@nurix/etna 0.19.49 → 0.19.51

package/data/harnesses/code/skills/design/data/corpora/operations-review/corpus.md

@@ -1,87 +0,0 @@
----
-id: operations-review
-class: c1
-sourced: 2026-07-12
-source: "Distilled from AML/fraud alert vocabulary (tookitaki.com, veridas.com) and chargeback reason-code taxonomy (chargebacks911.com, chargebackgurus.com); number-shape recipes informed by the FiFAR fraud-review-queue research artifact (Kaggle/Springer Nature, structure only) and general seeding practice (Salesforce Trailhead seeding basics, Analytics8). See docs/research/design-skill-phase1/c1-demo-data.md."
-license: open
----
-# Operations review — demo corpus
-
-A fraud / trust-and-safety case queue. `people.csv` is the internal review team (analysts, compliance, risk ops — the ones who work the queue); `companies.csv` is the merchant accounts under review (the flagged parties whose transactions generate the queue, never staff). Every activity string below is written as an analyst-facing case note, not a system log line.
-
-## Activity vocabulary (25)
-
-1. Card testing pattern detected — 14 declined auths in 90s, same BIN
-2. Velocity alert: 6 transactions over $500 within a 10-minute window
-3. Structuring flag — three deposits of $2,950 within 24h, single account
-4. Geo-mismatch: billing address in TX, IP geolocation resolves to Lagos, NG
-5. Chargeback filed — reason 10.4, other fraud in a card-absent environment
-6. Dispute reason 4837: cardholder denies authorizing a $1,240 charge
-7. New account, first transaction $4,800 — no prior purchase history
-8. Device fingerprint reused across 5 distinct customer accounts this week
-9. KYC document mismatch — selfie does not match ID photo, confidence 41%
-10. Escalated to Tier 2 — customer disputes the fraud hold, requesting a supervisor
-11. Account takeover suspected — password reset followed immediately by a wire request
-12. Sanctions screening partial match — 82% name similarity against OFAC list
-13. Refund-abuse pattern — 4th refund request in 30 days, same shipping address
-14. High-risk merchant category code (MCC 7995) flagged for manual review
-15. Synthetic-identity indicator — SSN issue date postdates the reported DOB
-16. Multiple failed 3DS challenges before a successful authorization
-17. Chargeback ratio breach — merchant at 1.2%, inside Visa's monitoring threshold
-18. Case closed — false positive, legitimate purchase confirmed via callback
-19. Case reopened — customer submitted additional supporting documentation
-20. Friendly-fraud suspected — tracking confirms delivery, dispute filed anyway
-21. Round-tripping detected — funds in and out within 4 hours, no net gain
-22. Email domain registered 2 days before the account signup date
-23. Shared card fingerprint across 3 already-flagged accounts this week
-24. Analyst note: hold released after a manual callback confirmed the cardholder
-25. Repeat offender — account previously banned under a different email, same device ID
-
-## Number-shape recipes
-
-```yaml number-shapes
-queue_intake:
-  arrival_process: poisson
-  base_rate_per_hour: 4.2
-  business_hours_multiplier: 2.6      # 9am-6pm local pulls volume up
-  weekday_vs_weekend_ratio: "3.5:1"
-  monthly_growth: "+2 to +5% MoM, flat in Dec"
-alert_severity_mix:
-  low: 0.58
-  medium: 0.29
-  high: 0.10
-  critical: 0.03
-  shape: "power-law-like, rank-ordered by alert_type frequency (velocity most common, sanctions rarest)"
-risk_score:
-  distribution: "right-skewed beta(alpha=2, beta=6)"
-  range: [0, 100]
-  mode: "12-18"
-false_positive_rate: "0.72-0.85 of flagged cases resolve as legitimate"
-time_to_resolution_hours:
-  distribution: "log-normal(mu=1.5, sigma=1.1)"
-  median: 5.8
-  p95: 74
-chargeback_amount_usd:
-  distribution: "log-normal(mu=4.1, sigma=0.9)"
-  median: 82
-  p99: 2400
-analyst_throughput:
-  shape: power-law
-  concentration: "top 20% of analysts close ~55% of total case volume"
-case_reopen_rate: "0.04-0.08"
-```
-
-## Believability patterns applied
-
-- **Coherence beats correctness per-field** — every case note pairs a concrete method, magnitude, and disposition (never a bare "flagged"); the analyst who closes it is a named row in `people.csv`, never a placeholder.
-- **Aggregates must reconcile** — the queue total, the severity mix, and the chargeback totals below are declared as invariants precisely so a rendered dashboard's headline number is always re-derivable from the rows beneath it.
-- **Timestamps carry seasonality** — `queue_intake` above is a Poisson arrival process gated by a business-hours multiplier and a weekday/weekend ratio, never a flat or uniform rate.
-- **Corporate email conventions are a single-source-of-truth choice** — every `people.csv` row uses `first.last@ledgerpoint.io` consistently; no row shortens to initials or a different pattern.
-
-```yaml invariants
-aggregates-reconcile: [queue_total = open_count + closed_count + escalated_count, sum(daily_alerts_by_severity) = daily_alerts_total, sum(chargeback_amounts) = chargeback_amount_total]
-legal-ranges: { risk_score: [0, 100, right-skewed-beta], time_to_resolution_hours: [0.1, 168, log-normal], chargeback_amount_usd: [5, 15000, log-normal] }
-forbidden-repetition: { max_same_surname_visible: 2, max_identical_case_note_text: 1, max_consecutive_cases_same_analyst: 3 }
-edge-rows: fixtures-only   # nulls/zeros/outliers live in state fixtures, never the primary set
-reference-instant: 2026-07-12T09:00:00Z   # fixed — all dates vary around this instant
-```

package/data/harnesses/code/skills/design/data/corpora/operations-review/people.csv

