docs: his segment, stable cost of capital and value-weight methods; live results

docs/ANALYSIS_STATE.md

@@ -196,6 +196,12 @@ deterministic keyword mapping in `damodaran_data.SEGMENT_INDUSTRY`). SpaceX: β
 initial cost of capital 7.92% vs his 8.37%. When the provider returns no statements, the historical base is
 built from SEC XBRL quarters (`SEC_ONLY`).
 
+**Segment builds, stable cost of capital, value weights** (2026-09-23): a segment-by-segment story is
+required when filed business lines differ in margin (≥ 20 pts) or industry risk (unlevered beta ≥ 0.25);
+the stable cost of capital is a mature company's, rebuilt from his `wacc.xls` market row at today's rf and
+ERP, reached after a five-year hold and a straight-line fade; business-mix weights are revenue × industry
+EV/Sales. See `DAMODARAN_REGRESSION_2026-09-23.md` §5.
+
 **Remaining gaps against his method:**
 1. **Liquidation value.** Declining firms have no liquidation-value floor. He compares going-concern
    and liquidation value; that needs asset-level data (e.g. book value by asset class).
@@ -206,10 +212,8 @@ built from SEC XBRL quarters (`SEC_ONLY`).
 4. **Revenue-weighted country exposure.** The country risk premium follows the country of operations
    (his list's country); he weights it by revenue or operations for multinationals, which needs
    geographic segment data.
-5. **Segment stories are optional.** Storytellers see the filed segment table but none built a
-   per-segment story for SpaceX; he always values multi-business companies segment by segment.
-6. **Stable-period beta** moves to 1.0; he kept SpaceX's stable cost of capital near its initial level.
-7. **Mix weights** are revenue weights; he weights segments by value.
+5. **Stable-rate override.** The stable cost of capital is the mature-company rate for every firm (his
+   default); he overrides it by hand for companies whose risk stays different (regulated utilities).
 
 ## 7. Validation checks removed or redesigned
 

docs/DAMODARAN_REGRESSION_2026-09-23.md

@@ -141,9 +141,40 @@ its filings were available.
   probability, so the engine defaults were used (that story was not the median).
 * The engine is not the cause: on his inputs it gives ≈ $87/share (§3).
 
