The trust layer. 49% of slides ship it.
176k chrome atoms — subtitles, source notes, disclaimers. Almost half of every slide carries some, and almost none of it argues. Chrome is the structural layer that says 'you are here, and someone is responsible'. Present it badly and the chart becomes a sketch.
- Source note 118k · 35% who counted, when, where. Turns a number into evidence.
- Disclaimer 46k · 13% caveat, qualifier, legal — what the number does NOT say.
- Subtitle 12k · 3% the second line under the title. Carries n=, period, units.
Attribute. Caveat. Anchor.
Three jobs the chrome layer covers. The corpus does the first reasonably (35%); the second selectively (13%); the third almost never (3%). The deck that gets all three is in the top decile.
- Attribute (source-note) 35% adoption "Source: X, Y, Z, dated W". The trust handshake. Without it the chart is a drawing.
- Caveat (disclaimer) 13% adoption "Estimates assume A. Excludes B. Period C." The boundary of the claim.
- Anchor (subtitle) 3% adoption "Q3 2024 · global · n=12,349". The metadata the title cannot afford to spend words on.
Chrome is a contract with the audience.
The audience does not read chrome. They glance at it, register that it exists, and trust the slide more. Remove it and the same slide reads as advocacy. Chrome is the cheapest credibility multiplier in the deck — and it lives where nobody looks.
- What chrome promises three things someone counted this, someone is responsible, the boundary of the claim is disclosed.
- What it does NOT do argue chrome cites; it does not advocate. A persuasive footnote is a contradiction in terms.
- When it earns its slot every slide tables, charts, metrics, claims — all four warrant chrome. Decoration does not.
Where the corpus ships evidence without chrome.
For each slide-type, the gap between "% with evidence" (chart, table or metric) and "% with chrome" (source-note, subtitle or disclaimer) is the slide-type's editorial debt. Positive gap = unfinished trust work. The eight slide-types below carry the largest gaps.
- Appendix data n = 17,12685% evidence 65% chrome+20pp gap
- Traction n = 10,06657% evidence 54% chrome+3pp gap
- Industry trends n = 8,21468% evidence 71% chrome-3pp parity
- Comparison table n = 5,45659% evidence 66% chrome-7pp parity
- KPI overview n = 26,36764% evidence 76% chrome-12pp parity
- Data table n = 10,05270% evidence 85% chrome-15pp parity
- Key takeaways n = 10,05224% evidence 44% chrome-20pp parity
- Market sizing n = 5,13557% evidence 78% chrome-21pp parity
Six corpus chrome lines. Two earn their slot.
Source-notes range from rigorous (named source + date + scope) to evasive ("internal analysis"). The reader cannot verify what the chrome will not specify.
-
Source: Gareth et. al.: An Introduction to Statistical Learning, 2019; Trevor et. al.: The Elements of Statistical Learning, 2019
Strong — verifiabletwo named authors, two dated works — a reader can pull either book and verify.
academic source -
Source: Google-commissioned Dynata SEA-6 Digital Merchant Survey 2021.
Strong — verifiablecommissioning party + research firm + scope + year. The cite earns its line.
commissioned research -
Note: Adjusted EBITDA excludes one-off expenses and non-cash items. Unaudited quarterly results.
Medium — partialdiscloses the adjustment but not the magnitude — reader cannot reconstruct the unadjusted number.
methodology note -
(1) Source: McKinsey & Company... (8) Source: Gartner CIO Agenda 2021...
Medium — partialeight footnotes referenced, but the slide hides the body text — chrome without anchor.
multi-footnote -
(1) See Reference / Definitions
Weak — uncheckablea pointer to a glossary that may or may not exist. The reader is told to look elsewhere.
pointer-only -
Private & confidential. Do not reproduce without express written permission of AvidXchange.
Weak — uncheckablelegal boilerplate dressed as chrome. Adds no information about the data on the slide.
boilerplate
The corpus is twice as good at sourcing charts as diagrams.
Chrome adoption is not flat — it scales with the perceived 'numericness' of the slide. Numbers force chrome; structure escapes it. The gradient is a 32-point trust gap.
Top tier (charts, tables, metric) sources 47-57% of slides. Bottom tier (diagrams, images) sources 24-29%. The mid is text at 38%. The gradient is steep and consistent across the corpus.
The pattern is not random — it tracks audience expectation. A reader instinctively asks 'where did that number come from?' and accepts the answer as ritual. The same reader looks at a diagram or photo and assumes the writer's authority covers it. Chrome appears where the audience demands it.
Closing the bottom-tier gap is the corpus's largest unfinished trust work. A diagram source-note costs three lines. The credibility lift is closer to twenty-five points.