@@ -1,25 +0,0 @@
-name,email,role,seniority
-Amara Osei,amara.osei@ledgerpoint.io,Fraud Analyst,Mid
-Wei-Lin Chao,wei-lin.chao@ledgerpoint.io,Senior Fraud Analyst,Senior
-Diego Fuentes,diego.fuentes@ledgerpoint.io,Trust & Safety Lead,Lead
-Priya Raman,priya.raman@ledgerpoint.io,Risk Operations Manager,Manager
-Jonas Kowalski,jonas.kowalski@ledgerpoint.io,Fraud Analyst,Junior
-Fatima Al-Sayed,fatima.al-sayed@ledgerpoint.io,Compliance Officer,Senior
-Marcus Webb,marcus.webb@ledgerpoint.io,Fraud Analyst,Mid
-Ines Carvalho,ines.carvalho@ledgerpoint.io,QA & Escalations Specialist,Mid
-Tomasz Nowicki,tomasz.nowicki@ledgerpoint.io,Senior Risk Analyst,Senior
-Grace Adeyemi,grace.adeyemi@ledgerpoint.io,VP Trust & Safety,Executive
-Hiroshi Tanaka,hiroshi.tanaka@ledgerpoint.io,Fraud Analyst,Mid
-Camila Duarte,camila.duarte@ledgerpoint.io,Senior Fraud Analyst,Senior
-Nathan Reyes,nathan.reyes@ledgerpoint.io,Fraud Analyst,Junior
-Sanne Bakker,sanne.bakker@ledgerpoint.io,Risk Operations Analyst,Mid
-Oluwaseun Bello,oluwaseun.bello@ledgerpoint.io,Trust & Safety Analyst,Mid
-Anya Petrov,anya.petrov@ledgerpoint.io,Compliance Analyst,Junior
-Marco Belloni,marco.belloni@ledgerpoint.io,Senior Risk Analyst,Senior
-Yuki Sato,yuki.sato@ledgerpoint.io,Fraud Analyst,Mid
-Esteban Rios,esteban.rios@ledgerpoint.io,QA & Escalations Specialist,Senior
-Naledi Khumalo,naledi.khumalo@ledgerpoint.io,Compliance Officer,Manager
-Aleksander Wrobel,aleksander.wrobel@ledgerpoint.io,Risk Operations Manager,Manager
-Ingrid Solberg,ingrid.solberg@ledgerpoint.io,Senior Fraud Analyst,Senior
-Rohan Mehta,rohan.mehta@ledgerpoint.io,Fraud Analyst,Junior
-Bartosz Kaminski,bartosz.kaminski@ledgerpoint.io,Fraud Operations Lead,Lead

package/data/harnesses/code/skills/design/data/corpora/procurement-approvals/companies.csv

@@ -1,15 +0,0 @@
-name,industry,size_band
-Meridian Cloud Services,Cloud Infrastructure,Enterprise
-Ashgrove Office Supply,Office & Facilities,SMB
-Talbridge Consulting Group,Professional Services,Mid-Market
-Kestrel Security Systems,IT Security,SMB
-Vantree Staffing Solutions,Staffing & Contractors,Mid-Market
-Redwood Print & Signage,Marketing Services,SMB
-Corvid Analytics Software,Software/SaaS,Mid-Market
-Highmark Freight & Logistics,Logistics,Enterprise
-Juniper Legal Services,Professional Services,SMB
-Atlas Facilities Management,Office & Facilities,Mid-Market
-Brightline Travel Solutions,Travel,SMB
-Fenwick Hardware Supply,Office & Facilities,SMB
-Novacore Data Centers,Cloud Infrastructure,Enterprise
-Silverline Training Group,Professional Services,SMB

package/data/harnesses/code/skills/design/data/corpora/procurement-approvals/corpus.md

