Gated predictions on AI/robotics × cost of living

Codelco officially capitulated on Jul 28 2026: chairman Bernardo Fontaine ruled out the 1.7Mt-by-2030 recovery goal ('no possibility'; 'for seven years, Codelco hasn't met its projections — and this year is no exception'), guided output flat at ~1.33Mt for the coming years (2026 guidance 1.331-1.357Mt), suspended approvals for new projects pending a capital-spending review, and will present a profitability-first recovery strategy in Oct/Nov (Bloomberg via Mining.com.au/TheDeepDive; Reuters Jul 16 precursor). Headline P50 ($15K canonical tier) HOLDS at 2028: that tier is demand-anchored, China is soft (Jul imports down on weak demand; ICSG spring forecast even shows a 96kt 2026 surplus), and the analyst bull-case midpoint (Goldman $13,735 end-2026, Citi $15,000 within a year, UBS $14,500 year-end) still frames 2027-2029. Sub-gate copper-sustained-20k P50 moves 2032 -> 2031: the gate's own calibration put full 1.7Mt-failure at ~400kt/yr = ~2y on this tier, but consensus skepticism was already partially priced in ('analyst consensus already skeptical', Jul 16 state) and the Peru +475kt/y unlock (on track — Fujimori inaugurated Jul 28, pro-mining agenda confirmed) still nets against it; the genuinely un-priced increment — official flat guidance vs realistic ~1.45-1.5Mt consensus recovery paths (~150-250kt/yr) plus the new-project approval freeze that uncovers grade decline in exactly the post-2030 window this tier fires in — is worth ~1y in, not the full 2y, avoiding double-count with the Jul 21 Peru move. Sub-gate copper-sustained-25k P50 moves 2034 -> 2033 to maintain tier ordering; the structural-project freeze bites hardest in the deep 2030s. Corroborating tape since Jul 16: LME stock 238,350t on Aug 4 (roughly halved since late May), spreads flipped to backwardation late Jul, cash $14,195/t (Westmetall).

Reverts the June pull-in: both legs of it failed re-verification, and the first direct measurement of the gate's own criterion landed below the bar. (1) METR — the dashboard's SOTA has been frozen at 17.41h (Claude Mythos Preview, released 2026-04-07, added 2026-05-08) with no new model added in four months, above METR's own posted ceiling 'Measurements above 16 hrs are unreliable with our current task suite'; the published doubling estimate is now 128.7d (2023-on, CI 104-158) and 187.8d all-time-stitched, not the 89d this gate assumed, and METR pivoted to a new 'expenditure horizon' metric in July 2026. The gate's own METR dependency prices a doubling-rate reversion at +1-2y; taking the low end gives +1y. (2) SWE-bench Pro — Scale's standardized public set (identical scaffolding) tops out at 61.5% (Muse Spark 1.1), not the 77.8/80.3% vendor-harness figures the June entry read off it, so the contamination-resistant 75% sub-gate is not cleared. (3) Anthropic's 2026 Agentic Coding Trends Report measures the criterion directly: engineers use AI in ~60% of their work but can 'fully delegate' only 0-20% of tasks; Anthropic's own 'When AI builds itself' reports >80% of merged code Claude-authored while naming human code review the new bottleneck — the 'without human review on each step' clause is exactly what has not happened. GPT-6 Astra (2026-09-03) is a real capability jump but ranks 22nd on GDPval-AA v2 (1,582 vs Fable 5.1's 1,766) and 6th on AA-Briefcase, and OpenAI omitted GDPval from launch materials. Confidence stays medium: capability and commercial momentum are intact, so P90 holds at 2033. — https://metr.org/time-horizons/; https://labs.scale.com/leaderboard/swe_bench_pro_public; https://resources.anthropic.com/2026-agentic-coding-trends-report; https://www.anthropic.com/institute/recursive-self-improvement; https://artificialanalysis.ai/evaluations/gdpval-aa

