AI Stock Analysis by Investment Committee: Five Personas, One Evidence Ledger, One Bottom Line

AI stock analysis by investment committee means several AI investor personas each build a case for one stock from the same sourced evidence ledger, challenge each other's cases, and a neutral chair writes a plain-English bottom line with the dissents on record. It is the method inside the SentiSense us-stocks-analysis skill, and this post shows what it produces on a real ticker, run on live data on 2026-09-26.

Why a committee instead of one answer?

Ask a model "is META a good long-term holding" and you get one confident paragraph. Nothing in a single pass forces it to source its numbers or argue against itself.

The committee is a six-step pipeline, each step finished before the next: restate the question as one falsifiable thesis; fill an evidence ledger from live API calls (price, reported financials, analyst consensus, insider and institutional filings, options positioning, recent stories), with a source and as-of date on every row and [NOT AVAILABLE] on anything that cannot be sourced; write a neutral base case from ledger rows only; seat the personas; run the debate; hand the chair the verdict.

The five default seats are built from publicly documented investing philosophies and named by what they look for: the Quality Owner (is this a durable business at a sensible price), the Inverter (how could this decision be stupid), the Story Checker (do we know what we own, and is the growth already paid for), the Macro Trader (is the liquidity and the tape with us), and the Short-Seller (if this were a fad or a broken model, where would it hide). Each seat is a worksheet of specific checks, not a personality to imitate, and the Short-Seller seat is mandatory.

Two rules do most of the work. Every seat must file at least one challenge against a different seat, and at least one challenge must be serious enough to change a stance. And no seat may reply without first restating the challenger's strongest version of the point. Conceding on evidence is scored as a win.

What the committee produced on META

The thesis tested: "At today's price, META is a reasonable three-to-five-year holding." The ledger was filled from the SentiSense stock snapshot, analyst consensus, annual and quarterly financials, news stories, options, insider and 13F tools, the screener and the earnings calendar. The ten-year Treasury yield row was not available on this host, and the output says so. Two figures below are derived from ledger rows rather than reported: the sales and earnings multiples, computed from the screener's $1.915 trillion market value and the four most recent reported quarters. What follows is abbreviated; the full run also prints the evidence table.

Bottom line. The committee is split on whether META at today's price is a comfortable three-to-five-year holding. The advertising business earns far more cash than it reports as profit and grew revenue about 22% in fiscal 2025, but the company is now spending most of that cash on infrastructure: investing outflows more than doubled to $102.0B in fiscal 2025, long-term debt went from $28.8B to $83.7B in three quarters, and free cash flow in the June 2026 quarter came in at $1.7B. The question the table could not settle is whether that spending earns its return before the market stops paying roughly 28 times trailing earnings for it.

Confidence: low, because the four voting seats split 2-2 on the one question that decides the thesis, and the rate environment could not be read on this host.

How they voted: For 0 / Leaning for 2 / Leaning against 1 / Against 1 / Pass 1.

Opening case, the Quality Owner (is this a durable business at a sensible price): Leaning for.

1. Operating cash flow of $115.8B in fiscal 2025 was nearly twice the $60.5B of reported net income, so the profits are real cash. Gross margin has held above 80% for three fiscal years.
2. The share count is falling, not rising: basic weighted-average shares went from 2.69B in fiscal 2022 to 2.54B in the June 2026 quarter, about 5% fewer shares over the period, a derived figure.
3. Long-term debt of $83.7B against $261.2B of equity, with a current ratio of 2.23, is a balance sheet that survives a recession. At roughly 28 times trailing earnings and 8.4 times trailing sales (both derived), the price is full but not detached from a business growing revenue at 22%. What would change my mind: "Free cash flow below $5B for two more quarters while the investing outflow keeps rising."

Opening case, the Inverter (how could this decision be stupid): Leaning against.

1. Assume the thesis fails in three years. The most plausible cause is that the infrastructure spend never earns its cost of capital, and it would show up first in free cash flow, which went from $14.8B in the December 2025 quarter to $13.2B in March to $1.7B in June.
2. Fiscal 2025 net income of $60.5B came in below fiscal 2024's $62.4B even though operating income rose from $69.4B to $83.3B. The cause of that gap is not in the ledger, and a committee that cannot explain it should say so.
3. The bull case's exposed bias is narrative: the Muse AI agent launch at Connect 2026 on September 23 drew coverage from 26 sources, and none of it is a revenue line the ledger can check yet. What would change my mind: "A reported segment or product line that shows the new spend producing revenue, at any scale."