@@ -1,83 +0,0 @@
----
-id: procurement-approvals
-class: c1
-sourced: 2026-07-12
-source: "Distilled from purchase-order approval-workflow vocabulary (knack.com, ziphq.com, procurify.com, stampli.com) and SaaS spend-category taxonomy (finops.org, ardura.consulting); escalation mechanics from cflowapps.com. See docs/research/design-skill-phase1/c1-demo-data.md."
-license: open
----
-# Procurement approvals — demo corpus
-
-A purchase-requisition and approval-chain workflow. `people.csv` is the internal requesters, approvers, and finance staff; `companies.csv` is the vendors/suppliers a request or PO is issued against.
-
-## Activity vocabulary (25)
-
-1. Purchase requisition submitted — 25 Figma seats, annual plan
-2. Auto-approved — under the $500 department threshold
-3. Escalated to Finance — request exceeds the $10,000 single-purchase limit
-4. Approved by department manager, pending Finance sign-off
-5. Rejected — duplicate request, existing contract already covers the need
-6. Returned for revision — missing cost center code
-7. Vendor onboarding required before a PO can be issued
-8. Three-bid comparison attached — selected the mid-price vendor per policy
-9. Sole-source justification submitted for specialized equipment
-10. Budget check failed — exceeds the Q3 marketing allocation by $4,200
-11. Approved with conditions — 12-month term cap, no auto-renewal
-12. Contract routed to Legal for MSA redline review
-13. PO issued — Net 30 terms, delivery expected in 3 weeks
-14. Invoice mismatch — PO amount $8,400, invoice received for $9,150
-15. Emergency purchase approved retroactively — server outage remediation
-16. Renewal flagged 60 days out — auto-renewal clause, opt-out deadline approaching
-17. Approval delegated — primary approver on PTO, alternate assigned
-18. Request withdrawn by requester — need resolved internally
-19. Compliance hold — vendor missing a signed security questionnaire
-20. Second approval required — crosses the $25,000 dual-authorization threshold
-21. Spend category reclassified from Software to Professional Services
-22. Escalated to CFO — capital expenditure exceeds department authority
-23. Approved — annual software audit recommended vendor consolidation
-24. Change order submitted — scope increase adds $6,000 to the existing PO
-25. Reminder sent — approval pending 4 business days, SLA breach imminent
-
-## Number-shape recipes
-
-```yaml number-shapes
-request_amount_usd:
-  distribution: "log-normal(mu=6.8, sigma=1.4)"
-  median: 900
-  p90: 12000
-  p99: 145000
-approval_tiers:
-  - { max_amount: 500, approvers: 1, chain: "Manager", auto_approve_share: 0.9 }
-  - { max_amount: 10000, approvers: 2, chain: "Manager + Department Head" }
-  - { max_amount: 50000, approvers: 3, chain: "+ Finance" }
-  - { max_amount: null, approvers: 4, chain: "+ CFO/Executive" }
-approval_cycle_time_hours:
-  distribution: "log-normal(mu=2.1, sigma=1.3)"
-  median: 8.2
-  p95: 96
-rejection_rate: "0.05-0.12"
-escalation_rate: "0.08-0.15"
-spend_category_mix:
-  software_saas: 0.34
-  professional_services: 0.18
-  cloud_infrastructure: 0.15
-  hardware_equipment: 0.10
-  travel: 0.09
-  marketing_services: 0.08
-  office_facilities: 0.06
-vendor_count_growth: "+1 to +3 new vendors/month for SMB; +5 to +10/month for enterprise"
-renewal_auto_flag_window_days: 60
-```
-
-## Believability patterns applied
-
-- **Realism needs imperfection, deliberately placed** — `rejection_rate` and `escalation_rate` above are authored non-zero bands, so the primary set already carries the friction a real approval queue has; true edge cases (a null cost center, a $0 PO) still live only in fixtures per `edge-rows`.
-- **Relationships carry more realism than any single field** — `approval_tiers` above routes each request to a chain of roles (Manager → Department Head → Finance → CFO) that must resolve to real `people.csv` rows; a request approved by no one in the chain reads as broken, not just less real.
-- **Name diversity is deliberate, not incidental** — `people.csv` draws names across cultural backgrounds by hand rather than from a frequency-ranked census list, so no single surname or origin dominates the roster.
-
-```yaml invariants
-aggregates-reconcile: [requests_submitted = requests_approved + requests_rejected + requests_pending, po_total_amount = sum(line_items.amount), total_period_spend = sum(spend_by_category)]
-legal-ranges: { request_amount_usd: [25, 250000, log-normal], approval_cycle_time_hours: [0.25, 240, log-normal], approver_count: [1, 4, tiered-discrete] }
-forbidden-repetition: { max_same_surname_visible: 2, max_identical_requester_amount_same_day: 1, max_consecutive_pos_same_vendor: 5 }
-edge-rows: fixtures-only   # nulls/zeros/outliers live in state fixtures, never the primary set
-reference-instant: 2026-07-12T09:00:00Z   # fixed — all dates vary around this instant
-```

package/data/harnesses/code/skills/design/data/corpora/procurement-approvals/people.csv

@@ -1,25 +0,0 @@
-name,email,role,seniority
-Isabela Rocha,isabela.rocha@cordillera.com,Procurement Specialist,Junior
-Nathaniel Cross,nathaniel.cross@cordillera.com,Finance Manager,Manager
-Aditi Krishnan,aditi.krishnan@cordillera.com,Department Head - Marketing,Senior
-Samuel Okoro,samuel.okoro@cordillera.com,Controller,Senior
-Freya Halvorsen,freya.halvorsen@cordillera.com,AP Coordinator,Junior
-Benjamin Achterberg,benjamin.achterberg@cordillera.com,CFO,Executive
-Leila Haddad,leila.haddad@cordillera.com,IT Manager,Mid
-Connor Whitfield,connor.whitfield@cordillera.com,Facilities Coordinator,Junior
-Ngozi Eze,ngozi.eze@cordillera.com,Sourcing Manager,Mid
-Piotr Zielinski,piotr.zielinski@cordillera.com,VP Operations,Lead
-Mateus Lima,mateus.lima@cordillera.com,Procurement Specialist,Junior
-Elin Kristiansen,elin.kristiansen@cordillera.com,Senior Procurement Specialist,Senior
-Farrukh Aliyev,farrukh.aliyev@cordillera.com,AP Coordinator,Junior
-Priya Chandran,priya.chandran@cordillera.com,Department Head - Engineering,Senior
-Owen Delacroix,owen.delacroix@cordillera.com,Contracts Manager,Mid
-Simone Fischer,simone.fischer@cordillera.com,Category Manager - Software,Mid
-Dario Moretti,dario.moretti@cordillera.com,Procurement Manager,Manager
-Amina Toure,amina.toure@cordillera.com,Sourcing Manager,Mid
-Grant Fairweather,grant.fairweather@cordillera.com,IT Manager,Mid
-Renata Ibarra,renata.ibarra@cordillera.com,AP Manager,Manager
-Kofi Mensah,kofi.mensah@cordillera.com,Legal Counsel,Senior
-Wendy Lachance,wendy.lachance@cordillera.com,Department Head - Sales,Senior
-Rasheed Malik,rasheed.malik@cordillera.com,Controller,Senior
-Bianca Ferreira,bianca.ferreira@cordillera.com,Procurement Specialist,Junior

