Appendix C. Industry Learnings, visualization library, exports, intake and notes
Five modules added after the first prototype. Each follows the same rules as the rest of the spec: evidence before adjectives, anonymity enforced by the engine, and nothing client-facing without partner approval.
C.1 Industry Learnings (Firm HQ)
Purpose. Turn the firm's book of business into reusable insight: what recurs, what differs by sector, what is changing, and what the firm can publish.
Inputs. Only k-safe aggregates from engagements whose client contract opts in to benchmarking. Respondent-level data never leaves a client workspace.
Rules.
| Rule | Value |
|---|---|
| Minimum contributing clients per group | 5 |
| Minimum respondents per group | 1,000 |
| Statistic | Client-weighted median: each client counts once, so none can dominate |
| Withheld groups | Shown by name and client count only; no scores, no findings, no titles derived from their data |
| Client names | Visible to partners in the contributor table only; never in learnings, charts or exports |
Learning categories (grouped the way consultants use them):
| Category | What it answers | Example |
|---|---|---|
| Industry learnings | Where a sector is strongest and weakest against the book | Healthcare: strongest on meaningful work, weakest on trust & candour |
| Recurring patterns | Which diagnoses recur, and where | The middle-layer squeeze in 16 of 24 clients |
| Universal levers | Which practices are top-five drivers almost everywhere, and which only in one sector | Priority clarity is a top-five driver in 88% of clients |
| Emerging signals | What is moving across fielding years | The inward turn: external orientation falling since 2024 |
| Themes by service line | Open-text themes across every survey, grouped by the offering that answers them | Operating model: handoff failures, a top-three theme in 16 of 24 clients |
Thought leadership. Any publishable learning can seed a piece (article, benchmark report, webinar, LinkedIn post, roundtable). Pieces move Idea → Drafting → Partner review → Approved. A linter blocks client names, site labels, withheld cohorts, and numbers that do not trace to a linked learning; export is blocked until it is clean.
Engagement benchmark. Each engagement's benchmark is computed from the same book: opted-in peers in the relevant sectors, excluding the client itself.
C.2 Visualization library (Firm HQ)
37 chart types in ten families: Comparison, Ranking, Change over time, Composition, Distribution, Survey response, Relationship, Matrix, Flow & hierarchy, Headline & text.
- Each type declares the data shapes it can draw (category, matrix, Likert, XY, distribution, hierarchy, flow, KPI, text), when to use it, when to avoid it, and how it exports to PowerPoint.
- Renderers are pure SVG with no runtime state, so the same code draws on screen, in slide previews and in PNG/SVG exports.
- Hover layer on every mark. Suppressed cells are hatched, never blank or zero.
- Default series palettes pass the colour-blind and normal-vision separation checks. The brand-kit editor re-runs the checks on any custom palette.
C.3 Brand kits and exports
Brand kits set who is on the page (firm only, co-brand, client only), accent, text and background colours, a fixed-order series palette, headline and body fonts (limited to faces installed with Office), footer, and firm and client logos. The software vendor never appears.
| Export | What it contains |
|---|---|
| PowerPoint | Brand master (accent rule, brand line, logos, footer, slide numbers); title slide with the anonymity statement; one slide per chart with action title, takeaway, source line and speaker notes. Native, editable PowerPoint charts for bars, columns, grouped and stacked bars, lines, areas, doughnut, scatter, bubble, radar, histogram and Likert; native tables for heatmaps and bar tables (paginated with the header repeated); images for the rest |
| Excel | About sheet (scope, anonymity rule, brand, contents) and one styled sheet per chart with source, header row in the brand colour, number formats, filters and frozen header |
| Power BI | Star-schema CSVs (engagement: FactScores, DimUnit, DimTarget, DimWave, FactThemes, FactDrivers; book: FactBookMedians, FactLearnings), VisualData for every chart, a report theme JSON in the brand's colours and fonts, DAX measures and an import README |
| PNG / SVG / CSV | Any single chart, rendered in the chosen brand kit |
Every export contains aggregates only. Distributions export summary statistics, never respondent rows. Firm-book exports are blocked if any text names a client or site.
Report builder. Decks are lists of slides bound to a dataset and a chart type. Consultants edit the action title, takeaway and speaker notes; swap the chart among types compatible with the data; sort, toggle labels, reorder, duplicate; and choose the brand kit. A check flags any number in a title or takeaway that is not in the chart's data.
C.4 Client intake (Engagement)
Seven sections, worked through with the sponsor before configuration: purpose and decisions, audiences, scope and population, hypotheses, constraints and sensitivities, deliverables and branding, success and timeline.
- Recommendations follow from the answers: pack and modules, anonymity mode, translation review, shift-cut simulation, deliverable set-up, timeline checks.
- Hypotheses carry a test (leader–frontline gap, top driver, tenure gap, common theme, below benchmark, site gap) and are evaluated against the data: Supported, Partly supported or Not supported, with the evidence line.
- Client view. When shared, the sponsor sees "Goals & scope" in the portal, can confirm each section and leave comments that appear in the firm's intake.
- Brief downloads as Markdown.
C.5 Team notes and consultant insights
- Notes attach to an insight, slide, learning, intake section, thought-leadership piece or the engagement. They carry author, date, tags and pin state, and are firm-only: never on the portal or in exports.
- A note can be promoted to a draft insight. Consultants can also write insights from scratch.
- Consultant insights attach evidence from the engine (score or % favourable for any k-safe cut, driver share, theme, quote). Values resolve live, so claims can quote them exactly.
- They then follow the same linter and approval flow as AI drafts. No insight can go to review without at least one piece of bound evidence.