Research Substrate
Performix makes a transparency claim: when the product surfaces an evidence-backed insight or a survey item, there should be a public path from that surface back to the peer-reviewed source. This part documents that chain — not as marketing architecture, but as an auditable data model.
The chain has four entity types:
Source → Finding → Construct → Item
Each link is explicit. Nothing in the adaptive layer (diagnostics, insight player, action loop) should reference a construct that cannot be traced backward through this graph.
Source
A Source is a citable research artifact: peer-reviewed paper, meta-analysis, authoritative handbook chapter, or — where honestly labeled — a high-quality working paper.
Sources live in the anchor registry (cams-anchor-registry.json in the Performix repo, PFX-53). Each entry carries:
- Stable identifier and slug
- Bibliographic metadata (authors, year, DOI where available)
- Topic tags and CAMS dimension mapping where applicable
- Quality tier (peer-reviewed vs working paper vs practitioner synthesis)
Sources are not survey items. They are the evidentiary floor.
Finding
A Finding is a discrete, citable claim extracted from one or more Sources — stated narrowly enough to support or refute a measurement decision.
Examples of finding granularity (illustrative placeholders):
- "Perceived organizational support predicts discretionary effort above and beyond job satisfaction in field samples."
- "Goal specificity improves performance when commitment is high; under low commitment, specificity can backfire."
- "Role ambiguity suppresses both task and contextual performance; effects are larger when Support is low."
Findings aggregate in science-library-seed.json (PFX-58). Each finding references its source IDs and carries a confidence / replication posture where the literature supports one.
Findings are not constructs. A finding is what the literature said. A construct is what we measure.
Construct
A Construct is an operationalized latent variable in the CAMS model — the thing a diagnostic dimension or subdimension intends to measure.
Constructs bridge findings to items:
- Defined in plain language (what it is and is not)
- Linked to one or more supporting findings
- Mapped to a CAMS dimension (Capability, Alignment, Motivation, Support) or cross-cutting theme
- Associated with expected directionality (higher = better, unless inverted)
Constructs are where theory becomes measurement intent. Two findings might support the same construct; one finding might inform multiple constructs with different weight.
Item
An Item is a survey or embedded-feedback prompt shown to a user — the atomic unit Reincarnation selects, the insight player narrates, and the action loop follows up on.
Each item:
- References its parent construct(s)
- Has calibrated wording for the target population (leader vs individual contributor, etc.)
- Carries metadata for adaptive selection (difficulty, information value, fatigue budget)
- Links to public evidence pages (
/learn/evidence/[id], PFX-65) when the chain is complete
Items are the only entity type most users ever see directly. The substrate exists so that visibility is not arbitrary.
The full chain (example shape)
Source: [Author et al., Year] — DOI placeholder
└─ Finding: "Role clarity mediates performance variance in cross-functional teams"
└─ Construct: role_clarity (Capability subconstruct)
└─ Item: "My team knows what success looks like for our role this quarter"
└─ Insight card / diagnostic surface (adaptive layer)
The Edmondson (2018) chain is live — browse the drill-down from Source through survey items. Other sources in the seed resolve to finding lists; full construct/item chains ship as the corpus grows.
Two renderers, one corpus
The same JSON substrate feeds:
| Renderer | Audience | Route |
|---|---|---|
| Reference layer | Public web, SEO, partners | /learn/*, /learn/evidence/[id] |
| Adaptive layer | Leaders in product | /diagnose, insight player, action loop |
Markdown parts (like this guide) link to anchor registry DOIs. Evidence chain pages read from seed JSON at build time. No duplicate authoring in consumers — Performix repo owns the SSOT; performix-site and PA-site syndicate.
What ships when (honest sequencing)
| PFX card | Substrate deliverable | Public surface |
|---|---|---|
| PFX-53 | Anchor registry | Citations in guide parts + glossary |
| PFX-58 | Library v0 seed | Evidence chain pages |
| PFX-59 | Item seeds | Footnotes in this part |
| PFX-60 | Intervention linkages | Part VI + /learn/interventions |
| PFX-65 | Evidence page template | /learn/evidence/[id] |
This part (VII) is intentionally drafted last in the editorial cadence — substrate transparency after the model exists.
Pre-chasm honesty
Counts on marketing surfaces must match seed data. This guide will not claim "847 cited sources" until the anchor registry contains 847 entries. Stats strips on evidence pages will read from seed.stats — honest counts only (brief-06 discipline).
Counts on the evidence drill-down stat strip reflect the current V0 seed (59 sources · 72 findings · 3 constructs · 4 survey items as of PFX-58) — honest numbers only.
Further reading (external)
Selected methodology and bibliography will syndicate to peopleanalyst.com/research as manifest entries land (PFX-64).
For the AI-human-interaction analog to this reference layer, see the AI Human Interaction Guide on peopleanalyst.com — same _parts.ts registry pattern, different domain.
Status: draft. Substrate architecture documented; Edmondson evidence chain browsable at /learn/evidence/source.edmondson_2018_fearless_organization.