package/data/harnesses/code/skills/design/data/corpora/revenue-analytics/companies.csv

@@ -1,15 +0,0 @@
-name,industry,size_band
-Northwind Traders,Wholesale Distribution,Mid-Market
-Brackenridge Health Group,Healthcare,Enterprise
-Pinnacle Freight Systems,Logistics,Mid-Market
-Solstice Financial Partners,Financial Services,Enterprise
-Fielding & Cole Law,Legal Services,SMB
-Marrow Robotics,Manufacturing,Mid-Market
-Cobalt Learning Systems,EdTech,SMB
-Havencrest Insurance,Insurance,Enterprise
-Ashworth Biotech,Life Sciences,Mid-Market
-Cascade Retail Group,Retail,Enterprise
-Ternion Aerospace,Aerospace & Defense,Enterprise
-Lighthouse Media Partners,Media & Entertainment,SMB
-Granite Peak Construction,Construction,Mid-Market
-Vellum Publishing House,Publishing,SMB

package/data/harnesses/code/skills/design/data/corpora/revenue-analytics/corpus.md

@@ -1,89 +0,0 @@
----
-id: revenue-analytics
-class: c1
-sourced: 2026-07-12
-source: "Distilled from B2B SaaS deal-size and win-rate benchmarks (culta.ai, optif.ai, Velocity Partners' 200-company dataset) and CRM call-note / pipeline-stage vocabulary (Weflow- and HubSpot-class sales-call-report guidance); segment bands from Martal and GTMnow's SMB/mid-market/enterprise definitions. See docs/research/design-skill-phase1/c1-demo-data.md."
-license: open
----
-# Revenue analytics — demo corpus
-
-A CRM sales pipeline. `people.csv` is the internal sales/CS org (AE, SDR, CSM, RevOps — seniority is meaningful here); `companies.csv` is the CRM accounts: prospects and customers a deal is opened against.
-
-## Activity vocabulary (25)
-
-1. Intro call with Elena — pricing discussion, budget confirmed at $40K
-2. Discovery call: identified pain point around manual reconciliation
-3. Demo delivered to VP Eng and 2 stakeholders — positive reaction to SSO
-4. Sent proposal — 3-year term, 12% discount for annual prepay
-5. Champion (Marcus, Head of Ops) went quiet — no response in 9 days
-6. Negotiation: legal redlines returned, DPA clause under review
-7. Closed won — signed an 18-month contract, ARR $86,400
-8. Closed lost — chose the incumbent, price was the deciding factor
-9. Renewal call — flagged a 22% QoQ usage drop, account at risk
-10. Upsell opportunity: 15 additional seats, add-on module discussed
-11. Economic buyer identified — CFO joining the next call
-12. Security review kicked off — SOC 2 report requested
-13. Referral from an existing customer (Northwind Traders)
-14. Trial extended 14 days — low activation, nudged onboarding
-15. Competitive deal — up against Rippling, differentiating on support SLA
-16. Contract sent via DocuSign — awaiting counter-signature
-17. Budget confirmed for Q3, procurement process to follow
-18. No-show on the scheduled demo — rescheduled for next Tuesday
-19. Champion change — original contact left the company, re-engaging new VP
-20. QBR completed — NPS 8/10, expansion conversation opened
-21. Free trial converted to paid — 22 seats at $49/seat/mo
-22. Objection raised on implementation timeline — offered a dedicated CSM
-23. Multi-year deal restructured to annual due to a budget freeze
-24. Stakeholder map updated — 4 decision-makers, 2 influencers identified
-25. Deal slipped to next quarter — legal review taking longer than expected
-
-## Number-shape recipes
-
-```yaml number-shapes
-deal_size_usd:
-  distribution: "log-normal(mu=9.4, sigma=1.1)"
-  median: 12400
-  p90: 68000
-  p99: 310000
-win_rate_by_segment:
-  smb: "0.30-0.35"          # <$10K ACV band
-  mid_market: "0.20-0.28"   # $10-50K ACV band
-  enterprise: "0.12-0.18"   # >$50K ACV band
-sales_cycle_days:
-  smb: "log-normal, median ~34"
-  mid_market: "log-normal, median ~95"
-  enterprise: "log-normal, median ~210"
-open_pipeline_stage_mix:
-  prospecting: 0.35
-  discovery: 0.22
-  demo: 0.16
-  proposal: 0.12
-  negotiation: 0.08
-  closing: 0.07
-activity_volume_per_rep:
-  shape: power-law
-  concentration: "top quartile of reps generates ~60% of pipeline value"
-mrr_growth:
-  monthly_delta: "+3 to +9% for growth-stage accounts, -2 to +2% for mature accounts"
-net_revenue_retention: "0.95-1.15"
-logo_churn_monthly: "0.008-0.02"
-stakeholders_per_deal:
-  smb: "1-3"
-  mid_market: "3-7"
-  enterprise: "7-13"
-```
-
-## Believability patterns applied
-
-- **Distributions have a known shape, not uniform or normal** — deal size is modeled log-normal (many small deals, a long right tail of large ones) and rep activity is power-law (top quartile generates ~60% of pipeline value); drawing either as uniform-random or bell-curve reads as generated.
-- **Domain vocabulary beats generic placeholders** — the activity vocabulary above names actual pipeline stages, dispositions, and stakeholder roles ("Champion... went quiet," "Economic buyer identified") instead of generic "Task 1"-style phrasing.
-- **Never repeat identical values across records** — `forbidden-repetition` below caps identical open-deal amounts and same-day close dates so no two rows read as copy-paste.
-- **Measured amounts should pass a Benford-style sniff test** — `deal_size_usd`'s log-normal generation naturally skews leading digits toward 1–3, unlike a naive uniform-random amount generator.
-
-```yaml invariants
-aggregates-reconcile: [total_open_pipeline = sum(pipeline_value_by_stage), arr_net_change = arr_new + arr_expansion - arr_churn - arr_contraction, total_deals = count(closed_won) + count(closed_lost) + count(open_deals)]
-legal-ranges: { deal_size_usd: [500, 500000, log-normal], win_rate: [0.05, 0.45, beta-segment-dependent], sales_cycle_days: [7, 400, log-normal-segment-dependent] }
-forbidden-repetition: { max_same_surname_visible: 2, max_identical_deal_amount_open_deals: 2, max_deals_same_close_date: 4 }
-edge-rows: fixtures-only   # nulls/zeros/outliers live in state fixtures, never the primary set
-reference-instant: 2026-07-12T09:00:00Z   # fixed — all dates vary around this instant
-```

