Case Studies

A Verdigris case study is the proof artifact for a single customer engagement. It carries one anchor metric, one named operator’s voice, and one closing invitation for the reader to replicate the work in their own facility. The Notion Sales Collateral Cheat Sheet lists four published case studies (Rectifier, Cooling Tower, Fortune 50 Telecom CapEx, Fortune 50 Telecom LL97) — all pre-cell, eligible for refresh under this spec.

This cell is single-genre. Unlike slides (4 genres), one-pagers (2 genres), or whitepapers (3 genres), case studies have one rhetorical structure that consistently lands across audiences. Fragmenting into multiple genres would underspecify each variant; the single-genre framing keeps the discipline tight.

The case study is dual-target

Every case study renders to two targets, both required:

1. Web canonical — verdigris.co/resources/case-studies/{slug}. The version a prospect lands on from search, from the website nav, or from a sales link. Interactive elements (live charts, embedded video) live here.
2. PDF mirror — committed alongside examples/{slug}.html, distributed via Drive (/Verdigris/Collateral/case-study/{slug}/), attached to outbound emails. Static rendering of the same content.

The web is canonical for content; the PDF is canonical for distribution as an attachment. The PDF mirror MAY drop interactive elements but MUST NOT add or modify content claims. The composition.persuade-case-study.web-pdf-parity rule enforces this — divergence between the two is a real failure mode (a case study with one anchor metric on the web and a different one in the PDF erodes credibility on first comparison).

Required structure (single genre)

Case studies have six required sections, in this order:

1. Header — eyebrow + headline + deck + byline (customer name, segment, engagement window)
2. Executive summary — one paragraph in a teal-rule callout box; the entire case in 4-6 sentences. A reader who reads ONLY the executive summary should grasp the engagement, the outcome, and the relevance.
3. Problem — what the customer needed to solve, in their language and at their scale. Specific to this customer; never generic (“data-center reliability is hard”).
4. Approach — what Verdigris did in this engagement. Specific actions, specific decisions, specific instrumentation. Avoids “best practices” framing — a case study is one execution, not a methodology lecture.
5. Outcome — section anchored by the single bolded anchor metric. Surrounding prose carries secondary observations; secondary observations are NEVER bolded as competing metrics.
6. Quote — named operator voice; required attribution including organization (or explicit redaction).
7. Replicate — closing CTA inviting the reader to do this in their own facility. Dark-banded, action-oriented (mirrors the one-pager CTA strip pattern).

(Quote technically sits inside or after Outcome; the six “required sections” frame is Executive summary, Problem, Approach, Outcome, Quote, Replicate per the rule block.)

Decision framework: where to land in the bounds

If you cannot reduce the engagement to one bolded anchor and a 4-6 sentence summary, the artifact may need to be a brief or a short whitepaper rather than a case study. Case studies that try to cover two engagements at once consistently land as marketing rather than evidence.

The anchor metric

One bolded outcome metric per case study. This is the load-bearing rule.

A case study that presents two competing bolded metrics (“$1.3M-$3M recovered AND 21 days advance warning AND 800+ rectifiers monitored”) dilutes the proof — the reader scanning the page sees three numbers competing for attention and the win loses its definition. Pick the one number that defines the engagement; let the others support in surrounding prose.

Examples of legitimate single anchors (illustrative; numbers should be re-verified against the canonical EVD atom set in the consuming repo before any production case study ships):

• $1.3M-$3M — Fortune 50 Telecom CapEx deferral over a 3-year horizon (anchored to EVD-005; the dollar range reflects fleet-projection sensitivity, not per-facility — see Z2O-1310)
• 19% HVAC energy reduction — Fortune 50 Telecom LL97 compliance (verify against the published case study at verdigris.co/resources/case-studies/ before reuse)
• 14 days of advance warning — Rectifier failure precursor identification (anchored to EVD-104)
• Zero penalties — LL97 compliance outcome (claim type: count; pair with the regulatory window in surrounding prose)

Per-instance qualifiers (“per facility”, “per rack”, “per circuit”) are silent metric multipliers and require explicit context. A $3M per facility anchor without stating fleet size N silently multiplies the canonical claim by an unknown N. See LEARNINGS.md for the per-instance qualifier discipline (Z2O-1310 in the consuming verdigris repo).

Template vs. produced

Case studies are the highest-stakes surface for the rendering-layer-never-invents-facts principle: a fabricated quote, a guessed customer name, or a plausible-sounding metric ships as evidence and erodes trust on first comparison. The template-vs-produced contract is concrete here.

The template stage is what an agent generates from the spec; the produced stage is what a human (or evidence-grounded agent) fills in. Never ship the produced stage without source evidence for every filled placeholder — quotes require a recorded interview or written attribution; anchor metrics require an EVD canonical claim ID; customer names require co-marketing approval (or explicit redaction). A case study with a plausible-sounding placeholder still in production form is the failure mode the dual-target parity rule cannot catch — only the hand-off note can.

Quote attribution

Every customer quote requires:

1. Named individual OR explicit role label. “Mark Chu, VP of Operations” or “VP of Operations, Fortune 50 Telecom”. Never “the customer told us.”
2. Organization name OR explicit redaction. “Fortune 50 Telecom” / “Hyperscale colocation customer” / “named operator” all acceptable when the customer has not granted public attribution.

A quote attributed only to the company (“Per Acme Corp”) without a human voice fails the trust register: the case study reads as marketing rather than evidence. Quotes attributed to “anonymous customer” are similarly weak — the redaction is fine, the absence of a role label is not.

Voice recipe

Uses the existing case_study recipe in voice/recipes.yaml:

• Seren primary — customer perspective, diplomatic precision. The customer’s voice is THE primary voice in a case study; Seren’s people-intelligence translation makes the customer voice land authentically rather than as scrubbed marketing speak.
• Mike supporting — field credibility, competitor comparison context.
• Jimit accent — market-trend connection, why this win matters beyond one customer.