The September FOMC fired this gate's own pre-registered bearish re-flag, and the canonical $7,000 tier gives the year back: P50 2028→2029, P10 2027→2028. The hike itself is the smaller half of the story — 25bp to 3.75–4.00% on 16 September, unanimous 12–0, the first increase since July 2023 [73]. THE PATH IS THE MOVE: the Fed's own SEP median now puts the funds rate at 4.1% at end-2026 and 4.1% again at end-2027 (June: 3.8% and 3.6%), 3.9% at end-2028 and 3.6% at end-2029, with PCE inflation not reaching 2% until 2029; 16 of 18 participants see at least one further hike this year, and the 27–28 October meeting is ~55–60% priced [74][79][81]. Warsh read it out himself — the appropriate rate is 4.1 percent at the end of this year, and to remain there next year — alongside 'this summer's inflation readings do not tell me that underlying trends have meaningfully improved' and being 'hard-pressed' to call financial conditions restrictive [75]. This gate defines the $7K tier as requiring the Fed cycle to turn to cuts in 2027; the Fed has removed that, a ~12-month delay in the tier's own named precondition, which is what clears the ≥1y bar. The August upgrade rested on four falsified premises and explicitly held itself to one year because premise (d) — a hold is not a cut — survived. That premise is now void, and premise (c) collapsed with it: the labour-market evidence behind 'the hike was priced out' was revised away, with July nonfarm payrolls restated from −23,000 to +21,000, June from −20,000 to +31,000, and August printing +162,000 against a ~53,000 consensus at 4.1% unemployment [76]. August CPI was +0.4% m/m and 3.4% YoY, core +0.3% m/m and 2.4% YoY, with energy +2.1% m/m contributing over a third of the monthly gain [77]. WHY ONLY ONE YEAR, AND WHY NOT $10K: the demand side strengthened in the very same window. Gold-ETF holdings hit an all-time record 4,189t in August on a US$18bn / 121t inflow — the second-largest month ever, with Europe's strongest month on record and North America's third-largest, returning the region to positive YTD — which resolves this gate's 'fragile Western bid' bear case against itself [82]. The PBoC added ~20t in August to 2,386.6t, its largest monthly buy since October 2023 for a second consecutive month on a 22-month streak [85]. And Goldman reaffirmed $5,400 for end-2027 on 18 September *after* the hike, with tighter policy hitting 'the near-term appreciation path rather than a lower terminal gold price' [86]. The trigger narrative that gold 'didn't care' does not survive the tape: spot fell ~5% from $4,538 on 21 August to $4,306 on 14 September as hike odds went 31–36% → 83–85%, then steadied to $4,383–$4,394 by 18 September for its first weekly gain in four — sell the rumour, buy the fact, not immunity to rates [81][87]. $6,000 moves 2027→2028 and $8,000 2029→2030 in train. DELIBERATELY NOT MOVED: $10,000 held at 2031 for a second refresh (a regime call, not a cyclical one) and P90 held at 2032 — the structural floor is *firmer* than in August, so the far tail did not get worse even as the median moved right. TWO CORRECTIONS TO THIS GATE'S OWN NUMBERS: WGC full-year 2025 official-sector buying is 863.3t, not the 1,237t carried since May — −21% from 2024's 1,092.4t and the lowest since 2021, so the 'three consecutive years above 1,000t' floor claim was never right [84]; and Manifold's end-2027 book, whose $5,893 expected peak anchored August's $6,000-in-2027 call, has not traded since — identical distribution, same 14 holders and 43 trades — so it is stale, while the two books that did trade fell to 14% and 5% [37][38][39]. WATCHLIST: the 'Fed real rate trajectory' entry is marked FIRED and its re-flag reset; 'BRICS gold-backed settlement scaling' resolved exactly as predicted at the 12–13 September New Delhi summit — BRICS Pay endorsed, no common currency, The Unit not advanced — and therefore carries no timeline effect [89]. CONFIDENCE STAYS LOW: Treasury's doubled long-end buybacks went live on 9 September and the 30-year still sits at 5.29–5.36% while the 10-year blew out ~51bp to a 2007 high near 5.00%, so the fiscal-repression leg is pressing but not working; the Q3 central-bank print is not out until late October; and the gate's own $6,000 tier now sits ~10% above the live sell-side 2027 cluster [78][86][90][91].