package/data/harnesses/code/skills/design/data/corpora/revenue-analytics/people.csv

@@ -1,25 +0,0 @@
-name,email,role,seniority
-Elena Vasquez,elena.vasquez@fernhollow.io,Account Executive,Mid
-Kwame Asante,kwame.asante@fernhollow.io,Sales Development Rep,Junior
-Sofia Marchetti,sofia.marchetti@fernhollow.io,Senior Account Executive,Senior
-Raj Patel,raj.patel@fernhollow.io,VP Sales,Executive
-Hannah Lindqvist,hannah.lindqvist@fernhollow.io,Customer Success Manager,Mid
-Yusuf Demir,yusuf.demir@fernhollow.io,Sales Engineer,Mid
-Chloe Bergstrom,chloe.bergstrom@fernhollow.io,RevOps Analyst,Junior
-Daniel Osei-Mensah,daniel.osei-mensah@fernhollow.io,Enterprise Account Executive,Senior
-Mei-Lin Tan,mei-lin.tan@fernhollow.io,Renewals Manager,Mid
-Oliver Bancroft,oliver.bancroft@fernhollow.io,Sales Manager,Lead
-Isabelle Fontaine,isabelle.fontaine@fernhollow.io,Account Executive,Mid
-Tariq Nasser,tariq.nasser@fernhollow.io,Sales Development Rep,Junior
-Brynn O'Callaghan,brynn.ocallaghan@fernhollow.io,Customer Success Manager,Mid
-Felix Andersson,felix.andersson@fernhollow.io,Sales Engineer,Mid
-Priyanka Deshmukh,priyanka.deshmukh@fernhollow.io,Senior Account Executive,Senior
-Marcus Lindgren,marcus.lindgren@fernhollow.io,RevOps Manager,Manager
-Aaliyah Johnson,aaliyah.johnson@fernhollow.io,Sales Development Rep,Junior
-Sebastian Kruger,sebastian.kruger@fernhollow.io,Enterprise Account Executive,Senior
-Noor Hassan,noor.hassan@fernhollow.io,Customer Success Manager,Mid
-Viktor Petrenko,viktor.petrenko@fernhollow.io,Deal Desk Analyst,Mid
-Camille Dubois,camille.dubois@fernhollow.io,Renewals Manager,Manager
-Adebayo Okonkwo,adebayo.okonkwo@fernhollow.io,Account Executive,Mid
-Ingrid Solheim,ingrid.solheim@fernhollow.io,Senior Customer Success Manager,Senior
-Julian Marsh,julian.marsh@fernhollow.io,Sales Director,Lead

package/data/harnesses/code/skills/design/data/elevation/ant-design-directional-shadow.md

