01
Input ticker list
Tickers, CSV, Excel, or an existing standardized financial metrics table.
Buffett's four filters, run in parallel
Winners, picked by four filters
Circle of competence. Moat. Management. Margin of safety.
A research-aid Skill that turns a long US-equity list into a defensible research queue: LoomLoom handles parallel review, while the local Agent gates evidence, provenance, and re-run conditions.
Research aid only · no target prices, position sizes, or buy/sell instructions · every row keeps a source URL, price timestamp, and falsification anchor.
RESEARCH PRIORITYNot a buy signal
01 / 4 FILTERSBuffett's four filters, in parallel
Circle of competence, moat, management, and margin of safety — each on its own evidence audit, run side-by-side per row.
02 / CLOUDAuditable cloud review
Four parallel research dimensions per ticker; quote first, then explicit user confirmation before any paid execution.
03 / LOCALProvenance-tiered evidence
Every fact carries a source URL, an as-of date, and an S/A/B/C tier. C-tier sources are not evidence; tier-S is the only audit-passing bar.
04 / OUTPUTFalsifiable, never certain
Every row keeps the price timestamp, fiscal period, risks, unknowns, and re-run conditions. Three-tier valuation calibration: deep research, watchlist (evidence thin), or vetoed.
One chain from data
to evidence
Models do the parallel reading; the local program refuses to let unverified content into the ranking. Every step keeps a boundary for the next re-run.
02
Prepare and preflight
Align fiscal periods, price timestamps, source links, and unknowns. Incomplete data is blocked first.
03
Parallel cloud review
Structured batch research via the public LoomLoom SkillBot; quote first, then confirm.
04
Local audit and ranking
Verify evidence paths, vetos, and falsification conditions. Keep only candidates ready for human deep dive.
Buffett's four filters
Look at the business
and the people first,
then at the price.
We don't let the model "pick stocks." Each filter runs its own evidence audit, and a hard local veto gates the ranking.
The framework emphasizes understanding the business model, long-term cash flow, moat, management, and capital allocation. Price judgment must bind to both a price timestamp and a fiscal period; without that anchor, the row is flagged evidence_thin in the report instead of being silently approved.
01Circle of competence & business modelFirst answer how the company makes money
02Moat & management / capital allocationFacts, not stories; vetting red flags with weak scores
03Financial quality & survival riskCash flow, leverage, dilution, and accounting red flags
04Valuation with margin of safetyNo target price; three-tier calibration (eligible / evidence_thin / vetoed)
Pick the few —
and review every one.
Download the Skill to your Agent and add your own CogFoundry credentials. From then on, just hand over a ticker list or a metrics table. Before any cloud run, the Agent will surface the current quote and wait for your confirmation.
First check whether the Skill is installed at [gold3bear/loomloom-buffett-screener](https://github.com/gold3bear/loomloom-buffett-screener). If not, install it from the GitHub Release, then use the installed "US Equity Long-Term Value Batch Screening" Skill and strictly follow its SKILL.md. Analyze the US tickers I provide, or the CSV/Excel/standard financial metrics table I upload. First check that the LoomLoom CLI is ready; if not, follow the Skill's install guide, install the CLI via `loomloom login`, and run `loomloom doctor`. Next, locally prepare, fill, and validate the data, then run preflight. Once preflight passes, generate the Market request and fetch the server quote. Show me the task count, currency, estimated cost, data-as-of time, and the main risks. Wait for my explicit confirmation in the current conversation before calling the cloud. When the run completes, perform local evidence audit, hard-veto review, and three-tier valuation calibration, then deliver an auditable Markdown report. Do not output target prices, position sizes, or buy/sell instructions.