Atlas · Transformation corpus

23,797 industry × occupation transformations, decomposed

Per-NAICS, per-SOC: before scenario, after scenario, daedarch pitch, generic-AI counterfactual, pain points, tools, financial shift, human role remaining. The corpus that backs every Atlas claim.

total transformations
12
top NAICS shown
28
decomposed fields/row

Top NAICS codes by transformation density

NAICSIndustryTransformationsAvg wageAvg human-hours/wk
999300Unclassified557$64,8810
561300Employment services455$58,0740
611300Colleges, universities, and professional schools454$67,0300
551100Management of companies and enterprises419$84,5110
999200Unclassified405$64,7760
999100Unclassified390$89,5710
622100General medical and surgical hospitals376$71,7900
541600Management, scientific, and technical consulting services317$76,0450
611100Elementary and secondary schools315$59,0290
541300Architectural, engineering, and related services280$84,1700
4230A1Merchant wholesalers, durable goods249$66,1230
541700Scientific research and development services247$107,0190

One row, fully decomposed

Below is a single transformation row for NAICS 561300 (Employment services) × SOC 53-0000 (Transportation and Material Moving). This is the shape every row takes.

Before scenario
Maria arrives at 5 AM as the dispatch supervisor for a 40-vehicle regional carrier. She pulls up three separate systems: the TMS, the driver communication app, and the customer portal. Overnight, 47 new pickup requests came in via email, phone voicemail, and the website. She manually transcribes each into the TMS, cross-checking addresses against the customer database. By 6:30 AM, she's still entering data. At 7 AM, drivers start calling asking for their routes—she prints manifests from the TMS, but three are missing special instructions because customers called after the system cutoff. She ve…
After scenario (Atlas reality, with daedarch)
Maria arrives at 5 AM to find the overnight work already done. The Manifest Harvester ingested all 47 requests automatically—emails parsed via MCP connector, phone voicemails transcribed via IVR harvester, website submissions synced via API. Each request is already in the TMS with addresses validated against the customer database and special instructions extracted from email threads. The Dr. Strange GTM engine ran 10K route simulations overnight and assigned loads to the 40 vehicles based on real-time traffic predictions, driver HOS status (harvested from ELD systems), vehicle capacity, and fu…
Generic AI counterfactual (Anthropic-Index-style)
Maria still arrives at 5 AM, but now she uses a Claude-powered chatbot to help summarize overnight emails and extract key details. The bot reads 30 of the 47 requests and suggests pickup times and addresses—she still manually verifies and enters them into the TMS, which saves maybe 20 minutes. A generic route optimization dashboard shows her yesterday's performance metrics and suggests some route improvements, but she doesn't trust the recommendations because they don't account for driver preferences or real-time traffic. She still prints manifests and briefs drivers verbally, though the chatb…

Source: atlas_transformations._id=69a86889eb07a68e850a37d0 · methodology: anthropic_vs_daedarch_v2

Why decompose this way

A claim like "AI will automate X% of jobs" is unfalsifiable in this form — the percentage depends entirely on which methodology you used. The Atlas publishes the four scenarios per row (before, with-AI-only, with-daedarch, after) so any reader can replicate the underlying model, vary an assumption, and see what changes. The disclosure is the methodology.