A first sourcing pass across all seven strategy decks

Why Care?

A pitch deck full of impressive-looking stats is only as good as the stats themselves. Reach University's seven strategy decks — Literacy & Numeracy, Frontier Job Demand, Agent Workflow Maxxing, Rural Income, the Agentic Product Lab, Reverse Engineer Available Funding, and Workforce Development — were built fast, with plenty of numbers dropped in as directionally-right placeholders. That's normal for a first draft. It's not normal for something that goes in front of a funder.

We'd just shipped a hover-citation system (the small [n] superscript marker you can hover or tab to for the source), so the moment was right to actually use it: go stat by stat, deck by deck, and either find the real source or admit we couldn't.

What's New?

• Workforce Development is live as the seventh Strategy deck (/scroll/workforce-development/wd-v1), with its own one-pager, joining Rural Income, Literacy & Numeracy, the Agentic Product Lab, Agent Workflow Maxxing, Reverse Engineer Available Funding, and Frontier Job Demand on /strategies. The through-line: the U.S. workforce system was built for employer demand, not worker mobility, and Brookings/Child Trends research names the active ingredient the Apprenticeship Degree implements — relationship-based work-based learning that predicts job quality a decade later.

Every numeric claim across all seven decks — roughly 30 slide files — now either carries a verified citation or an honest flag. Concretely:

• New citations added wherever a stat checked out against a primary source: OECD/PIAAC wage-premium data, NCES literacy proficiency levels, BLS wage benchmarks, DOL/GAO apprenticeship reporting, Brookings/Child Trends longitudinal findings, Giving USA and Altrata philanthropy data, and more — each one fetched and read, not just assumed from a headline.

• Wrong numbers corrected in place, not just re-cited. A few examples:

  • Frontier Job Demand claimed ~650K U.S. drone-industry jobs, sourced to PwC — except that PwC figure is a UK economic-impact study, not a U.S. one. Swapped for a real FAA U.S. remote-pilot growth forecast.

  • Workforce Development said "25% of tasks displaced by 2030," citing the WEF Future of Jobs report — the report actually says 47%. Fixed.

  • Reverse Engineer Available Funding was still running last year's UHNW population figure (426K); the current Altrata 2025 report says 511K.

  • Two Workforce Development slides linked to a DOL PDF that 404s. Both now point at the real document.

• Filler softened or flagged where no real source exists — a district quote in Rural Income and two funder "prospects" in Literacy & Numeracy couldn't be substantiated, so they're now labeled honestly ((unverified), prospect · no grant history confirmed) instead of dressed up as fact.

• Agentic Product Lab came back clean — every stat in that deck already matched its source once checked. Good to know the placeholder-stats problem wasn't universal.

How it Works

We ran seven agents in parallel, one per strategy, each scoped to its own slide directory so there was no risk of them stepping on each other's files. Each agent had three inputs: the deck's own narrative doc in context-v/narratives/strategies/<strategy>/README.md (which already flagged most figures ⚠ VERIFY and listed candidate sources from the original research pass), the client's research corpus in ai-labs/augment-it/clients/reach-edu/corpus/, and open web access to fetch and read the actual source page — not just trust a search snippet or a citation's title.

The rule given to every agent was the same: never invent a URL. If a number couldn't be confirmed against something actually fetched and read, it got softened, flagged with a SOURCING TODO comment, or swapped for a verified substitute — never silently left as an uncited "fact," and never given a fake-looking citation just to make the slide look sourced.

What's Next

This is a first pass, not a final audit — a few line items are explicitly flagged for follow-up (the unconfirmed Rural Income quote, the two Literacy & Numeracy funder prospects, the assessment-instrument choice still TBD). Worth a second pass before any of these decks goes in front of a funder for real, and worth checking whether the corrected figures change any of the surrounding narrative copy in the context-v/narratives/strategies/ READMEs.