June MOU collapses — $55 tier goes back to 2028 and p10 to 2028; canonical $50 P50 held at 2029, confidence held medium. Verified: the Jun-14/19 US-Iran MOU broke down within weeks (Iranian drone hit a cargo ship Jun 25, truce collapsed Jul 8, IRGC re-declared Hormuz closed Jul 12, US naval blockade reinstated); an Aug 4-5 interim Hormuz deal failed to land; Treasury Secretary Bessent said Aug 13 the blockade could run indefinitely with 'measures like have never been seen in the history of economic isolation'; Trump said Aug 14 he will declare the strait 'a territory of the United States' after 'defeating Iran' — a verbal claim with no legal instrument attached [34][35][36][37]. WTI front-month is $82.40 (Aug 14, +5.4% w/w) vs ~$75.5 on Jun 16 [34]. So the war premium is back — but only at the front of the curve. **What moved:** the $55 sub-gate 2027 → 2028 and p10 2027 → 2028, because the *specific premise* of the June pull-forward is falsified — Goldman's 'full Persian-Gulf export recovery by end-July 2026' is now EIA's 'near pre-conflict averages in early 2027' with ~0.6 Mbpd residual disruption through end-2027, and Polymarket/Kalshi price only ~45% on normal Hormuz traffic by Dec-2026/Jan-2027 [32][41]. That pushes the post-normalization overshoot window out of calendar 2027. **What did not move:** the canonical $50 P50 (2029), the $40 P50 (2033) and confidence (medium) — because the *destination* moved the other way and roughly cancels the delay. The Aug STEO cuts 2027 WTI to $65.39 with Q4'27 at $60.97 (from ~$68-70 in June) on a 4.78 Mbpd 2027 inventory build (vs Goldman's 3.2 Mbpd); the IEA now has 2026 demand *contracting* 1.6 Mbpd with only +2.4 Mbpd back in 2027; OPEC cut 2026 growth to 580 Kbpd; China is at ~50 Kbpd growth with gasoline -5.5% and EVs displacing 1.4 Mbpd in H1'26; OPEC+ finished unwinding the 1.65 Mbpd 2023 tranche with the September quota; and CME WTI Dec'27 at $70.52 is *below* its May print of $72.71 [32][33][38][40][42]. The offsetting reason $50 does not come forward: the 2027 'surplus' mostly refills an emptied tank — OECD commercial stocks fall to 2,477 Mbbl by end-2026 and only reach 2,905 Mbbl by end-2027, and the SPR is at 298.7 Mbbl, the lowest since January 1983, so strategic restocking (>1 Mbpd in 2027 per Goldman) absorbs much of it, which is exactly why EIA's own 4.78 Mbpd build still lands at $61 WTI rather than $50 [32][43]. Net: later path, lower destination, wider variance — a one-year slip in the near tier, no move in the canonical.

P50 pushed back one year (2028→2029) because the OECD-market-access assumption that carried the May pull-in has been falsified, and all three Western consumer paths slipped in the same quarter. Regulatory: the FCC added 'advanced robotic devices' (humanoids + quadrupeds) to its Covered List on July 28, barring **new** foreign-produced models from the equipment authorization required to import, market, or sell in the US — grandfathering previously authorized models, exempting foreign-owned firms that manufacture domestically (Boston Dynamics, 1X Hayward), and offering only a Department of War conditional-approval waiver; this follows the DoD adding Unitree to the Section 1260H Chinese Military Companies list June 8 and Korea's March final antidumping ruling of 15.96-19.85% on Chinese robots. May's explicit premise — that humanoid protectionism 'may form slower because there's less existing domestic industry to protect' — is now wrong in two OECD jurisdictions, with the EU Cyber Resilience Act (applies Dec 11 2027; 'smart robots' in scope) a third wall arriving inside the P50 window against a Unitree stack carrying the documented UniPwn root exploit, hardcoded pi/123 credentials, and undisclosed 5-minute CloudSail telemetry. Product: 1X has still not made a verified NEO consumer delivery (softened to 'some this year, some later'), the WSJ hands-on found every household task teleoperated, Figure 03 has no public price/pre-order/retail channel and home use remains internal alpha, and Optimus V3 was still unrevealed with no date set at the July 22 Q2 call. Metaculus sub-$20k-by-2030 eased 79%→70%. Offsets are real but do not restore the conjunction: Unitree launched commercially in Europe July 22, R1 fell to RMB 29,900 (~$4,150), Morgan Stanley doubled 2026 China shipments again to 50,000 units, and π0.7 showed zero-shot cross-embodiment household generalization. The slip is one year, not the 2-4 years this gate pre-registered for a national-security designation, precisely because grandfathering plus the open EU/Japan lanes leave the route narrowed rather than eliminated. P90 widened back to 2034: if security-driven exclusion coordinates across the OECD, the gate falls back on a Western BoM whose actuators cost $58k without Chinese content.

