Editorial policy

AI authorship and accountability

Stocksembly research is generated, challenged, audited, and synthesized by software-driven AI roles. Their names and visual personas do not represent human analysts. Unless a report explicitly says otherwise, readers should not assume that a human editor independently reviewed every sentence or source before publication.

SERN operates the service and is responsible for maintaining the product rules described here. Automation reduces neither model risk nor the reader's duty to verify material facts.

Evidence and provenance

Material claims are expected to carry accepted evidence artifact identifiers. Stored evidence includes run and snapshot identity, content hashes, retrieval timing, and source locator metadata so the publication boundary can verify lineage and detect cross-run or altered content.

This provenance is a control, not a guarantee of truth. Regulatory filings can be amended, providers can revise data, publisher labels can be imperfect, and a cited passage can still be interpreted incorrectly.

Disagreement before consensus

Full-committee research separates specialist memos, department consolidation, blind challenges, owner responses, ballots, audits, and chair synthesis. The process is intended to preserve opposing evidence, revisions, abstentions, open questions, and falsifiers rather than manufacture unanimous confidence.

Multiple AI roles do not constitute independent human institutions and may share model weaknesses, training biases, or source errors. Agreement among agents is not proof that a conclusion is correct.

Publication quality rules

A report can be persisted only after the required workflow artifacts and source lineage pass structural checks. Public bilingual fields are then evaluated for evidence ownership, supported numbers, comparator typing, anticipated-question evidence, unsafe investment language, repetition, information value, and leakage of internal metadata or provider terminology.

Content quality findings do not block publication. Targeted repairs and wording normalization use existing evidence without inventing new references or numbers. Remaining concerns are recorded and the report is published with disclosed limitations. Source authentication, run isolation, and storage integrity checks still apply.

Investment language and reader safety

Public output must not present an immediate buy or sell instruction or a guaranteed return. Stocksembly does not provide target prices or individualized recommendations. Scenario values, valuation frameworks, confidence labels, and investment actions are analytical constructs, not trade instructions.

Time, limitations, and updates

Every report should be read as a snapshot of the evidence available to its run. Market prices, filings, estimates, news, and company conditions can change after publication. Missing, stale, unavailable, or rights-restricted capabilities should be surfaced as coverage or limitation information where the report schema supports them.

A change in the market is not automatically a correction to an earlier dated analysis. Factual, persistence, lineage, or projection defects are handled under the corrections policy.