Opening case, the Short-Seller (hunts for what could go wrong): Against.

1. Every one of the twenty insider rows in the last 90 days is a sale by one of two officers, with zero open-market purchases; the three largest are $14.0M, $13.5M and $12.1M.
2. Long-term debt rose by about $24.9B in the June 2026 quarter alone (derived from the two reported balances) while free cash flow for the same quarter was $1.7B on $31.9B of operating cash.
3. Fifty-five buys, seven holds and zero sells across 62 analysts, with an average target of $790.27, is perfection priced in. Eight of the twenty largest 13F holders trimmed in the June quarter. What would change my mind: "One open-market insider buy, or free cash flow back above $10B a quarter with the debt balance flat."

Opening case, the Story Checker (do we know what we own): Leaning for.

1. Classified as a stalwart with fast-grower revenue: 22% growth in fiscal 2025 on a $201.0B base, a two-minute story anyone follows (ads on apps billions of people use daily).
2. The growth roughly covers the multiple, at 28 times trailing earnings (derived) for a business that grew earnings per share from $8.59 diluted in fiscal 2022 to $23.49 in fiscal 2025.
3. Insiders are not buying with their own money, and the crowd has clearly found this name.

The Macro Trader (is the tape with us) passed: without the Treasury row it cannot run half its worksheet, and it would not vote on the half it could. It recorded what it saw: eleven of the twenty largest 13F holders added in the June quarter, eight trimmed, one entered new; in the options chain, put/call volume sits at the 73rd percentile of the stock's own history while 25-delta skew sits at the 21st, so more puts are trading but nobody is paying up for them; the biggest open-interest walls are calls at $900 and $825.

Challenge, the Short-Seller challenges the Quality Owner: the facts say otherwise, serious. "Your cash-machine case is fiscal 2025's. The June 2026 quarter converted $31.9B of operating cash into $1.7B of free cash, the company borrowed about $25B in that one quarter, and the insiders in the file sold on all twenty rows and bought on none."

Reply, the Quality Owner answers the Short-Seller: partly concedes. Their best point: "Free cash flow of $1.7B on $31.9B of operating cash means the spending, not the business, sets the return for the next several years." Answer: "Granted that the cash is being spent rather than kept, and that a $25B borrowing quarter is a choice I did not weigh enough. But it is a choice made from $261B of equity and a shrinking share count, not a business that needs the money to survive. The test is whether the September and December quarters rebuild free cash flow." Stance now: Leaning for, unchanged.

Challenge, the Inverter challenges the Story Checker: blind spot, serious. "Fifty-five buys and zero sells across 62 analysts is not evidence for the thesis; it is the crowd you are supposed to be checking for." The Story Checker conceded the point in one line and kept its stance, on the grounds that the growth, not the crowd, carries its case.

Where they split: whether the infrastructure spend earns its return. Both sides cite the same rows; they disagree on how to read a $102.0B investing year from a company that also produced $115.8B of operating cash.

Who disagrees and why: the Inverter, because the failure mode is visible in the free cash flow trend and the fiscal 2025 profit gap has no explanation in the ledger. The Short-Seller, because every insider row is a sale and the debt balance nearly tripled in three quarters. The Macro Trader declined to vote without a rate read.

What to watch: the next quarterly report, estimated for October 28, 2026, for free cash flow and the investing line; the insider file for any open-market purchase; and what penalty follows the New Mexico jury's September 25 finding of about 43.9 million violations of the state's Unfair Practices Act, which is not in the ledger.

What we don't know: the rate environment; why fiscal 2025 net income was below fiscal 2024's despite higher operating income; the dollar consequence of the New Mexico verdict.

How to run it

Install the skill into the agent you already use, set a free API key, and ask a judgment question. Factual questions ("what is META's price target") route to a five-line quick read; judgment questions ("should I hold NVDA for five years", "red-team my COIN thesis") convene the committee.

# Cursor, Codex, Claude Code, Gemini and other coding agents
npx skills add SentiSenseApp/skills
export SENTISENSE_API_KEY=...

The skill is also listed on ClawHub for OpenClaw. It runs as parallel sub-agents where the host supports them and as a sequential pass where it does not, with a three-seat lite path for small local models. Only the seat count degrades; the rounds, the mandatory challenge and the dissent section never do.

Ready to convene one? Get a free API key → and paste it into the export line above.

---

SentiSense provides market information and analysis for informational and educational purposes only. The committee output above is a research summary of a thesis derived from AI-analyzed data and should not be relied upon for investment decisions. It is not investment advice.