Confidence medium → high and P90 2034 → 2033; P50 holds at 2029. Three watchlist factors flipped accelerating since May, all verified at the primary source. (1) Regulation: NHTSA granted Zoox the first-ever passenger-carrying ADS exemption (Exemption No. 2026-01, Federal Register 2026-07-31, effective through 2028-07-31) — up to 2,500 purpose-built vehicles/yr with paid commercial deployment authorized under a new Operational Authorization oversight regime; paid Las Vegas service starts August 2026. That resolves this gate's only both-direction regulatory factor ~6 months ahead of the sub-gate's 2027 P50, and the same-day AV Framework interim-guidance RFC plus pending FMVSS 108/135 amendments make the purpose-built pathway systemic — killing the 'denial extends reliance on retrofitted platforms' branch that fed P90 2034. (2) Capability: the Zeekr-built 6th-gen Ojai opened to public riders 2026-05-28 (~$32K base vehicle vs $150-200K outfitted I-Pace per FutureSearch), and Waymo announced four new markets 2026-07-08 (Las Vegas driverless immediately; Denver, San Diego, Tampa validating on the ~$50K Hyundai Ioniq 5) — 15 announced US markets, 3x the shots on goal for a 5-city profitability threshold. (3) Cross-gate: autonomous-freight-delivery moved P50 2033 → 2032 (2026-07-29) on Aurora's gen-2 hardware cost halving and incumbent-carrier commitments — shared-stack corroboration that the AV cost curve is tracking projections. P50 stays 2029 because the trigger's confirmation half did not move: Alphabet's Q2 (2026-07-22) kept Waymo aggregated in Other Bets ($382M revenue / $1.8B loss, widened YoY), weekly rides plateaued at ~500K March→July behind a ~3,900-vehicle construction-zone recall with a ~5-week freeway suspension (resumed end-July), and the one new independent quantification — FutureSearch (2026-05-27) — puts Q4-2026 rides at median 775K and 'meaningful profit will not show up before 2029', landing exactly on this gate's P50. Teleop ratio (~1:41) and AV-fleet insurance pricing showed no new prints; the disclosure sub-gate (P50 2028) is unmoved.

P50 2033 → 2032 on four verified state-changes in the trucking leg, which is ~75-80% of the combined freight + last-mile dollar denominator. (1) Cost curve: Aurora's second-generation Aurora Driver, launched 2026-07-22, cuts hardware cost by more than half versus gen-1 and is built to a 1M-mile design life [23][24] — direct evidence the $/mile curve is tracking or beating Goldman's 2028 human-parity crossing. (2) Manufacturing: Roush is targeting a 1,000 Aurora-equipped-truck annual run-rate by end-2026 [25], 3-5x the 'hundreds per year' ramp this gate assumed, with OEM series production behind it (Volvo VNL Autonomous, Daimler/Torc Cascadia L4 with Innoviz LiDAR selected May 2026). (3) Demand is now the non-binding side: Aurora's commercial capacity is fully committed through Q3 2026, Hirschbach signed a non-binding MOU for 500 trucks from 2027 [26], and TFI International — a top-5 North American carrier — told analysts on 2026-07-28 it will roll out autonomous US linehaul in 2027 with testing starting within months [27]. (4) Labor displaced per truck is ~2x what this gate modelled: Aurora's ~1,000-mile Fort Worth-Phoenix lane is the first driverless freight route running beyond Hours of Service limits, cutting transit nearly in half versus a single human driver [28], so one autonomous tractor substitutes for roughly two human tractors in ton-miles and roughly halves the fleet needed to move a given share of national ton-miles off driver labor. Regulatory adds to this: federal AV-truck preemption migrated from the party-line SELF DRIVE Act to Subtitle E of the must-pass surface transportation reauthorization (H.R. 8870), reported out of House T&I 62-2 on 2026-05-22 [29], while California finalized heavy-duty AV rules on 2026-04-28 [30] and the Teamsters' in-cab-operator bill AB 33 died in the Senate inactive file. Partly offsetting, and why this is one year and not more: the FAA's Part 108 BVLOS comment period was reopened twice in early 2026 and no final rule had published as of 2026-07-29 [31], sidewalk-robot state authorization is flat at ~24 states, and Aurora's FY2026 revenue guide is unchanged at $14-16M so the $100M-revenue sub-gate has not moved. P90 2040 → 2039 because the specific worst case this gate named — Teamsters winning the political war state-by-state — has partially falsified in the largest blocking state. P10 held at 2030: nothing made the fast tail more achievable, since ~250 driverless trucks in service today cannot compound to ~35% of national ton-miles by 2030. Confidence stays medium — the trucking leg's uncertainty narrowed but the last-mile leg's widened, and the decisive federal action is unresolved.

