Signal
A dated public event or piece of evidence. CPI preserves the source, source class, entity, date, domain, and evidence/significance score.
Loading intelligence…
Resolving the latest stored evidence and product state.
MethodologyGuest mode
CPI methodology
The value is not collecting more headlines. It is making the reasoning inspectable.
CPI moves from public evidence to an evidence-backed industry thesis while keeping raw evidence, automated interpretation, experimental forecasting, and expert judgment visibly separate.
Underlying event
CPI clusters duplicate and syndicated coverage so five articles about one announcement do not become five independent events. Corroboration is preserved separately from event count.
Trend
A pattern built from connected underlying events and measured source-state change. Trend Engine V2 calculates velocity, acceleration, breadth, durability, verified progression/reversal balance, primary support, and counter-signal pressure before automated interpretation.
CPI Analysis
Automated interpretation through CPI's clinical-product lens, including workflow, intended users, implementation, evidence, human factors, and adoption.
Forecast
A measurable hypothesis with probability, deadline, supporting evidence, counter-signals, and a resolution rule. Early forecasts are explicitly experimental and are backtested over historical cutoffs.
CPI Reviewed
A separate human layer. An authorized CPI expert can publish a CPI Take, score the clinical-product Lens qualitatively, and add a forward CPI Outlook. Machine analysis is never silently relabeled as expert judgment.
Conviction ledger
Published CPI Outlooks preserve the base case, confidence, expected window, strengthen/weaken conditions, and eventual outcome so forward-looking judgment can be resolved instead of quietly rewritten later.
Discovery
CPI can look beyond predefined themes
The platform starts with transparent CPI trend definitions, but it also tests co-occurring technologies, themes, settings, and organizations for candidate patterns. Dynamic candidates stay internal until a CPI expert accepts them, preventing an automated cluster from becoming a public thesis by accident.
Trend confidence
What moves a thesis
Evidence count alone is not enough. Trend Engine V2 compares 7-, 30-, and 90-day canonical-event windows and measures velocity, acceleration, company and evidence-class breadth, durability, primary-source support, verified progressions and reversals, and counter-signal pressure. Regulatory, clinical, research, capital, safety, coverage-policy, commercial, and market evidence remain distinguishable so one evidence class cannot impersonate broad convergence. The measurements are deterministic; Foresight explains them afterward rather than inventing the movement.
Hybrid retrieval
Search uses language, graph context, and optional semantics
Research and Ask CPI use lexical expansion plus the CPI entity graph. An optional public-signal embedding index can improve semantic recall. Private Product Workspace document text is intentionally excluded from the public semantic embedding pipeline.
Forecast discipline
No made-up accuracy claims
Forecasts are stored with creation time, probability, horizon, and resolution criteria. Historical backtests and resolved outcomes feed calibration metrics, so CPI can build a real track record instead of marketing predictive accuracy before it exists.
Expert separation
Emma is a layer, not a hidden prompt
CPI's expert knowledge bank contains reviewed principles and decision rules. Human-reviewed perspective is visibly labeled and can correct automated interpretation without rewriting the observed evidence.
Private workspace
Public intelligence works without confidential uploads
Private Product Workspace is optional. Extracted document text can be application-encrypted and is analyzed server-side with deterministic methods. It is not added to the public intelligence graph or public embedding index. V1 is not intended for PHI.
Strategic Radar
Attention is not attractiveness
Strategic Radar ranks which companies deserve investigation based on measured evidence change, breadth, primary-source support, verified state transitions, connected Trend Engine V2 movement, and capital activity. Negative reversals can increase attention while weakening trajectory. The score is not a valuation, expected-return estimate, investment recommendation, or probability of company success.
Company trajectory
Direction stays separate from urgency
Foresight classifies company trajectory as Advancing, Building, Mixed, Pressured, or Quiet from source-backed evidence. Progressions, reversals, safety or coverage changes, and cross-evidence confirmation remain visible so a high-attention company cannot be mistaken for a positive recommendation.
Comparative company intelligence
Urgency and evidence strength are different measures
Strategic Attention answers which companies deserve investigation. Comparative Evidence Position answers which observed company trajectory is stronger on measured progression, breadth, primary-source support, Trend Engine V2 positioning, momentum, and counter-pressure. A pressured company can therefore remain high-attention without leading the comparison.
Peer context
Peers come from evidence overlap, not a hard-coded competitor list
Foresight suggests adjacent peers from overlap in observed domains, measured trends, and evidence dimensions. The result is a research comparison set, not a market-share ranking or a claim that every suggested company competes directly in every product line.
Decision Intelligence
The evidence state is assembled before automated synthesis
Decision Intelligence combines direct source evidence, measured company state, Trend Engine V2, Strategic Radar, verified counter-signals, and evidence-based peer context into a structured decision packet first. OpenAI receives that packet afterward and is instructed not to recalculate or override the deterministic measurements.
Conclusion changers
A useful brief should say what could prove the current read wrong
Every decision packet separates the strongest supporting evidence, counter-signals, unresolved questions, next milestones, and the events that would strengthen or weaken the current interpretation. The output remains a diligence and strategy aid, not a valuation, expected-return estimate, investment recommendation, or probability of company success.
CPI Lens
Human judgment is structured, not hidden in a paragraph
Published CPI Reviews can address clinical value, workflow fit, implementation readiness, adoption friction, evidence maturity, economic alignment, and competitive implication. The reviewer can mark a dimension favorable, balanced, watch, concern, or unknown and explain why. These are human judgments, not automated scores.
CPI Outlook
Forward-looking calls have to be testable
A CPI Outlook states a base case, confidence, expected next state and time window, what would strengthen the view, what would weaken it, and what CPI is watching next. Resolved Outlooks remain in the conviction ledger as correct, partial, incorrect, or inconclusive rather than being rewritten after the fact.
Editorial Intelligence Engine
Foresight drafts. CPI decides.
Foresight continuously scores evidence changes for editorial significance, matches them against the CPI Pattern Bank, and can prepare internal Draft CPI Takes and Draft CPI Outlooks from the source-linked evidence packet. Drafts are never customer-facing CPI judgment. Only an authorized CPI reviewer can edit and approve a draft into CPI Reviewed content.
Editorial memory
CPI review improves future selection without rewriting history
Approvals, edits, rejections, and needs-more-evidence decisions are stored as structured editorial feedback. That feedback adjusts how strongly matching patterns are prioritized in future draft generation. Published Takes and resolved Outlooks remain immutable historical records of what CPI approved at the time.