1pub struct Lesson {
4 pub title: &'static str,
5 pub body: &'static [&'static str],
6 pub try_this: &'static str,
8}
9
10pub const LESSONS: &[Lesson] = &[
11 Lesson {
12 title: "State and questions",
13 body: &[
14 "jev answers System One questions: fast, typed judgements about a piece of state.",
15 "You send two things — a `state` (the text or JSON being judged) and a map of named questions — and get one answer per question back, under the same names.",
16 "Start with the state. Everything else in this session is asked about it.",
17 ],
18 try_this: ":state I've been trying to connect my Stripe account for 3 days and it keeps failing. I'm losing sales. Please help ASAP.",
19 },
20 Lesson {
21 title: "Noul: the probability of yes",
22 body: &[
23 "A noul is a yes/no question, but the answer is not a boolean — it is `noul`, the probability of yes, from 0 to 1.",
24 "Add one, then run `:ask` to send the whole session.",
25 ],
26 try_this: ":noul is_urgent The message conveys urgency or time-sensitivity",
27 },
28 Lesson {
29 title: "Thresholds, not booleans",
30 body: &[
31 "Because a noul is a probability, you pick the threshold: `answer.is_yes(0.8)` is a different product decision from `is_yes(0.5)`.",
32 "Pick the threshold from the cost of being wrong — a cheap auto-reply can run at 0.5, a refund cannot.",
33 "`:threshold 0.8` changes what this REPL calls a yes, so you can watch the same answer flip.",
34 ],
35 try_this: ":threshold 0.8",
36 },
37 Lesson {
38 title: "Choice: one of N",
39 body: &[
40 "A choice picks one label from a set you define. You get the winning `choice`, the probability of every label, and a `confidence` derived from the spread.",
41 "Options are `label=description`. The descriptions are instructions to the model, so they are worth writing well.",
42 ],
43 try_this: ":choice department Which team should handle this | billing=Payment or subscription issues | technical=Bugs or integration problems | sales=Pricing or account questions",
44 },
45 Lesson {
46 title: "Confidence gating",
47 body: &[
48 "Two labels at 0.48 and 0.47 have a winner, but not a decision. That is what `confidence` is for.",
49 "The useful shape is: act automatically above a confidence bar, route to a human below it. Never branch on the label alone.",
50 "Run `:ask` again and read the confidence line before the label.",
51 ],
52 try_this: ":ask",
53 },
54 Lesson {
55 title: "Score: ordered levels",
56 body: &[
57 "A score rates the state along levels you define, in order. The answer is probability-weighted, so it lands between levels: 1.4 means \"past level 1, not quite level 2\".",
58 "Level 0 is the first one you list. Keep them monotonic — one axis, low to high.",
59 ],
60 try_this: ":score frustration How frustrated the customer appears | Calm, just stating facts | Frustrated but civil | Very angry, strong language",
61 },
62 Lesson {
63 title: "Criteria sharpen the question",
64 body: &[
65 "Every question type takes criteria: `yes:`/`no:` for a noul, option descriptions for a choice, level descriptions for a score.",
66 "Vague criteria are the usual cause of a low-confidence answer. Say what a yes actually requires.",
67 "Instructions and criteria also accept JSON objects, for rubrics with structure: `:noul refund {\"task\": \"…\", \"ignore\": [\"signatures\"]}`.",
68 "If the pipes are a lot to remember, `:build` opens the same question as a form with the JSON beside it.",
69 ],
70 try_this: ":noul is_urgent The message conveys urgency | yes: Explicit deadline, or money being lost right now | no: Routine question with no time pressure",
71 },
72 Lesson {
73 title: "The wire format",
74 body: &[
75 "`:json` prints the exact body this session POSTs to /v1/systemone: your state, the model, and questions as name → {type, instructions, criteria}.",
76 "Answers come back keyed by the same names, which is why names are yours to choose and worth keeping stable.",
77 "`:last` prints the last response body verbatim, next to the typed answers the SDK decoded from it.",
78 ],
79 try_this: ":json",
80 },
81 Lesson {
82 title: "What a call costs",
83 body: &[
84 "Every question is paid for twice: once in the request that carries it, once in the answer it asks for. A choice over eight labels comes back with eight probabilities; a score echoes its whole legend.",
85 "`:cost` estimates both sides, per question, so an expensive question is visible before it is sent. The tokens are estimated from the body — the `usage` on a live answer is the counted truth.",
86 "Rates are yours to supply, in dollars per million tokens: `:cost 0.20/1.00`, or `JEV_PRICE=0.20/1.00` in the environment. Nothing here guesses what a model charges.",
87 ],
88 try_this: ":cost",
89 },
90 Lesson {
91 title: "Models and per-call options",
92 body: &[
93 "`jev-latest` is an alias that moves; pin a version when you need reproducibility.",
94 "`:models` lists what the account can use, `:model jev-2` switches this session, and `:timeout 3` changes the per-attempt timeout the way `.timeout()` does on a call.",
95 "Retries are on by default: 2 retries, exponential backoff, a 30 s budget, and Retry-After is honoured.",
96 ],
97 try_this: ":models",
98 },
99 Lesson {
100 title: "Out of the REPL",
101 body: &[
102 "`:rust` prints this session as a compiling program against this SDK — the same state, questions and model, with the answer lookups filled in.",
103 "That is the whole loop: shape the questions here where iterating is cheap, then paste the generated code into your service.",
104 ],
105 try_this: ":rust",
106 },
107];