Three independently-measured state changes, all adverse, all post-dating the May refresh. (1) FAA type certification — this gate's #1 dependency, previously P50 2027 — is crawling, not closing. Joby's own published stage table (Q2 2026 shareholder letter, data as of 2026-07-31) shows Stage 5 'Show & Verify' at 20% Joby / 10% FAA, up from 15% / 6% in the Q1 letter (data as of 2026-05-01): +5pp and +4pp in a quarter. Stage 4 is also *not* closed (83% / 77%), correcting this gate's prior claim that Joby 'cleared Stage 4' in March 2026 — it entered Stage 5 with Stage 4 still open. Joby has dropped any type-certificate date from guidance; Archer's Q2 release names none either. At the observed FAA-side rate a 2027 certificate needs ~6x the Q2 pace sustained for six quarters; TC P50 moves 2027 → 2028. (2) Vertiport reality is thinner than the buildout narrative: 7 vertiports operational worldwide as of 2026-07-26 (9 under construction, 82 planned), and *no metro has more than one operational* — NYC 1, LA 0, Dubai 0. The '10+ pads in one metro by 2030' sub-gate is not reachable on a 2-4yr/site pace; it moves to 2032. (3) Fleet math, which is what actually generated P50 2032, was falsified: Joby has five aircraft flying, 12 in production and **two planned for delivery in all of 2026** against this gate's assumed 2/month now → 4/month by 2027; the Toyota JV (2026-06-30) is explicitly 'groundwork' for high-volume production, i.e. the ramp has not started. The EU/UK fallback path also slipped — Vertical Aerospace re-baselined Valo certification to 2029 (FlightGlobal, 2026-07-13). The gate's own dependency wording prices a 12-month TC slip at +1-2y on the parent; I take the low end (+1y → P50 2033, P10 2030) because the accelerants are real and address the downstream bottlenecks: the eIPP gives a pre-certification operational-learning path across 11 states, Atoms (Kalanick, $1.7B raised) is financing multimodal hubs in FL/NY/TX/CA, Florida legislation funds vertiport projects up to 100%, and demand is better than this gate's bear case — Joby-owned Blade grew seats flown >50% YoY in Q2 2026 with Joby stating aircraft availability, not passenger demand, is now the constraint. P90 holds at 2038 and confidence holds at medium-low: the structural failure mode is unchanged, and we are better instrumented than in May (a published quarterly stage table, actual delivery counts, a vertiport census). The trigger that prompted this refresh is a non-event for the gate — the coast-to-coast 'autonomous' flight is the J208, a converted Cessna Caravan running supervised autonomy with a ground-based remote Pilot in Command for freight/medical/defense, not the S4 powered-lift air taxi, and the Texas eIPP week (2026-09-10/14) was piloted demonstrations with no paying passengers. — https://www.sec.gov/Archives/edgar/data/1819848/000181984826000432/joby-8kx20260805xexx99_2.htm; https://www.sec.gov/Archives/edgar/data/1819848/000181984826000321/joby-8kx20260505xexx99_2.htm; https://www.jobyaviation.com/news/joby-launches-eipp-flights-in-texas; https://www.jobyaviation.com/news/joby-autonomous-aircraft-begins-cross-country-tour; https://evtol.travel/vertiports; https://www.flightglobal.com/archive/2026/07/vertical-aerospace-confirms-new-delay-to-valo-certification/

Re-analysis after humanoid-10m-households was added; P50 pulled in by one year because (a) Figure explicitly confirmed Figure 03 units are now performing component assembly inside BotQ (not only logistics), (b) NVIDIA GR00T N1.7 introduced the first dexterity scaling law on 20k+ hours of human video, and (c) IDTechEx's May 2026 model puts humanoid payback at 6 months under high utilization in 2026 — collectively tightening the BoM-and-capability stack that gates self-build economics. The new humanoid-10m-households gate is synergistic (volume demand creates factory pull) but not on the critical path.

