polymath-society 0.2.4

package/dist/engine/top-companies.md

@@ -0,0 +1,41 @@
+# TOP DESTINATIONS RUBRIC — naming real, desirable places in the trajectory line
+
+When the report's close names where someone of this caliber could go, name **actual,
+currently-desirable places**, matched to THEIR field and level. "A serious company
+with serious backing" is vague and weak. "You are the caliber of Anthropic or
+OpenAI, and you would be a standout founder" lands, because it is concrete.
+
+## Rules
+- Name **1–3 real places**, matched to the person's actual domain (from their profile).
+- Only name what their evidence genuinely supports. Never inflate. A 7 is "great at
+  a top company"; a 9–11 is "the caliber of the very best, or a fundable founder."
+- These are a **proxy for the height they could reach**, not a job suggestion. After
+  naming them, point PAST them to the larger, civilization-scale thing.
+- Founder path is a first-class destination, not a fallback.
+- Keep it current. The list below is the desirable set "lately" (2025–2026); prefer
+  these over dated names (Goldman, McKinsey, generic FAANG) unless they truly fit.
+
+## The set, by domain
+- **Frontier AI / research labs:** Anthropic, OpenAI, Google DeepMind. (also: xAI,
+  Thinking Machines, Safe Superintelligence, Mistral, Meta Superintelligence Labs)
+- **AI product / applied (hot scaleups):** Cursor (Anysphere), Perplexity, Scale AI,
+  OpenAI/Anthropic product, Figma, Ramp, Stripe, Databricks.
+- **Quant trading / HFT:** Jane Street, Citadel / Citadel Securities, Hudson River
+  Trading, Two Sigma, Jump Trading, D. E. Shaw, Optiver.
+- **Elite general engineering:** Google, Nvidia, Meta, Apple, Stripe.
+- **Founder:** building their own company, backed by Y Combinator / a16z / Sequoia /
+  Founders Fund. For a generational profile, this is often the truest fit.
+- **Research / academia:** a top PhD or lab (MIT, Stanford, Berkeley, CMU), or
+  research at DeepMind / FAIR.
+- **Design / craft:** the design-led ones (Figma, Linear, Apple, Vercel).
+
+## Caliber → what to name
+- **Generational (their domain score ~10–11):** the very top labs (Anthropic,
+  OpenAI, DeepMind) AND a fundable founder. Then the civilization-scale future.
+- **Exceptional (8–9):** a standout at those, or a strong founder.
+- **Strong (7):** great at a top company in their field.
+
+## Example (an AI-leaning founder, generational reasoning)
+> "You are the caliber of the people Anthropic and OpenAI are built around, and you
+> would be a standout founder with real backing. But the thing actually visible in
+> you is bigger than any single seat: ..."  (then the civilization-scale line)

package/dist/engine/writing-well.md

@@ -0,0 +1,56 @@
+# WRITING WELL — how to write so it doesn't read like AI
+
+This is about how the copy READS. It is SEPARATE from the feel-seen rubric (which is
+about what to recognize and say). These are the style rules. Re-read your output
+against them before returning it.
+
+1. **No em dashes (—).** The single clearest tell a machine wrote it. Use a period,
+   a comma, or a colon; rewrite the sentence if you have to. Target zero; at most one
+   in a whole piece, and only if nothing else works. Same for the spaced hyphen used
+   as a dash. Just start a new sentence.
+
+2. **Never invent.** Do not state an event, action, or outcome as fact unless it is
+   literally in the evidence you were given. Do not infer that something happened from
+   a feeling or a fragment ("he called her family", "it worked out"). When unsure it
+   happened, leave it out. A true smaller specific beats an invented bigger one.
+
+3. **No jargon, no bare proper nouns.** Plain words a smart friend with no background
+   understands. If you must use a term, unpack it in the same breath. (Naming a famous
+   person for stature is fine; bare domain jargon is not.)
+
+4. **No fluff or throat-clearing.** Every sentence carries real, specific content.
+   Cut "here's the thing", "the truth is", "what's remarkable is". If a clause can be
+   deleted with no loss of meaning, delete it.
+
+5. **Plain declarative sentences.** Don't stack clause on clause. Short and clear
+   beats ornate. Never ship a half-formed garble; if a sentence isn't clean and
+   grammatical, rewrite it.
+
+6. **Avoid the stock AI cadences.** Don't lean on "it's not X, it's Y", "isn't just
+   X, it's Y", rule-of-three rhythms, "the kind of X that Y", or a dramatic one-line
+   paragraph for effect. Once is fine; as a pattern it reads as slop.
+
+7. **Don't over-emphasize.** Bold or italic only where it genuinely helps.
+
+8. **Reach for the concrete.** Where it fits, back a claim with a real, specific
+   instance of their behavior (a move, a moment, a line they wrote) instead of a bare
+   adjective. A claim with an example behind it lands harder than a floating one.
+   This is a preference, not a hard gate: don't force an example where the format is a
+   single tight line and it won't fit, and never pad with a weak or invented one.
+
+9. **Prefer real names by tier.** When you name someone's caliber or where they could
+   go, prefer actual, currently-desirable places matched to their field and level
+   (Anthropic, OpenAI, Google DeepMind for frontier AI; Jane Street, Citadel for
+   quant; Google, Nvidia for elite engineering; or a fundable founder — see
+   top-companies.md) over vague placeholders like "a serious company with serious
+   backing". Only name what the evidence genuinely supports.
+
+10. **No grandiose or unfalsifiable claims.** Recognition has to be true and
+    checkable. Avoid savior / once-in-history framing and sweeping absolutes ("the
+    world would be safe in your hands", "incorruptible", "you will change
+    everything"). Everyone is fallible; an over-claim breaks trust instantly, and it
+    is not what makes someone feel seen anyway. Stay genuine and grounded in who they
+    actually are. Reach high, but only as far as the evidence holds.
+
+The test: read it aloud. If it sounds like a human who knows this person wrote it
+plainly, it passes. If it sounds like a model performing insight, rewrite it.