Blog

What survived the audit

We spent a week deleting our own best numbers. Here is the track record that was left standing — including the tiers that are not working.

Read post →

Long-horizon benchmark by tier: T+60 tested, T+90 pending

A tier-separated long-horizon audit centered on T+60 and T+90. No mature tier beat baseline at T+60, T+90 remained pending, and T+30 is retained only as exploratory noise.

Read post →

We found the May archive. Here is what Quorum actually got right.

A provenance-first audit of Quorum predictions since May: live calls separated from pre-launch snapshots and later hindcasts, with T+30, T+60, T+90, tier results, and the full GOOGL correction.

Read post →

493 right, 278 wrong: Quorum's live direction ledger at T+30

We froze 1,644 live equity signals, kept every miss and abstention, and tested Quorum against the simplest honest benchmark. T+30 showed directional discrimination. T+60 did not beat the bearish majority. T+90 is still running.

Read post →

The first 28 days of receipts: how our council called the market on 60 stocks

Three weeks ago we promised to publish our track record honestly — predicted direction vs realized return, scored against the market. The first version of this post said 75% direction-correct at T+14d for our top tier. The honest number, scored straight from the GMM posterior instead of a stale chip label, is 58% (n=24). We're leaving the original claim AND the correction both visible.

Read post →

Ground-truth, finally: how we're building a long-term track record from day one

Every stock-research tool dodges the question: were you right? Not the headline signal — the post-hoc 'if you had, you'd be up Y%.' We just shipped the storage layer that will answer it for every verdict the council issues from today onward. The first publicly verifiable LLM-stock-research track record in the sector starts populating now.

Read post →

A worked DCF on AAPL — and the three places the Council disagrees with the textbook model

5-stage DCF on Apple: WACC build, terminal-growth sensitivity, and three places the Council's bear seats refuse to accept the textbook intrinsic-value story at face value. The intrinsic-value swing from a 1% shift in g_terminal is ~25% — and that's the entire active investing debate.

Read post →

Why P/E lies on capital-intensive cyclicals — running the EV/EBITDA comp Quorum's Council uses

A debt-heavy industrial vs a debt-light tech comp will always look mis-priced on P/E. Walk the lease-adjusted EV/EBITDA comp the way the Council's MoatStrategy and CapitalAllocation seats read it — with Carrier (CARR) as the worked example.

Read post →

Reading a 10-Q like the Cartographer seat — the DSO/A-R stretch that flipped BYRN before the drawdown

Days-Sales-Outstanding stretching +258% over four quarters is not noise — it's the customer-concentration tell hiding in page 47 of the 10-Q. Walk the working-capital flags the Council's Cartographer seat surfaced on BYRN, and the named falsifier that flipped the Bull seat under rebuttal.

Read post →

Bloomberg $24k. Koyfin $99/mo. Tikr $50. Seeking Alpha $239/yr. Quorum: the honest line-by-line.

What you actually pay for stock research and where Quorum's four research tiers fit: Quorum Lite, Quorum Pro, Quorum Max, and Quorum Ultra-Max.

Read post →

The bell curve you've never seen on a research tool — visualizing a 4-1 dissent on AAPL

Quorum's verdict isn't a chip — it's a shape. Ten seats, ten little hills, one mountain range that tells you not just where the council landed but how much it agrees with itself. Here's what that chart actually looks like, side-by-side, when the council is unanimous vs when one seat is dissenting hard.

Read post →

Four engines, one verdict — pick the tier that fits the question

Quorum has four research engines — Quorum Lite, Quorum Pro, Quorum Max, and Quorum Ultra-Max — with the same canonical verdict and Executive Briefing contract.

Read post →

Why your free seat is actually free forever

Most SaaS free tiers cost you something — your data, your attention, a hostile upgrade nag. Quorum Lite costs us GPU time and nothing else. The mechanism that lets us keep it free is an operator policy we wrote into the codebase.

Read post →

Capping Claude Opus at 10 runs a day — and why that's a feature

Quorum Max runs a 4-prompt frontier-model sweep per ticker. We capped it at 10 runs per day for every account, owners included. The reason isn't budget — it's that 10 deep verdicts you actually read beats 50 you skim.

Read post →

Profile-weighted ENTER/WAIT picks for your watchlist

Same stock, different conviction depending on who's asking. A balanced retiree and an aggressive small-cap chaser shouldn't get the same verdict on PLTR. Quorum's Watchlist Council weighs five different signal channels per profile and downgrades anything that can't clear the profile's margin-of-safety floor.

Read post →

The dissent is the signal — how a bell curve preserves the lone bear

A 5-0 council vote is suspicious; a 4-1 with a well-cited dissenter is useful. Quorum's verdict math agrees. We use a bounded Gaussian mixture so one confident dissenter actually shifts the curve instead of getting clipped to a spike at the boundary.

Read post →

We taught the Council to grade itself

Stage 2.5 — every seat now blind-ranks the others on argument quality, identities stripped. After enough runs, the question "is the Skeptic actually adding value?" stops being a gut call and becomes a number. Plus a general-purpose Ask-the-Council surface and a faster Cartographer.

Read post →

Why your stock analysis should debate itself

Quorum convenes a 10-seat adversarial Council — five bull seats versus five bear seats, plus a Synthesizer Chair — before issuing any non-trivial chip. Quorum Lite, Quorum Pro, Quorum Max, and Quorum Ultra-Max share one canonical verdict and briefing contract.

Read post →