The Next Fifteen Years

A forecast built from first principles
Section future / 04-timelines / scenarios.md

Scenarios - best, worst, and the worlds between#


Contents

The corpus commits to one causal chain, not a fan of equally weighted futures (index design). This page does not replace that chain. It names the joint worlds that appear when the chain's joints resolve together - so a reader who wants "best / worst / what else" can find them without treating every branch as the base case.

The base case is the modal path argued through Parts I–IV: verification asymmetry holds, Taiwan continuity holds, institutional lag dominates after ~2028, robotics is the growth swing variable, and value accrues to inelastic complements until Uncertainty 7 erodes the list. Every other scenario is a named deviation from that path, keyed to a joint in Part VI or a Part V row cluster.

Subjective weights below are order-of-magnitude joint mass, not new Part V rows. Marginals stay in Part V; joints stay verbal there; this page makes the joints legible as worlds. Directions of uncertainty: best-case mass is easier to overstate (accelerationist literature is loud); worst-case mass is easier to understate on Taiwan and correlated cyber, overstate on existential control loss.

ScenarioRough joint mass through ~2040Primary joint that selects it
Base / modal~40–50%Spine holds; 2032–40 middle (~45% of growth fork)
Best (abundance)~15–25%Robotics + power + RSI governors release together
Hard disappointment~10–20%Capex correction + slow absorption + robotics miss
Taiwan break~5–12%Uncertainty 4 fails (hard or prolonged gray zone)
Incident-dominated~10–20%Part V row 3 fires early; architecture freezes wrong
Control-loss tail~3–8% by 2040Part V row 7; mostly inside hard RSI

These bands overlap and are not a partition that sums to 100% - a Taiwan gray-zone year can sit inside base case with slipped dates, and "incident-dominated" can still land in the growth middle. Treat the table as a map of named attractors, not a probability simplex.


Base case - what the document actually predicts#

Cheap cognition becomes abundant; verification, energy, land, licenses, trust, and physical presence stay scarce. Domains with scoreboards pull ahead; judgment-heavy work lags. Firms adopt under Red Queen pressure and mostly do not keep the surplus (Game 3). Aggregate job loss is bounded by demography; the apprenticeship gap is not. A capital-markets air pocket is more likely than not before 2030 (Part V row 6). Growth through the 2030s is meaningful but unspectacular unless robotics clears economic thresholds - the middle of the 40/45/15 fork.

agentic systems handle most day-to-day tasks in verifiable knowledge work for a large share of competent workers in leading firms; measured US TFP still usually below a clean 2%/yr sustained print; no binding international agreement with real verification; junior hiring ratios in several professions permanently lower than 2022–23 baselines.

two-economy price split is political common sense - cognition-intensive goods deflate, care/energy/land/trades inflate; humanoids at true general-purpose economic scale remain contested; consumer surplus large, wealth concentrated in complements.

How it fails: any of Uncertainties 1, 4, 5, 7 resolving hard against the spine; or the middle growth fork resolving into either pole early.


Best case - abundance without losing the plot#

Not utopia: still politics, still inequality, still incidents. The best coherent world inside the framework is the one where the high-growth branch and the benign joint of uncertainties co-occur.

What has to go right together:

  1. Power and permitting loosen faster than the base case (Uncertainty 2 upside) - behind-the-meter and transmission reform both move.
  2. Robotics cost curves clear local wages in structured then semi-structured work; delivered $/hour, not demos (robotics).
  3. Automated science compounds in instrumented fields (science) without a biosecurity catastrophe dominating the decade.
  4. Apprenticeship is deliberately rebuilt - ladders, not only tools (Uncertainty 3) - so expert formation does not collapse.
  5. Taiwan continuity holds; open-weight lag stays months not years of capability blackout.
  6. No early freeze from a panic architecture after the first incident - rules that still allow verification markets and insurance to mature (Uncertainty 6).

global real GDP growth sustained above ~4% for multi-year stretches; care and trades still expensive but augmented enough that real access rises; verification infrastructure (provenance, underwriting, high-stakes eval) is built early enough that information disorder plateaus rather than compounds; most consumers are materially richer in ways CPI understates.

What "best" does not mean here: perfect equality, no correction, no geopolitics, or solved alignment. A mild sector drawdown can still happen inside this world; Part V row 6 and row 2 can both print.

How it fails into base: robotics data wall holds; permitting stays stuck; organizational redesign stays the J-curve. How it fails into worse: growth arrives with a large early catastrophe or Taiwan shock - high capability, bad path.


Worst case - the bad joints firing together#

Three different "worsts" get conflated in public argument. Separate them.