P10 pushed back one year (2031→2032); P50 2035, P90 2043 and confidence medium all held. The pre-registered trade-restriction factor — flagged in key_dependencies as the most structurally dangerous P90 path — partially materialized: the FCC added 'advanced robotic devices' (humanoids, quadrupeds, and consumer robots down to vacuums) to its Covered List on July 28, 2026, barring new foreign-produced models from the equipment authorization required to import, market or sell in the US — prospective-only (existing Unitree G1/R1 authorizations grandfathered), production-location-based (any foreign country, even US brands assembling abroad), domestic manufacturers exempt (1X Hayward, Boston Dynamics), and a Department of War conditional-approval path demanding a full component country-of-origin BoM plus a time-bound US onshoring plan. It followed DoD adding Unitree to the Section 1260H list (June 8) and Korea's final 15.96-19.85% antidumping duties (March); China's Ministry of Commerce threatened 'resolute' retaliation on July 30, with rare earths the analyst-flagged lever. Sized once to avoid double-counting the two arrival paths: the indirect path via humanoid-retail-20k nets to exactly zero — that gate went 2029→2028 (May 24) and back to 2029 (July 29, same FCC action) while this gate's frontmatter carried 2029 throughout — and the direct path is worth well under a year on the global 10M integral, since the US slice of the pre-2036 cumulative is ~15-25% and only its cheap-import wing (the sub-$10k tier that 69%-wont-pay-over-$5k US survey economics require) is blocked. Offsets since May run through the China-led engine that dominates the volume integral anyway: Morgan Stanley doubled its 2026 China shipment forecast to 50,000 (June 24), Unitree's ~$620M STAR Market IPO opened subscription in early August to fund its 75k/yr capacity plan, R1 sits at RMB 29,900 (~$4,150, under the $5k US consumer-survey ceiling), and JD.com opened the first consumer e-commerce humanoid channel (618 auction; >10B RMB 2026 robot-sales target; UBTech UWorld consumer pre-orders). P10 still moves because its pre-registered conjunction is broken: leg (b) required Chinese consumer humanoids extending to OECD markets without trade restrictions (falsified in the US and Korea), and leg (a) — Optimus ~100k/yr by end-2028, NEO 100k/yr by end-2027 — degraded in the same quarter (zero verified Optimus units at the July 22 Q2 call with V3 unrevealed; first NEO customer deliveries still pending as of mid-July). A 2031 crossover would need China-domestic plus EU adoption to carry nearly the whole 10M in five years — no longer a 10% path. P90 holds at 2043 because the tail was already priced for exactly this scenario (grandfathering, open EU/Japan lanes and the untouched Chinese domestic market keep partial materialization from extending it), and P50 holds because the verified change nets below the ≥1-year threshold and the analyst anchor is unmoved (BofA 2033-34 / Citi 2034-36 / SNE ~2037 / MS 2038-40), now joined by a first Metaculus anchor (~4.11M cumulative all-sector units by 2030) consistent with a mid-2030s household crossover.

State-ban update. South Dakota became the 8th US state to ban cultivated meat (SB124 — a 5-year moratorium after Gov. Rhoden vetoed an indefinite version), and Mississippi became the first state to ban cultivated *dairy* (HB1153, signed 23 Mar 2026, effective 1 Jul 2026) on top of its existing 2024 meat ban. The 11th Circuit's March-2026 affirmation of Florida's ban (surviving federal preemption) makes these statutes durable, and the category-ban model is now broadening beyond meat — P90 bear tail nudged 2045→2046. Central P50 held at 2036: at 8 states the count is still below the 12-15-state threshold the gate models as the +2y trigger, and accelerants (Mission Barns' on-track Q3-2026 Sprouts launch) offset. Confidence stays low.

Re-analysis after home-grown-veg-beats-store was added; P50 unchanged at 2042, P10 pulled in by one year (2034→2033) on stronger 2026 enablers (US-Israel ag deal in force, EDF NIS 0.065/kWh solar PPA — lowest ever, agrivoltaic framework approved, robotic-harvest seed wave), P90 pulled in by two years (2055→2053) because home-grown demand-side substitution now applies a (modest) ceiling on retail margins. Same-week Carrefour 'Basket of Israel' data showing competing chains raising prices rather than matching cuts confirms structural stickiness — P50 doesn't move.