- Charts 56.5%
- Tables 49.5%
- Metric 46.6%
- Text 37.7%
- Diagrams 29.2%
- Images 24.3%
Chrome owns the bottom edge.
Across 176k chrome atoms — subtitles, source notes, disclaimers — the bottom 10% of the canvas is where almost everything lives. The eye does not visit it. The reader trusts it.
How to read this
- Heatmap
- Surface coverage. Each bbox in the corpus contributes to every 40×22 cell it overlaps; opacity is power-scaled (γ=0.6) so subtle bands stay visible. Darker = more atoms cover that point.
- Sample box
- The modal cluster. Atoms grouped by centre into 8×8 bins; the densest bin is plotted at its centroid, sized by the average width × height of atoms in it. Reads as: "the typical atom of this family looks roughly this big, here."
- Source
corpus.components.bbox— atoms with measured bounding boxes only. Inline mentions extracted from body text are excluded from this analysis.
Chrome tracks accountability.
Chrome presence is the corpus's accountability thermometer. Disclaimer slides hit 89%; data tables 85%; appendix data 65%. Cover and section divider — under 16%. The pattern: the more the slide commits to numbers, the more chrome it ships. Where there is nothing to source, there is nothing to chrome.
Chrome follows charts more than tables.
Chrome-bearing slides are evidence-heavy: 38% ship a chart, 27% ship a table, 25% ship an image. Where the chart is, the source-note follows — and where the table is, the disclaimer follows. The corpus already wires the trust layer to the evidence layer; the gap is in coverage, not pattern.
- 01 Title 96.6%almost universal — chrome anchors a titled slide
- 02 Chart 38.3%— charts attract sources. The strongest chrome-bearing combo.
- 03 Paragraph 33.3%narrative + footnotes — the report pattern
- 04 List 31%list of points + source — the briefing pattern
- 05 Table 26.9%table + disclaimer — the finance pattern
- 06 Image 25.2%image + caption — the documentary pattern
- 07 Callout 15.1%callout + source — extracted claim with provenance
- 08 Metric 10.2%metric + chrome — the trust loop. Should be 100%.
Five chrome failures. All erode the contract.
Chrome failures are quiet — the audience does not protest a missing source-note, they just trust the slide less. Five corpus-recurring failures, each one cheaper to fix than to suffer.
-
∅ The unsourced slide
A slide with numbers, claims, or named entities — and no source.
65%- Bad
- Chart of "market growth" with no source line.
- Fix
- Three-line footer: source, date, methodology. "McKinsey · 2024 · 6,500 respondents".
-
¶ The legal paragraph
A 200-word disclaimer in 6pt type. Nobody reads. Defeats its own purpose.
~8%- Bad
- Six lines of "forward-looking statements" boilerplate at 6pt.
- Fix
- One line of substantive caveat ("Pro-forma. Excludes restructuring."), boilerplate moves to appendix.
-
◷ The fake source
Cites "company analysis" or "internal data" — uncheckable, uninformative.
~12%- Bad
- "Source: Internal analysis"
- Fix
- Cite the actual data: department, dataset, period. "Source: Sales ops · CRM extract · Q3 2024".
-
⌐ The broken citation
A URL or footnote anchor that does not resolve.
common in PDFs- Bad
- Footnote "(1)" with no corresponding "(1)" in the body, or a URL truncated mid-line.
- Fix
- Audit cross-references before export. Run footnotes through a link-checker.
-
⊘ The orphan disclaimer
A disclaimer for a claim that is no longer on the slide.
~5%- Bad
- "Estimates assume FY2023 baseline" — but the chart on this slide shows FY2024.
- Fix
- Disclaimers ship with their claim, and only their claim. Audit when slides are reordered.
Four rules. Honour the contract.
Chrome is the structural layer that turns a slide into evidence. These four rules protect the contract every slide makes with its audience.
- 01
Source every claim
Numbers, named entities, market sizes, growth rates — all warrant a source. Three lines of footer carry the trust of the entire slide.
35% currently sourced - 02
One disclaimer per substantive caveat
Disclaimers are useful; legal-paragraph wallpaper is not. One substantive caveat per material claim. Boilerplate moves to appendix.
13% slides have disclaimers - 03
Subtitles carry the metadata
Title carries the takeaway; subtitle carries n=, period, scope, units. The corpus underuses this slot — only 3% of slides ship a subtitle.
3% slides ship subtitles - 04
Chrome lives at the foot — never anywhere else
Source notes at top of slide are intrusive. Disclaimers in headers are confusing. The bottom 10% of the canvas is the universal contract zone.
bottom 10% of canvas