The recipe was audited 2026-05-02 against the new Mark Chung and Jon Chu profile YAMLs and held up — neither founder voice belongs as primary in a case study because the customer is the protagonist, not Verdigris.

Voice at a glance

A producer reading this cell should be able to answer “what voice mix am I writing in?” without leaving the page. Pulled directly from the case_study recipe and the linked profile YAMLs.

Seren — primary (Profile: voice/team/seren-coskun.yaml). The customer’s voice is the primary voice in a case study; Seren’s people-intelligence translation makes the customer voice land authentically rather than as scrubbed marketing speak. She names what is positive before what is missing and frames opinions as perspectives, which is how a real operator describes a vendor’s work.

“From my perspective”

Carries: executive summary, problem, outcome prose around the anchor metric, quote framing. The body register throughout.

Mike — supporting (Profile: voice/team/mike-mahedy.yaml). Field credibility and competitor-comparison context. Mike knows what the customer was comparing Verdigris against because he was at the same conferences, looking at the same vendor product pages.

“Interesting that it doesn’t talk about…”

Carries: approach section, the technical specifics that establish “we did this work, not just described it.”

Jimit — accent (Profile: voice/team/jimit-shah.yaml). Market-trend connection — why this win matters beyond one customer. Jimit’s outside-in voice connects the customer outcome to the broader category shift, which makes the case study feel strategic rather than anecdotal.

“Hot new update from Semianalysis…”

Carries: the closing prose around Replicate that gestures past this one engagement. Used sparingly.

Citation discipline

Inherits citation conventions from categories/whitepapers/cover.md:

• Numbered citations with full source (author, title, year, URL or report ID) — no bare domains
• arXiv preprints require year
• Government and lab reports require report number (e.g., LBNL-2001637)
• Verdigris-internal data sources cite the EVD atom or canonical claim ID where applicable, plus a plain-language description

Charts and figures use the .vd-figure pattern from cover.css (figure + caption with label + credit line). Uncredited proprietary charts fail the trust register — every chart cites its data source.

Confidentiality

Case studies default to PUBLIC tier. Customer-redacted variants (“Fortune 50 Telecom” / “Hyperscale colocation customer”) are still PUBLIC — the redaction is content discipline, not a confidentiality tier. A case study that needs to be CUSTOMER-CONFIDENTIAL is probably a brief or an internal team retrospective, not a case study.

Spacing and typography

All values from tokens/spacing/print.json (frame primitives) and build/print/case-study.css. Case studies use the same paper-page frame as whitepaper inside-pages: 0.95-1.0in margins, Inter 11pt body, 1.6 line-height, 6.6in max content width.

Why fixed point sizes (not ranges)

Print artifacts assume a constant reading distance — roughly 18 inches for a printed case study held at a desk or pulled out of a folder. Fixed point sizes lock the visual hierarchy at that distance: every printed copy and every downloaded PDF renders identically, including the 56pt anchor metric that has to land the same way for every reader. The slide cells use ranges (32-44pt headlines, 18-22pt deck) because projection distance varies — a boardroom monitor at 8 feet versus an auditorium screen at 30 feet calls for different sizes, calibrated per venue. Don’t borrow the slide-cell range pattern back into case studies; the read-distance assumption is fixed, and the dual-target web/PDF parity rule depends on the typography scale staying constant across both renderings.

Reference exemplars

The first refresh under this cell should be Rectifier — its anchor metric is specific, its customer is willing to be named in some channels, and its problem-approach-outcome arc is the cleanest of the four for testing the spec.

What this cell does NOT cover

• Multi-customer reports. A “5 customers, 5 stories” report is a brief or a whitepaper, not a case study. Case studies are one engagement.
• Internal post-mortems. When the audience is Verdigris-only, use categories/slides/internal-team.md (deck) or a future internal-retrospective sub-cell.
• Pre-engagement projections. A “what we expect to deliver” artifact is a pilot kickoff deck (categories/slides/pilot-kickoff.md), not a case study.
• Marketing campaigns / collateral series. When multiple case studies appear together as a campaign, this cell governs each individual case; a future “case-study-series” cell can govern series-level patterns when a real surface needs one.

Why this guide exists

Verdigris has four published case studies and at least two more in flight. The published four don’t share a structural template — anchor-metric placement varies, quote attribution varies, executive-summary discipline varies. Codifying the single genre and the dual-target render makes refresh and net-new case studies a template-fill, not a blank-page redesign.

The single anchor metric rule resolves a specific pre-existing tension: the homepage and platform pages already have a “per-facility / per-rack” qualifier risk that PR #281 surfaced (see Z2O-1310). Case studies are the highest-risk surface for that failure mode because the bolded outcome is what the reader takes away. Forcing the single-anchor discipline at the rule level catches the multiplier risk before it ships.

The web/PDF parity rule reflects how case studies actually circulate: a prospect lands on the web canonical, downloads the PDF for forwarding, and the two artifacts diverge over time as edits land in one but not the other. Codifying parity as a rule (not just a process) makes divergence visible in evaluator runs.

See also

• workflows/sales-collateral — production guide spanning all collateral types; see the versioning-vs-refresh section for when to bump the version vs. edit in place
• categories/slides/ — sister cell; sales-collateral universals (logomark, roles, dates) inherited here
• categories/whitepapers/cover — sister cell; print-rendering patterns + citation discipline inherited
• categories/one-pagers/guide — sister cell; CTA-strip pattern inherited
• build/print/case-study.css — reference stylesheet (dual-target)
• voice/recipes.yaml — case_study recipe (Seren primary)