — computed from the data layer, never written by hand.
The method, applied to its hardest case
We tested astrology. It failed. Here is the result anyway.
The astrology lens is the easiest thing on this desk to sell and the easiest to fake, which makes it the honest test of whether any of the rules above are real. So it ran the full gauntlet: traditional claims recorded first, then a circular-shift permutation test on the condition mask — preserving both the autocorrelation of returns and the block shape of the astro windows — stock by stock, never averaged into sectors where a single name’s effect could hide.
143 raw hits against 129.5 expected by chance is not a signal — it is what noise looks like. Nothing survived Bonferroni, nothing survived FDR, and nothing replicated. The desk publishes this rather than hiding it, and the lens is barred from generating or gating a single trade setup. If a condition ever does survive, the evidence will be published on exactly the same terms.
The rules
The rules the desk holds itself to
Research, not advice
The desk never says buy or sell, never promises a return. It briefs; you decide.
Numbers only from the data layer
No price, date or dividend is quoted from memory. Sources back every figure.
Every call is dated and falsifiable
A view without a date and a disproof condition is an opinion, not research.
The desk grades itself in public
Its own calls sit on the same leaderboard as brokers’ — hits and misses both.
It never places orders
Execution is manual and yours. The desk has no path to your brokerage.
See the reasoning on any PSX name
Start free and read a full research note, working and all.