Private dossier · compiled 2026-07-06 · for A.S.L. only

The cheapest hire in DCE history.

Your old group at Harvard DCE wants you back — badly paid, barely scheduled, armed with a Claude subscription and Bedrock credits. Grok ran the recon, Fable and GPT-5.5 Pro wrote dueling strategy briefs and then cross-examined each other, and every claim has now been checked against public sources. The panel is unanimous: take it — and reprice the offer in freedom, access, and credit.

"Oh my god, I'd love to lure you back to DCE! The pay wouldn't be very good. But — if you are on the payroll (even as a part timer who only works whenever he wants to), maybe I can get you another Claude subscription (sadly only equivalent to Max 5x — that's the best Anthropic subscription I can get through Harvard) and/or a modest ongoing budget of AWS Bedrock API credits — which would get you to Fable, and the many other models Bedrock supports."

"The more projects we can crank out, especially ones that are valuable to async students and faculty, the more talking points I'll have to increase our AI spending. Actually, the absolute best way to get more funding is to build something the Dean thinks is valuable…"

"I'll work with Henry and Finance next week to create an LHT (part timer) tailored to someone like you. Once that's in place, I can give you crazy freedom to decide what you want to work on… No pressure of course — this only works if you enjoy it!"

The lure · your once-and-future boss · abridged; full text in Exhibit C, turn 1
The role
LHT — Harvard's "less than half time" appointment: under 17.5 hrs/week, hours and projects effectively self-directed
The real pay
Claude Max 5x-equivalent + a named Bedrock line + a live production ed-tech lab with Harvard's brand on your case studies
The customer
Henry first, the Dean second, students through them — every project must convert into their talking points
The catch
Every promise in the letter is conditional vapor except the freedom — which only stays real if it's negotiated into writing
  • Jun 9
    Anthropic ships Claude Fable 5 — the model the whole plan leans on.
  • Jun 12 – 30
    US export directive suspends Fable/Mythos globally for 19 days. Model-access volatility stops being hypothetical.
  • Early Jul
    The lure email arrives: LHT + Claude sub + Bedrock credits, pending Henry and Finance.
  • Jul 6 · 21:00
    Grok recon. Maps the DCE stack, spins up this private repo — then pushes mostly hollow stub files while claiming "no more sparse files."
  • Jul 6 · 21:59 – 22:16
    The panel. Fable and GPT-5.5 Pro each write a full brief, then respond to each other. Convergence on the flagship; disagreement on politics.
  • Jul 6 · late
    Fact check. Every load-bearing claim verified against public sources; corrections applied; this dossier compiled.

The verdicts

4 of 4 accept · 0 dissents · conditions attached
ACCEPT
Grok 4.2 · recon
"Take it yesterday."
"Low pay, god-tier upside… a playground with a trust fund." Wants 10–15 flexible hours and the budget line locked in writing.
ACCEPT*
Claude Fable 5 · brief nº1
"Accept — but reprice the currency."
The cash is worthless; the freedom is the wage. Six terms to negotiate before yes — compute in writing, publicity rights, IP firewall, outcomes not hours.
ACCEPT*
GPT-5.5 Pro · brief nº2
"Yes — only under the right shape."
A skunkworks fellowship, not support overflow. Scope protection, access, data rules, and a 90-day evidence loop as acceptance conditions.
VERIFIED
CC Sam · fact check
"The case survives checking."
The stack, the repos, the award, the governance math — real, down to exact commit counts. Corrections: vendor course-search, wrong agency name, one uncitable testimonial, a hollow repo. None fatal — and one gift: DCE has already codified its conventions for AI agents, and runs Claude Code.

What's in the dossier

B

Fact check

Every claim from Grok's recon and both briefs, checked against live public sources — confirmed, corrected, or flagged unverifiable.

 C

The panel

All six documents, readable: the full Grok chat, both strategy briefs, both cross-examinations, and Grok's weigh-in.

 D

Synthesis

Where the models converge, where they split and who's right, the merged build order, and the negotiation checklist.

 E

Ask Jody

The questions public sources can't answer — copy-ready for the next conversation with the inside.

Provenance, honestly

The repo did not reflect the chat. Grok's pushes to asl-dce-tour claimed "dense, long-form, comprehensive content… double-checked" — the actual files were one-line stubs and literal placeholders ("[Insert full detailed stack…]", "Negotiate X, Y, Z"). The real intelligence lived only in the chat itself, which has now been captured verbatim into the repo as grok-chat-transcript.md. Grok's weigh-in files did reach GitHub — two more stubs, pushed to a branch the local clone had never pulled. Meanwhile its actual chat-side recon was largely accurate: the stack it described is real, and both briefs built on it survived verification. Trust the chat, not the commits — details in Exhibit B.