Type: thesis-shift (subtype: corrects — completing yesterday's structural correction with the right ticker)

event_id: 2026-05-11-crs-promotion-ati-removal-from-cohort-key-tickers-completing-yesterday-correction

Source: Internal — follow-through on the 2026-05-10 ATI structural correction; user delegated the call ("you make the choice"); ATI deep-dive revised, CRS deep-dive written, watchlist updated, README key-cohort section refreshed.

Shift: Closing out yesterday's correction with concrete actions rather than leaving the open question hanging:

1. CRS added to ai-power.json watchlist (45 stocks total). Deep-dive at 2026-05-11-crs-deep-dive captures the alloy-layer-for-IGT thesis under the correct ticker. Two converging surface paths: (a) 2026-05-06 blind research-trace surfaced CRS at 2/3 convergence independently of any human-narrative-down framing; (b) 2026-05-10 ATI FY2025 10-K explicitly names CRS as the HPMC-segment competitor in "legacy nickel-based alloys and superalloys" — the IGT-blade-relevant alloy class. Two independent paths converging on the same name without coordination = substantive-class convergence pattern. CRS financial trajectory consistent: FY26 Q3 (Mar 2026) rev $811.5M +11.6% YoY, op income $186.5M +35.3% YoY, op margin 23.0% (+400bp YoY). $21.3B mcap. First ticker added via derived-then-verified path rather than narrative-down path — Track B/B4 cold-problem-practice loop validation case. The own-thesis (CRS as alloy-layer-above-HWM) was generated before sources converged; we can now track whether substantive-class sources (@Gaurab / @ChurchillWw / @vikramskr / @vmaxpax) eventually surface CRS independently in public posts. If yes within 30/90/180 days → we matched lead time. If no → either we found something they missed, or the market doesn't reward it.

2. ATI removed from cohort-key-ticker status, retained in watchlist. ATI's 2026-05-06 deep-dive carries a 2026-05-10 correction block. ATI is no longer framed as "alloy layer above HWM in 3-layer IGT-blade stack" (structurally wrong per ATI's own 10-K). Reframed as "diversified specialty-alloy with commercial-jet-engine as the actual growth driver" — Q1 2026 HPMC Jet Engines - Commercial +8.5% YoY is real and works as a commercial-aerospace-recovery position. ATI deep-dive conviction revised medium → medium-low. Sizing should reflect cyclical-aerospace exposure, not IGT-stack exposure. The watchlist description in ai-power.json carries the re-framing language explicitly so future-session ticker reads see the correction.

3. CLAUDE.md guardrail promoted from the 2026-05-10 meta note. The "vertical-reality check for stacked supply-chain theses" pattern (pull upstream ticker's 10-K Item 1 competitor list + customer-concentration disclosure before publishing a multi-layer-stack thesis) is now a standing rule. Per CLAUDE.md lifecycle: observation → identifies gap → tool/prompt upgrade → CLAUDE.md update so future sessions use the new rule. This is the CLAUDE.md update step — completing the methodology insight rather than leaving it as a meta note that doesn't fire.

Effect on thesis:

• HWM key_ticker conviction unchanged — already locked in at "vertically-integrated value capture" framing after the FY25 10-K + Q1 2026 reads.
• CRS new key_ticker, medium conviction. Sized as a sub-position (~1/2 to 1/3 of HWM weight) given (a) the IGT-pull-through hypothesis is still inferred-by-elimination not directly confirmed by CRS customer disclosure; (b) +110% 12m multi-year-run reversion risk; (c) PCC (Berkshire-owned, private) is the larger competitor — CRS captures share at the margin.
• ATI removed from cohort-key-ticker. Watchlist retention captures the commercial-aerospace tilt without anchoring future sessions to the wrong "alloy layer above HWM" frame.
• Methodology validation case logged. The derived-then-verified path is now a working pattern, not theoretical. The 2026-05-06 own-theses spike (2026-05-06-derive-skill-smoke-test-gas-turbine) named CRS at 2/3 convergence; the 2026-05-10 ATI 10-K read independently confirmed the alloy-layer position. The Track B/B4 cadence question ("did we surface something our sources eventually land on") gets its first real test case with CRS.
• Open question forward-traveling: does any substantive-class source (Gaurab / ChurchillWw / vikramskr / vmaxpax) surface CRS publicly within 30/90/180 days? Calendar-track at TASKS.md.

Per-ticker: CRS ↑↑ (new key_ticker; derived-then-verified path; sub-position size). ATI ↓ (removed from cohort-key-ticker; retained as commercial-aerospace watch-name). HWM unchanged.