@@ -1,19 +0,0 @@
----
-id: ant-design-directional-shadow
-class: a3
-sourced: 2026-07-12
-source: "distilled from ant.design/docs/spec/shadow"
-license: open
-feel: "enterprise-friendly, directionally aware; drawers and fixed headers cast shadow toward the content they overlay, not just downward"
----
-Ant Design is unusual in defining shadow **direction** as a first-class variable: the same three-layer blur/opacity recipe is mirrored down/up/left/right depending on where the component sits (a right-side drawer casts left, a bottom-fixed table header casts up). Three layers per direction blend from a tight, high-opacity contact shadow to a huge, near-invisible ambient one.
-
-```yaml values
-radius: { sm: 4, md: 6, lg: 8 }
-hairline: "none at rest; inputs use a separate 1px border token, not shared with the shadow system"
-shadow:
-  layer1-down: "0px 1px 2px -2px rgba(0,0,0,.16), 0px 3px 6px 0px rgba(0,0,0,.12), 0px 5px 12px 4px rgba(0,0,0,.09)"
-  layer2-down: "0px 3px 6px -4px rgba(0,0,0,.12), 0px 6px 16px 0px rgba(0,0,0,.08), 0px 9px 28px 8px rgba(0,0,0,.05)"
-  layer3-down: "0px 6px 16px -8px rgba(0,0,0,.08), 0px 9px 28px 0px rgba(0,0,0,.05), 0px 12px 48px 16px rgba(0,0,0,.03)"
-dark-ladder: { step: "+~5L per level (v5 dark algorithm inverts the neutral scale)", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/apple-liquid-glass-material.md

@@ -1,18 +0,0 @@
----
-id: apple-liquid-glass-material
-class: a3
-sourced: 2026-07-12
-source: "distilled from developer.apple.com/design/human-interface-guidelines/materials"
-license: reference-only
-feel: "spatial, translucent, light-reactive; depth comes from a refractive material simulating glass thickness, not a fixed shadow token"
----
-Apple's newest materials model (Liquid Glass) ties elevation to simulated material *thickness* rather than a discrete shadow scale: as a glass element grows or is interacted with, it "flexes" to a thicker material with deeper shadow and more pronounced lensing/refraction. Darker, denser materials deliberately mute shadow legibility, trading depth for reduced visual noise — the opposite trade every other system here makes.
-
-```yaml values
-radius: { sm: 10, md: 16, lg: 28 }
-hairline: "none as a drawn line — edge definition comes from the material's own refractive/specular edge"
-shadow:
-  base: "shadow deepens dynamically as the glass element scales up (material 'flexes' thicker)"
-  scroll: "shadow intensifies as content scrolls beneath it, to preserve separation"
-dark-ladder: { step: "n/a — darker/denser materials intentionally reduce shadow legibility rather than laddering lightness", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/atlassian-surface-shadow-pairing.md

@@ -1,20 +0,0 @@
----
-id: atlassian-surface-shadow-pairing
-class: a3
-sourced: 2026-07-12
-source: "distilled from atlassian.design/foundations/elevation and /foundations/radius; radius px approximated from usage guidance, exact figures not published in source"
-license: open
-feel: "structured enterprise calm; every raised surface ships as a matched surface+shadow pair, never a shadow in isolation"
----
-Atlassian defines four discrete elevations — sunken, default, raised, overlay — and enforces that `elevation.surface.*` and `elevation.shadow.*` tokens are always used together, never mixed across levels. In dark mode the rule inverts intuition: higher elevation surfaces get **lighter**, not darker, simulating front-lighting rather than cast shadow.
-
-Approximation note: radius px approximated from usage guidance — exact figures are not published in the source and should be checked before shipping.
-
-```yaml values
-radius: { sm: 3, md: 6, lg: 8 }
-hairline: "none at rest — sunken/default read via fill color, not a border line"
-shadow:
-  raised: "0px 1px 1px #091E4240, 0px 0px 1px #091E424F"
-  overlay: "0px 8px 12px #091E4226, 0px 0px 1px #091E424F"
-dark-ladder: { step: "+~4L per level — \"the higher the elevation, the lighter the surface looks\"", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/fluent-2-key-ambient-ramp.md

@@ -1,20 +0,0 @@
----
-id: fluent-2-key-ambient-ramp
-class: a3
-sourced: 2026-07-12
-source: "distilled from fluent2.microsoft.design/elevation"
-license: open
-feel: "Windows-native layered light; every shadow is a paired sharp key + soft ambient cast, doubling in blur up the ramp"
----
-Fluent 2 generates its whole shadow ramp from one equation: each step doubles the prior step's blur (shadow2 → 4 → 8 → 16 → 28 → 64), always as a key+ambient pair. Dark mode doesn't lighten surfaces — it simply raises the shadow's own opacity (~0.14 → ~0.28) to stay visible against a dark canvas.
-
-```yaml values
-radius: { sm: 2, md: 4, lg: 6 }
-hairline: "none by default — shadow2 (near-hairline blur) covers the lowest resting state"
-shadow:
-  shadow2: "0 0.5px 1px rgba(0,0,0,.14)"
-  shadow8: "0 4px 8px rgba(0,0,0,.14)"
-  shadow16: "0 8px 16px rgba(0,0,0,.14)"
-  shadow64: "0 0.5px 1px rgba(0,0,0,.24), 0 0 8px rgba(0,0,0,.20)"
-dark-ladder: { step: "n/a — dark mode doubles shadow alpha (~.14→.28) rather than lightening surfaces", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/github-primer-sparing-shadow.md

