Basic Investor

Basic Investor

_{<a href="README.ko.md">한국어</a>}

Reads what happened, works out what it means, writes down a claim that could be proved wrong, and only then asks whether your portfolio should change.

In one paragraph

Basic Investor is the plainest manager in the catalogue: one analyst, reading one asset, in the order a careful analyst actually works. It gathers the facts first and refuses to have an opinion while it is doing so; then it interprets them; then it writes a thesis — a claim about the future, with the condition that would prove it wrong; and only at the end does it ask what any of that means for the portfolio you actually hold. Its usual answer is either WAIT or a BUY that comes with that written thesis attached.

It is also the reference package — the one the Aumos test suite runs on every commit — so it doubles as the worked example for anyone writing their own. That side of it lives in WALKTHROUGH.md (WALKTHROUGH.md), so this page can be what every other manager's page is: the page a person reads before installing something that will judge their money.

The methodology

FinRobot's three-stage chain of thought, extended with two Aumos stages.

Words this page uses, in plain terms:

Thesis	a written claim about the future, including the condition that would prove it wrong. Not a rating
Chain of thought	forcing the reasoning into fixed stages, so a conclusion cannot be reached before the facts are in
**asOf**	the instant the judgement is pinned to. The manager can only see what had been published by then
WAIT	"nothing here changes what the portfolio should hold" — a first-class answer, not a failure to decide
Evidence	the record of what the manager actually fetched, kept by Aumos rather than claimed by the manager

stage	what it establishes	where it comes from
1 — Data	the facts, and nothing else. Prices, filings, headlines, gathered before any interpretation is allowed	FinRobot
2 — Concept	what those facts mean — the mechanism, not the mood	FinRobot
3 — Thesis	a claim about the future that could be proved wrong, and the condition that would do it	FinRobot
4 — Portfolio context	an asset view becomes a portfolio view: what does the book already hold, and does this change it?	Aumos
5 — Decision	one action, one target, and the reasons	Aumos

Three things in that are decisions rather than structure.

WAIT is stated as a peer, repeatedly. Left alone, an analyst-shaped prompt produces analyst-shaped output: something to do. So the prompt says out loud that an unjustified BUY and a well-reasoned WAIT are not close to equally good, and stage 4 closes by asking "does this change what the portfolio should hold?" rather than "is this a good company?" — because those questions have different answers, and only the first one is the product.

The thesis is not the manager's. It arrives as read-only context and goes back as a proposal. Long-term memory belongs to Aumos rather than to the manager: a manager that kept its own would take the reason a position exists with it when it was swapped out. The thesis stage proposes the next revision of an existing thesis rather than restating it, because a revision chain full of no-op restatements loses the property that makes it worth keeping — that every version transition has a reason and a decision behind it.

It says the case against itself. Stated uncertainty is a required part of the answer, not a courtesy.

How a run works

One run, one proposal. Nothing below places an order.

flowchart TB
    classDef reads fill:#1e2a44,stroke:#6f9bf0,color:#cfe0ff
    classDef judges fill:#2f2f38,stroke:#9aa0b4,color:#e8eaf2
    classDef proposes fill:#1b4332,stroke:#40916c,color:#d8f3dc
    classDef person fill:#5c4813,stroke:#f6a609,color:#ffe8b0

    WAKE["Aumos wakes it on an asset,<br/>event or thesis review"]:::reads

    subgraph IN["1 · Data — facts before opinion"]
        direction TB
        PX["Price history"]:::reads
        FIL["Filings"]:::reads
        NEWS["Headlines"]:::reads
        BOOK["Your portfolio as it stands"]:::reads
        TH["The thesis already written,<br/>as read-only context"]:::reads
    end

    S2["2 · Concept<br/>what those facts mean —<br/>the mechanism, not the mood"]:::judges
    S3["3 · Thesis<br/>a claim that could be proved wrong,<br/>and the condition that would do it"]:::judges
    S4["4 · Portfolio context<br/>an asset view becomes a book view"]:::judges