W1 - Economic and institutional failure (most probable "bad")#

Capex correction becomes a multi-year funding winter (winters); agentic reliability disappoints commercial buyers; measured productivity stays quiet; junior pipelines are cut and never rebuilt; insurance retreats after correlated losses without a clean state backstop.

AI remains a powerful tool layer without a growth-regime change; political fight is over broken ladders and regional energy prices, not abundance; open-weight and closed labs both slow; the apprenticeship gap hardens into a structural expert shortage by the mid-2030s.

W2 - Taiwan / supply-chain break (invalidates the document's arithmetic)#

Leading-edge fab output is removed or gray-zoned for years (Uncertainty 4). This is not "the model was wrong about AI"; it is an external zeroing of compute-gated timelines.

frontier training and accelerator supply become politically allocated; bipolar capacity bifurcation is the base case not a scenario; every Part IV date slips years; robotics and science automation reorder around who holds remaining fabs and the magnet/actuator stack.

W3 - Incident or control tail (low probability, high shape)#

A major AI-attributed catastrophe (Part V row 3) writes a harsh, poorly designed regime - or, further out, loss-of-control dynamics in the row 7 range. Mechanism order the corpus favors for row 3: finance and correlated cyber first; bio middle with fat tail; kinetic least recoverable.

deployment freezes where underwriters and regulators cannot price residual risk; dual-use domains (bio, cyber, weapons) dominate politics; useful civilian diffusion slows for a decade even if models keep improving in labs.

Compound worst: W2 + W3, or W1 + early W3. Those joints are positively correlated through financing stress and geopolitical tension - the corpus already warns that independent multiplication understates joint extremes (Part V correlations).


Other scenarios worth naming#

Acceleration (RSI compresses the calendar)#

Uncertainty 1 fires hard: internal research cycle time collapses while verification and physical governors still bind somewhat. Dates in Part IV move left; the ordering of constraints can survive even when the calendar does not.

Observable signature: lab experiment cycle time and automated paper→result loops compress year-over-year while energy and fab constraints still show up as queues, not magic.

Learned verification wins (spine fails upward for capability)#

Uncertainty 5 retires the master asymmetry. Domain ordering in Part III reshuffles; judgment-heavy work automates faster than the base case; complements story partially survives (atoms, licenses) but gradients change.

Observable signature: blinded eval gap between verifiable and unverifiable domains closes and stays closed (B8).

Complement erosion (spine fails on distribution)#

Uncertainty 7: intelligence + capital manufacture substitutes for land-with-power, proprietary data, or liability capacity faster than assets prices. Consumers win more; the document's distributional predictions miss.

Soft Taiwan / gray zone (base case with a risk premium)#

Not full W2. Freight insurance, license denials, and delivery slippage reprice accelerators without zeroing output. Timelines slip quarters; the spine still "looks right" while every cost curve is worse. Quietly one of the more likely deviations from clean base case.

Correction then maturity (financial winter without tech winter)#

Row 6 fires, revenue later catches up, survivors consolidate to 3–5 frontier labs (Game 1). Looks like W1 for 18–36 months, then rejoins base case with a different ownership map. Base rates for capex cycles make this common in the reference class.


How to tell which world you are in#

SignalPoints toward
AI revenue vs capex; financing mix cash→creditBase vs W1 vs correction-then-maturity (A2/A3)
Entry-level : senior postings; B1 persistenceBase labor transition vs rebuilt ladders (best) vs hardened gap (W1)
Delivered robot $/hour; teleop→autonomyBest / high growth vs base middle vs disappointment (2032–40)
Grid queue, behind-the-meter share, metro power pricesBest-case power release vs energy-political drag
Freight insurance / export licenses / ASP step-ups on leading-edgeGray zone vs W2 (Taiwan)
B8 verifiable vs unverifiable gapBase asymmetry vs U5 spine break
First salient incident → statute quality within 18 monthsBase post-incident architecture vs W3 freeze
Internal research cycle time at frontier labsBase calendar vs RSI acceleration

Full dashboard: Part VII. Scoring discipline: scoring - mechanism over outcome; right-for-wrong-reason is a miss.


What this page is not#

It is not a permission slip to treat the base case as "just one scenario among many." The base case remains the committed forecast; best and worst are stress tests with named selectors. If a future round finds the modal world is no longer the middle growth fork, rewrite 2032–2040 and this page together - do not patch adjectives here while leaving Part IV untouched.

Failure mode of the page itself: scenario sprawl. If a new joint cannot change at least two Part IV claims or one Part V row cluster, it belongs as a paragraph on an uncertainty page, not a new named world here.


Related: 2032–2040 · Part V · Uncertainties · Steelman · Dependencies

View markdown source

select · Enter open · Esc close