@@ -1,19 +0,0 @@
----
-id: github-primer-sparing-shadow
-class: a3
-sourced: 2026-07-12
-source: "distilled from primer.style/product/css-utilities/box-shadow and primer.style/foundations/css-utilities/borders"
-license: open
-feel: "developer-tool restraint; flat-and-bordered by default, shadow is the deliberate exception, never the everyday chrome"
----
-Primer's stated philosophy is to use shadow *sparingly*: small shadows for things that need to feel slightly elevated (pricing cards), medium for editorialized clickable content, and large reserved almost entirely for marketing pages and screenshots — never for ordinary product UI, which relies on `border-default` instead.
-
-```yaml values
-radius: { sm: 4, md: 6, lg: 8 }
-hairline: "border-default, 1px solid — the primary separation mechanic; shadow is the exception, not the rule"
-shadow:
-  small: "subtle lift — pricing cards, attention-getting UI"
-  medium: "diffused, larger — editorialized clickable content"
-  large: "very diffused, used sparingly — marketing content and screenshots only"
-dark-ladder: { step: "+~3L per level (canvas-default #0d1117 → canvas-subtle/overlay)", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/ibm-carbon-layer-and-border.md

@@ -1,19 +0,0 @@
----
-id: ibm-carbon-layer-and-border
-class: a3
-sourced: 2026-07-12
-source: "distilled from v11.carbondesignsystem.com/guidelines/color/implementation and Carbon's elevation guidance"
-license: open
-feel: "flat, rectilinear, enterprise-IBM; depth is a color-and-border swap between layers, almost never a cast shadow"
----
-Carbon's default posture is nearly shadow-free: each of its layer tokens (`layer-01/02/03`) is paired 1:1 with a matching border token, so nesting a panel inside a panel is a background-color-plus-border swap, not a shadow stack. Shadow is reserved almost exclusively for true overlays — dropdowns and modals. Corners stay square almost everywhere, in keeping with Carbon's rectilinear IBM heritage.
-
-```yaml values
-radius: { sm: 0, md: 0, lg: 2 }
-hairline: "$border-strong-0X paired with $layer-0X — the layer+border pair IS the elevation cue"
-shadow:
-  layer: "none — flat; separation via layer-01/02/03 background swap + paired border"
-  overlay: "0 2px 6px 0 rgba(0,0,0,.30)"
-  modal: "0 8px 16px 0 rgba(0,0,0,.30)"
-dark-ladder: { step: "n/a — discrete named themes (White/g10 light, g90/g100 dark), not a continuous ladder", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/linear-dark-surface-ladder.md

@@ -1,17 +0,0 @@
----
-id: linear-dark-surface-ladder
-class: a3
-sourced: 2026-07-12
-source: "distilled from github.com/voltagent/awesome-design-md — linear.app/DESIGN.md"
-license: reference-only
-feel: "software-craft dark mode; restraint scales — one accent, hairlines, and a shallow surface ladder carry all the hierarchy, no shadows anywhere"
----
-Linear runs dark-only and resists drop shadows almost entirely; elevation is a **four-step surface-lightness ladder** (surface-1 through surface-4) paired with 1px hairlines of increasing strength (default → strong → tertiary) for nested content. The steps are deliberately small — a whisper of lift, never a spotlight.
-
-```yaml values
-radius: { sm: 6, md: 8, lg: 12 }
-hairline: "1px solid {colors.hairline} #23252a at rest; {hairline-strong} #34343a on focus; {hairline-tertiary} #3e3e44 for nested surfaces"
-shadow:
-  all-levels: "none — drop shadows resisted almost entirely on dark surfaces"
-dark-ladder: { step: "+~2L per level (surface-1 #0f1011 → surface-2 #141516 → surface-3 #18191a → surface-4 #191a1b)", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/material-2-key-ambient-dp-ramp.md

@@ -1,19 +0,0 @@
----
-id: material-2-key-ambient-dp-ramp
-class: a3
-sourced: 2026-07-12
-source: "distilled from m2.material.io/design/environment/elevation.html and m1.material.io/resources/shadows.html"
-license: open
-feel: "classic Google-Material paper-stack; every raise reads as a discrete, physical dp jump"
----
-The original Material model: a single dp scale (0–24) drives a fixed **key shadow + ambient shadow** pair at every level — sharp directional shadow layered under a soft diffuse one, both pure black at low opacity. No brand tint, no tonal color; the whole hierarchy is shadow-only, which is what later gave way to M3's tonal approach.
-
-```yaml values
-radius: { sm: 2, md: 4, lg: 8 }
-hairline: none
-shadow:
-  1dp: "0 1px 1px 0 rgba(0,0,0,.14), 0 2px 1px -1px rgba(0,0,0,.12), 0 1px 3px 0 rgba(0,0,0,.20)"
-  8dp: "0 8px 10px 1px rgba(0,0,0,.14), 0 3px 14px 2px rgba(0,0,0,.12), 0 5px 5px -3px rgba(0,0,0,.20)"
-  16dp: "0 16px 24px 2px rgba(0,0,0,.14), 0 6px 30px 5px rgba(0,0,0,.12), 0 8px 10px -5px rgba(0,0,0,.20)"
-dark-ladder: { step: "white-overlay 5%→16% across 1dp→24dp (~+1-2% per step)", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/material-3-tonal-elevation.md