-**Open methodology questions** (for the owner; not tuned):
-1. Require a segment build when two or more filed business segments have margins more than, say, 20 points
-   apart (his method), instead of leaving `segments` optional.
-2. Stable-period beta: move to 1.0 (current) or to the business-mix beta when it is below 1.
-3. Weights: revenue weights (current, from the filings) against his value weights (he weights segments by
-   their estimated value).
+**Method decisions implemented (2026-09-23, owner's instruction: "follow his method")**
+
+| decision | before | now (his method) | source |
+|---|---|---|---|
+| segment build | optional `segments` field; no storyteller used it for SpaceX | **required** when the filed business lines' operating margins span ≥ 20 points or their industries' unlevered betas span ≥ 0.25. Each segment carries its own market, share, margin and **sales-to-capital** (capital = Σ revenue ÷ sales-to-capital). A storyteller that omits a required build is asked once more and then rejected; a missing driver is asked for once, and a default that remains is disclosed (`defaults_used`) | his SpaceX, Amazon and Alphabet builds; his SpaceX update gives a lower sales-to-capital to the AI business |
+| stable cost of capital | beta moved linearly to 1.0 from year 1, at the company's own debt ratio and synthetic rating | **a mature company's cost of capital**: beta, debt ratio and default spread of his `wacc.xls` "Total Market (without financials)" row (US; `waccGlobal.xls` for non-US), rebuilt at today's riskfree rate and the company's ERP. The initial rate is held for the first half of the forecast, then runs in a straight line to the stable rate by maturity. Discounting uses the cumulated factor | video NQKIJU7TmTc: "8.37% for the next 10 years, and then to 8.25% beyond"; video WhY5EF1_LjQ: "about 8%, roughly that of an average US company"; his template's five-year hold and five-year fade |
+| business-mix weights | revenue | **estimated value** = segment revenue × its industry's EV/Sales (`psdata.xls` / `psGlobal.xls`); revenue weights only if an EV/Sales is missing (stated) | his bottom-up beta method |
+
+The method reproduces his numbers without fitting: the mature-company rate at his June inputs (rf 4.56%) is 8.17% (he used
+8.25%); at his April inputs (rf 4.2%) it is 7.91% ("about 8%").
+
+**Found and fixed during verification:**
+* The segment-build paragraph first printed the margin range as plain text; the report audit (R1) withheld the report.
+  The evidence is now structured and cited as claims.
+* Vistra's segments (Retail, East, Texas, West, with "Retail Energy Charge" product lines) mapped to general
+  merchandise retail, which moved its beta from 0.41 to 1.03. Region-named segments now keep the company's industry;
+  "retail" alone no longer maps to store retail; energy-sale lines map to Power; a mix that falls in one industry uses
+  the single-industry beta.
+
+**Live results after the change (2026-09-23):**
+
+| company | segment build | beta | cost of capital | value | result |
+|---|---|---|---|---|---|
+| SpaceX | required (margins −110% to +38%); all 3 storytellers built Connectivity / AI / Space | 0.96 value-weighted (AI 60% of value; revenue weights gave 0.69) | 9.02% → 8.38% (his 8.37% → 8.25% at a 0.39 pt lower riskfree rate) | **$23.27** (range $12.39–$31.73; was $12.52) | PUBLISHED SELL −12.3% |
+| Microsoft | required (margins 27% to 60%); all 3 built | 1.29 | 10.03% → 8.38% | $401.27 | PUBLISHED HOLD −1.4% |
+| Vistra | not required (one business) | 0.41 (Power) | 6.09% → 8.38% | $126.32 | PUBLISHED SELL −9.4% |
+
+SpaceX against his ≈ $100: the cost of capital now matches his method. The adopted story has $178B of year-10 revenue at
+a 28.9% margin, against his ≈ $400B at ≈ 40%. That is the storytellers' judgement, which is not tuned. The engine on
+his inputs gives ≈ $87 (§3).
+
+**Open point for the owner:** for a low-risk business the stable rate rises to the mature-company rate (Vistra
+6.09% → 8.38%). That is his template's default; he overrides it by hand when a company's risk will stay different
+(regulated utilities).
+
+**Replay after the change:** SpaceX Apr −8.7%, Jun −14.7% (was −14.0%), Nvidia (partial) −56.7% (was −49.5%, because his
+11.79% is now held five years as in his template), Anthropic unchanged. All inside the ±10-point regression band.

