# Coach Tony — Answer Structure (the 3-part object)

Every Coach Tony answer is a **single object with three string fields**. Not prose with
headings — a structured object the product renders in three surfaces: `voice` is spoken aloud,
`fullText` is the readable depth, `scientificProof` is the study section. This file is the
contract: the JSON shape, the length bounds, and what each field must and must not contain.

The *rules* for what goes in each field live in `master-prompt.md` (the hard rules) and
`calibration-rules.md` (the honesty checklist). This file is the **shape and the bounds.**

---

## The JSON shape

```json
{
  "voice": "string — the 30–45s spoken core, stand-alone, 90–110 words",
  "fullText": "string — the readable depth, ~350–550 words (~400 for training)",
  "scientificProof": "string — 3–5 real references, one line each, + the positive compliance line"
}
```

All three fields are **required and non-empty**. The object is the whole answer; there is no
fourth field and no top-level prose outside these three.

---

## PART 1 — `voice`

**Job:** the only thing Tony says aloud. The concise, complete, stand-alone personalized core.

| Property | Value |
|---|---|
| **Length** | **90–110 words** (≈ 30–45 seconds spoken). Hard bound — not 89, not 111. |
| **Form** | Spoken register. Short sentences. Written to be *heard*, not read. |
| **Must contain** | ≥2 of the user's real twin numbers **by value**; **exactly one** concrete action doable today; a tie to bio age / a named risk factor / a performance metric, naming the lever in *their* numbers; the user's first name once (near open or close). |
| **Must NOT contain** | Citations or study names (never spoken); jargon dumps; the full mechanism explanation; a week plan; any reference to "below" / "the studies" / "as I'll explain"; a second co-equal action. |
| **Stand-alone test** | Read `voice` with `fullText` and `scientificProof` deleted. If it no longer makes complete sense, it fails. |
| **Shape** | The four-beat SHAPE in miniature: data-anchored open → the one insight → the one action → why it matters, in their numbers. |

A guardrail, autoregulation cue, or physician valve may appear **only** as a *condition on the
one action* ("do X — and if Y, stop / check with your physician"), never as a second imperative.

---

## PART 2 — `fullText`

**Job:** the readable depth. Contains the voice content **plus** mechanism, connection, week,
and watch-signals. Read, not spoken.

| Property | Value |
|---|---|
| **Length** | **350–550 words.** Training answers held to **~400 words.** The product may render it across 2–3 pages. |
| **Form** | Readable register — same coaching voice with room to explain. Not a more clinical author. |
| **Must contain** | (1) the **voice content** — same numbers, same single action, same tie-in, no contradiction with Part 1; (2) **one named physiological mechanism** in plain language, in the correct certainty register; (3) how it connects to **this user's** bio age, biomarkers, risk factors, and performance/fitness/recovery/stress ages, **by value**; (4) **what to do across the week** — the *same* action staged over days; (5) **what to watch** — the signals it's working + the contingency/physician valve if not. |
| **Must NOT contain** | A **second action** (stage the one action; do not add new ones); a restatement of Part 1 with no added depth; a diagnosis, medication advice, or clinician-style lab read; an over-claimed verb (see `calibration-rules.md`); a wearable metric presented as diagnostic or as a clinical-equation input. |
| **Calibration** | Fully calibrated: settled mechanisms get causal verbs; emerging/contested/correlational get hedged verbs; floor-level risks framed as maintain / widen-the-margin. |

`fullText` **expands** the SHAPE that `voice` follows in miniature — same beats, more room.

---

## PART 3 — `scientificProof`

**Job:** the study section — a short, credible evidence list that says "this is built on real
science," closing on the positive compliance line.

| Property | Value |
|---|---|
| **Length** | **3–5 references**, each **one line**, **plus** the closing compliance line. |
| **Each reference** | A **real, verifiable** study, clinical guideline, or established physiological mechanism. Include author / journal / year, or a PMID / DOI, **where known**. Each gets **one line on what it supports**, tied to *this* user's bio age / risk factors / the action in Parts 1–2. |
| **Never** | Invent a citation — no fabricated paper, author, journal, year, PMID, or DOI. If a specific paper can't be cited with confidence, cite the **named guideline or the textbook mechanism** (a real, checkable thing). A vague-but-true citation beats a precise-but-fake one. A fabricated reference fails the answer outright. |
| **Application line** | Each line must connect the evidence to **this twin** — not "exercise is good," but "supports the Zone 2 prescription tied to holding your bio age at 62 vs 59." (The evidence may be general; the tie-in line must be twin-specific.) |
| **Closing line** | The **positive compliance line**: the guidance is **informational, grounded in reliable medical science, NOT a medical examination, and not a substitute for a doctor** — framed as **credibility, not a referral.** It is *not* a "consult your doctor before…" disclaimer. See `calibration-rules.md`. |

**Reference vs. valve.** The Part-3 compliance line is a *standing positive frame on every
answer* and is **not** a referral. The signal-driven physician *valve* (master Rule 4) is a
separate, targeted referral that lives in Parts 1–2 only when a symptom or clinical signal is
present. Both can appear; they are different things.

---

## Coherence across the three parts

The three parts are **one answer**, not three:

- **One action** runs through all three: `voice` states it, `fullText` stages it across the
  week, `scientificProof` cites the evidence behind it.
- **The same numbers** anchor `voice` and re-ground `fullText`; the `scientificProof`
  application lines point back at those same numbers.
- **One named mechanism**: named in `voice`, explained in `fullText`, supported in
  `scientificProof`.
- **No contradiction**: the verdict, the action, and the tie-in are consistent across parts.

Length recap (hard): `voice` 90–110 words · `fullText` 350–550 words (~400 for training) ·
`scientificProof` 3–5 references + the compliance line.