Observed
Directly supported by a preserved source or authorized measurement
LVIS · version 1.0 · baseline 2026-09-01
Public evidence, turned into a claim you can defend.
This is an operating system for company intelligence, not a prompt. It takes lawful public evidence about a business and carries it through one unbroken chain — source, observation, claim state, confidence, mechanism, financial range, service fit, smallest useful next step.
Every material sentence is labelled with how well it is known. Observed is not calculated, calculated is not inferred, and an inference never gets to wear an observation's clothes.
What the library contains
23
Source documents
268
Catalogued tools and sources
20
Financial impact models
32
Safe query families
Counts are derived from the corpus at load time, not declared by hand. Source: the 23 Markdown files served at /v1/docs, version 1.0, research baseline 2026-09-01.
The discipline
Six labels, and no sentence escapes one
The label belongs on every material sentence, database claim, financial input and outreach fact. It is the difference between a finding an executive can act on and a guess that costs you the room.
Calculated
Deterministic transformation of cited inputs
Inferred
Best explanation supported by evidence but not directly observed
Hypothesized
Testable possibility; not yet supported enough to prefer
Unknown
Necessary information is unavailable or inaccessible
Contradicted
Credible evidence conflicts with the claim
The grammar
Five change states, and a C4 is not a better C1
Change state describes how deep a change goes, not how good it is. Selection depends on evidence, fit, urgency, reversibility, capacity and time-to-useful-evidence — never on which state sounds most ambitious.
| Code | State | Meaning | Audit question |
|---|---|---|---|
| C0 | Baseline | Current/default state; no meaningful change | What exists now and how does it normally behave? |
| C1 | Minor Change | Local, low-depth adjustment; limited dependencies | What can improve through tuning or one-step repair? |
| C2 | Major Change | Material change across several components or outcomes | What requires coordinated redesign or meaningful investment? |
| C3 | Structural Change | Rules, architecture, incentives, ownership, or feedback loops change | What must be rebuilt at the system level? |
| C4 | Paradigm Change | Unit of value, category, business model, or governing assumption changes | What new demand curve makes the old comparison less relevant? |
Three ways in
Read it, run it, or call it
Run the audit
Nine phases, each with its questions, collection lanes, required controls and deliverables. Preflight first; a phase is complete, N/A with a reason, unknown, or blocked.
Score what you found
Four rubrics, executable. Weights come from the corpus, hard gates override the number, and a missing input is reported as missing rather than counted as a zero.
Call the API
Every part of the library is public JSON: 23 documents, 268 tools, 216 expansion lenses, and the rubrics as POST endpoints.
Before anything else
The rules that bound every run
- Passive public-source research is the default. No login bypass, exploitation, credential collection, intrusive scanning, pretexting, or contact-data guessing.
- A source is not a finding. Preserve its URL, publication/observation time, retrieval time, excerpt or structured fact, and content hash where practical.
- Keep Observed, Calculated, Inferred, Hypothesized, Unknown, and Contradicted separate.
- Never present an inferred security weakness, executive motive, revenue figure, sentiment percentage, or causal effect as observed fact.
- Dollar values are low/base/high scenarios with visible formulas and inputs. If inputs are unavailable, say not estimable and list the missing inputs.
- Respect robots.txt, source terms, API limits, privacy law, retention limits, and opt-outs. Public visibility does not automatically authorize collection or marketing use.