engine/damodaran_replay.json

@@ -0,0 +1,107 @@
+[
+ {
+  "id": "spacex_2026_04",
+  "company": "SpaceX (pre-IPO)",
+  "published": "2026-04-23",
+  "source": "blog 'To a Trillion(s) Dollars and beyond: A SpaceX IPO Odyssey!' + video WhY5EF1_LjQ",
+  "his_value": 1210000000000.0,
+  "his_value_basis": "value of operating assets (his: $1.2-1.21 trillion)",
+  "engine_value": 1104908897080.5183,
+  "gap": -0.08685215117312539,
+  "sales_to_capital": 1.64,
+  "partial": false,
+  "revenue_year_n": 320.0,
+  "margin_target": 0.472,
+  "terminal_growth": 0.04,
+  "pv_terminal_share": 0.834,
+  "engine_value_g_equals_rf": 1146873451727.6062,
+  "gap_g_equals_rf": -0.05217070105156507,
+  "assumptions": [
+   "current operating margin -13% (his pre-prospectus estimate of an operating loss of about $2B)",
+   "sales-to-capital: revenue-weighted from his US industry dataset (not published for this valuation)",
+   "tax 24% (engine convention; his marginal 25%)",
+   "no R&D capitalisation (he capitalises R&D)"
+  ]
+ },
+ {
+  "id": "spacex_2026_06",
+  "company": "SpaceX (post-prospectus)",
+  "published": "2026-06-04",
+  "source": "blog 'Revisiting the SpaceX Valuation: A Post-Prospectus Update!' + video NQKIJU7TmTc",
+  "his_value": 1220000000000.0,
+  "his_value_basis": "value of operating assets (his: about $1.22 trillion; ~$100/share after IPO cash)",
+  "engine_value": 1041196989323.8682,
+  "gap": -0.14655984481650153,
+  "sales_to_capital": 1.62,
+  "partial": false,
+  "revenue_year_n": 400.0,
+  "margin_target": 0.396,
+  "terminal_growth": 0.04,
+  "pv_terminal_share": 0.855,
+  "engine_value_g_equals_rf": 1093421222415.5911,
+  "gap_g_equals_rf": -0.1037530963806631,
+  "assumptions": [
+   "current operating margin from the prospectus: -$2.6B on $15.5B",
+   "sales-to-capital: revenue-weighted from his US industry dataset; he added reinvestment in years 1-5 from prospectus capex/R&D, which is not modelled here",
+   "tax 24%",
+   "no R&D capitalisation"
+  ]
+ },
+ {
+  "id": "nvidia_2025_01",
+  "company": "Nvidia (post-DeepSeek)",
+  "published": "2025-01-31",
+  "source": "blog 'DeepSeek crashes the AI Party: Story Break, Change or Shift?'",
+  "his_value": 1883000000000.0,
+  "his_value_basis": "$78/share \u00d7 ~24.5B shares, less cash ~$38B, plus debt ~$10B (approximate EV)",
+  "engine_value": 814726404095.1594,
+  "gap": -0.5673253297423476,
+  "sales_to_capital": 1.21,
+  "partial": true,
+  "revenue_year_n": 180.0,
+  "margin_target": 0.6,
+  "terminal_growth": 0.04,
+  "pv_terminal_share": 0.568,
+  "engine_value_g_equals_rf": 846036843690.9175,
+  "gap_g_equals_rf": -0.5506973745666928,
+  "assumptions": [
+   "PARTIALLY SPECIFIED: revenue base, non-AI revenue, share count, cash and sales-to-capital are not in the post",
+   "year-10 revenue = AI chips only ($300B \u00d7 60%); his model likely adds other businesses",
+   "sales-to-capital: his US Semiconductor industry value"
+  ]
+ },
+ {
+  "id": "anthropic_2026_08_10y",
+  "company": "Anthropic (breakeven at a $2T pricing)",
+  "published": "2026-08-20",
+  "source": "video AAi9QIl6gw8 'AI's Bar Mitzvah Moment?'",
+  "his_value": 1200000000000.0,
+  "his_value_basis": "revenue needed in year 10 to justify $2T (his)",
+  "engine_value": 1708682515507.9536,
+  "gap": 0.423902096256628,
+  "sales_to_capital": 1.54,
+  "partial": false,
+  "assumptions": [
+   "after-tax operating margin 30% at maturity (his), i.e. pre-tax 39.5% at the engine's 24% tax",
+   "revenue now = $65B annualised run rate (his figure for July 2026); margin starts at 0 and converges",
+   "sales-to-capital: his US Software (System & Application) industry value"
+  ]
+ },
+ {
+  "id": "anthropic_2026_08_15y",
+  "company": "Anthropic (breakeven at a $2T pricing)",
+  "published": "2026-08-20",
+  "source": "video AAi9QIl6gw8 'AI's Bar Mitzvah Moment?'",
+  "his_value": 2000000000000.0,
+  "his_value_basis": "revenue needed in year 15 to justify $2T (his)",
+  "engine_value": 1943346097171.3552,
+  "gap": -0.0283269514143224,
+  "sales_to_capital": 1.54,
+  "partial": false,
+  "assumptions": [
+   "after-tax operating margin 30% at maturity (his), i.e. pre-tax 39.5% at the engine's 24% tax",
+   "revenue now = $65B annualised run rate (his figure for July 2026); margin starts at 0 and converges",
+   "sales-to-capital: his US Software (System & Application) industry value"
+  ]
+ }
+]