    Q{"5 · Does this change what<br/>the portfolio should hold?"}:::judges

    WAIT["WAIT<br/>a first-class answer,<br/>with the reasons"]:::proposes
    ACT["BUY / SELL / RESIZE / WATCH<br/>with the thesis revision attached"]:::proposes
    UNC["Stated uncertainty — required,<br/>on every answer"]:::proposes

    subgraph HUMAN["Where a person decides"]
        direction TB
        MAND["Aumos judges it against your Mandate"]:::person
        YOU["You approve, or you do not"]:::person
        ORD["Only then does an order exist"]:::person
    end

    WAKE --> IN --> S2 --> S3 --> S4 --> Q
    Q -- no --> WAIT
    Q -- yes --> ACT
    WAIT --> UNC
    ACT --> UNC
    UNC --> MAND --> YOU --> ORD

Legend — 🟦 what it reads · ⬜ what it works out on its own · 🟩 what it hands back · 🟧 where a person decides.

⚠️ The five stages are prompt structure, not orchestration. There is one manager call per run; the stages are sections of one prompt sharing one context window, which is the entire reason the decomposition works. Aumos does not run a process per stage, and does not claim to know a manager's internals — what it knows about the inside of a run is what it observed: the tool calls, their timings, their evidence.

Cadence. It is woken by an event or a review rather than by a calendar, and the condition it arms on its answer is what brings it back.

What it needs

Markets	US-listed equities and ETFs (XNAS, XNYS)
Data	source:passthrough — prices, filings and headlines, asked of the vendors this machine holds credentials for. The response is theirs, unread by Aumos; this package does the reading
The book	portfolio:read — so a judgement is about this portfolio rather than about the asset in the abstract
Settings	how many days of price history to request, and how many articles at most. Neither can loosen the time gate, which is the gateway's and is not configurable from a package
Language	it answers in the language the invocation asks for. Field names and enum values stay English — those are the wire format — and quoted sources stay in the language they were published in
Your approval	it proposes and never trades. Every judgement enters your approval queue and a person approves it

What it is bad at

• Anything that needs more than one asset in view at once. It reasons about one subject per run. Comparing two candidates against each other is not something it does.
• Risk budgeting. It reads the book to place its judgement in context, not to manage the book's total exposure. prudent-allocator is the package for that question.
• Speed. Five stages of reasoning and a human approval gate are not a short path, and anything that only works if executed within hours will come back as a WATCH.
• Knowing whether it is any good. One run against fixture data says nothing about whether this manager invests well, and no number of such runs would.

Do not install this if you want a specialist. Its virtue is that it is the ordinary case done carefully, which also means it is nobody's edge.

Notes

Evidence, and what "it saw" means. A Decision's evidence is what Aumos observed the manager fetch, not what it claimed to have read. A citation is self-report; the observation is not. So a Decision leads to its evidence and its evidence to its provenance, answering "what did this judgement look at, where did it come from, and when was it published" whether or not the manager was honest — which is what makes it an audit trail rather than a bibliography.

How it is exercised. Every run in the test suite goes against a fixture world: earnings released at 2026-05-27T20:05:00Z, every run pinned to five minutes before. Post-event values sit nowhere near pre-event ones — closes of 412 against 100, revenue 999,999 against 100,000 — so a rationale mentioning a 412 close is visibly reading the future. The automated half is not a mock: a real coding CLI is spawned, it spawns the real gateway from the real configuration, and every tool call goes over the real protocol through the real time gate. The only faked thing is the judgement itself. The other half is the same bundle in front of a real model on a real subscription.

What that does not establish is that the judgement is any good. That is what a forward track record is for, and it takes calendar time rather than compute. What is established is narrower and worth stating exactly: that a package shaped like this loads, runs inside its permissions, cannot see past asOf, and produces something the Kernel will seal.

Writing your own. WALKTHROUGH.md (WALKTHROUGH.md) is this package read as a worked example — what asOf does to a prompt, why WAIT has to be repeated, who owns the thesis, how the prose is translated and the format never is, and what happens to an answer that fails validation. CONTRIBUTING.md (../../CONTRIBUTING.md) is where the rules are.

Aumos shows no returns for this package until it has earned some. A forward track record takes calendar time rather than compute, and nothing on this page is one.