@@ -1,21 +0,0 @@
----
-id: material-3-tonal-elevation
-class: a3
-sourced: 2026-07-12
-source: "distilled from m3.material.io/styles/elevation/tokens; shadow figures approximate Material Web's published elevation-level tokens"
-license: open
-feel: "quiet tech-forward calm; depth reads as a warm color shift more than a cast shadow, corners softly rounded"
----
-Material 3's headline move: elevation is expressed primarily through **tonal surface tint** (a translucent overlay of the primary brand color, stronger at higher levels) with a thin shadow underneath as a secondary cue. No borders — surfaces separate by tone, not line. Best where a single brand hue should visibly "light up" whatever floats highest.
-
-Approximation note: shadow figures approximate Material Web's published elevation-level tokens — check exact rgba/opacity figures against the live source before shipping.
-
-```yaml values
-radius: { sm: 8, md: 12, lg: 16 }
-hairline: none
-shadow:
-  level1: "0px 1px 2px rgba(0,0,0,0.30), 0px 1px 3px 1px rgba(0,0,0,0.15) + {primary-hue tint / ~5% overlay}"
-  level2: "0px 1px 2px rgba(0,0,0,0.30), 0px 2px 6px 2px rgba(0,0,0,0.15) + {primary-hue tint / ~8% overlay}"
-  level3: "0px 4px 8px 3px rgba(0,0,0,0.15), 0px 1px 3px rgba(0,0,0,0.30) + {primary-hue tint / ~11% overlay}"
-dark-ladder: { step: "+~3% primary-tint per level (approx 5/8/11/12/14%)", hue: drift }
-```

package/data/harnesses/code/skills/design/data/elevation/neobrutalist-hard-offset.md

@@ -1,19 +0,0 @@
----
-id: neobrutalist-hard-offset
-class: a3
-sourced: 2026-07-12
-source: "distilled from neubrutalism.com and community neobrutalism CSS references (community pattern, no single owning org)"
-license: open
-feel: "loud, tactile, intentionally raw; a solid offset shadow with zero blur reads as a physical card lifted straight off the page"
----
-Neobrutalism inverts the entire elevation vocabulary: no blur, ever. A thick (3–4px) solid black border plus a flat, hard-edged offset shadow in pure black (or a bold flat color) does all the depth work — the shadow is a solid silhouette displaced down-right, not a soft cast. Radius stays at or near zero to keep the "raw" read.
-
-```yaml values
-radius: { sm: 0, md: 0, lg: 4 }
-hairline: "not a hairline — a 3-4px solid black border replaces the border/shadow distinction entirely"
-shadow:
-  sm: "4px 4px 0px #000000"
-  md: "6px 6px 0px #000000"
-  lg: "8px 8px 0px #000000"
-dark-ladder: { step: "n/a — most kits binary-invert black/white rather than laddering surfaces", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/neumorphic-soft-dual-shadow.md

@@ -1,18 +0,0 @@
----
-id: neumorphic-soft-dual-shadow
-class: a3
-sourced: 2026-07-12
-source: "distilled from neumorphism.io and community soft-UI CSS references (community pattern, no single owning org)"
-license: open
-feel: "soft, extruded, monochrome; elements look pressed into or popped out of a background that shares their exact color"
----
-Neumorphism fakes physical extrusion with a **paired light+dark shadow** on an element whose background exactly matches its parent's — no border, no color contrast, only two offset shadows (one lighter, one darker than the shared base) doing all the modeling. Pressed/inset states simply invert the pair. Contrast-fragile by construction: raising shadow strength enough to pass accessibility checks destroys the soft illusion that is the entire point.
-
-```yaml values
-radius: { sm: 12, md: 16, lg: 24 }
-hairline: "none — a drawn border would break the illusion; edges exist only via the shadow pair"
-shadow:
-  raised: "10px 10px 20px #bebebe, -10px -10px 20px #ffffff"
-  pressed: "inset 6px 6px 12px #bebebe, inset -6px -6px 12px #ffffff"
-dark-ladder: { step: "n/a — dark variant is a manual per-surface retune (e.g. base #2d3239, shadows #23272d / #373d46), not a formula", hue: held }
-```

package/data/harnesses/code/skills/design/data/elevation/notion-warm-minimal-hairline.md

@@ -1,17 +0,0 @@
----
-id: notion-warm-minimal-hairline
-class: a3
-sourced: 2026-07-12
-source: "distilled from github.com/VoltAgent/awesome-design-md — notion/DESIGN.md"
-license: reference-only
-feel: "warm, paper-like minimalism; content over chrome, separation is almost always a thin line rather than a cast shadow"
----
-Notion's house rule is explicit: don't apply heavy shadows on flat documentation cards. Separation is carried almost entirely by a single hairline border token reused across nav, cards, and containers; the rare exception is a very soft lift on preview/mockup cards, kept subtle enough not to compete with content.
-
-```yaml values
-radius: { sm: 4, md: 6, lg: 10 }
-hairline: "1px solid {colors.hairline} #EDEEEC — the default separator for nav, cards, and containers; heavy shadows explicitly avoided"
-shadow:
-  rare-lift: "used sparingly, only on mockup/preview cards — e.g. 0 4px 16px rgba(0,0,0,.06)"
-dark-ladder: { step: "+~2-3L per level (approx #191919 → #202020 → #2f2f2f surfaces)", hue: held }
-```