Method
How Parallax reads a company.
Parallax doesn't treat every mention as an equal data point. It asks who is making a claim, what changed, and how that compares with what's normal for this company.
The evidence layer
Parallax maintains a continuously updated evidence layer around every company on a watchlist, public or private.
- Global and financial news
- Industry and niche trade publications
- Regulatory, legal and other public records
- Financial filings
- Company announcements and investor materials
- Earnings releases, transcripts and management commentary
- Analyst, specialist and short-seller research
- Advocacy, labor and stakeholder publications
- Public forums and professional communities
- Public employee commentary and review platforms
- Materials you authorize
Forums and employee sources count only when the material is publicly accessible or lawfully provided by you. Private Slack channels, internal email and restricted communities aren't part of it.
What it weighs
- Who is making the claim, and which stakeholder group they represent
- Whether different sources are genuinely independent
- Whether previously separate groups are converging around the same concern
- Whether causal explanations are changing
- Whether speculative language is becoming more certain
- Whether important claims or voices are disappearing
- Whether the narrative has moved materially away from the company's historical baseline
Five classes of language signal
- Epistemic
- How sure the language is. Hedged language hardening into statements of fact, and context dropping out. Measures: Certainty drift, modal shift, evidence density, omission against expected context.
- Discourse
- How the story is framed. An operational problem turning into a question of accountability. Measures: Framing shift, causal attribution, moral and emotional register.
- Trajectory
- How far it has traveled. A niche complaint moving into trade and national press. Measures: Velocity, acceleration, narrative mutation, migration across source tiers.
- Linkage
- Whether separate groups are connecting. Isolated complaints arriving at the same cause. Measures: Cross-source similarity, coordinated phrase reuse, content clustering, source independence.
- Actor
- Who is saying it. Anonymous posts giving way to people on the record. Measures: Who originates, who amplifies, new actor classes, anomalous involvement.
Read against each company's own baseline
A reading only means something relative to what's normal for that company. Parallax keeps an evolving narrative baseline for each one, then evaluates new evidence against it and against your risk hypotheses. A change reaches a person when it crosses the materiality and escalation rules set for your workflow.
Every finding keeps its provenance
You can inspect the original sources behind a finding, see why it was considered relevant, and question it through Parallax Analyst, including testing innocent explanations and evidence that contradicts the first interpretation. If the record can't support an answer, it should say so.
When coverage is thin
Parallax evaluates the depth and diversity of the available evidence. When coverage is limited, it makes that uncertainty explicit rather than manufacturing confidence.
What it doesn't do
It doesn't make the decision. It doesn't claim that narrative movement proves deterioration. It tells your team where to look sooner and shows why.