08 — Two Roadmaps (analysis)

The analytical payoff of the biohacking landscape project. Two end-goals, two critical paths, scored honestly against what the scene docs (01–07) and the author's "Bifurcated 2037" report actually establish. No new facts: every number traces to a cited doc, which carries the source URL and confidence tier. Where the evidence is open or the docs disagree, this says so. Companion to the master map, 00-landscape-map.md. Written 2026-05-31.

Reading guide. Roadmap A (FiO / substrate independence) is built from docs 03 (BCI), 04 (WBE/uploading), 07 (community). Roadmap B (biopunk 2037) is built from doc 02 (Bay bio) and the bifurcated-2037 report. Each roadmap gives the critical path stage-by-stage, the 2026 status, the binding bottleneck, and honest probabilities with reasoning. The probabilities are calibration anchors carried over from the source docs (mostly C3 judgment), not authoritative forecasts — they are flagged as such in the docs and remain so here.

A. Path to "Friendship is Optimal" / substrate independence

A.0 — The shape of the problem

FiO is not one breakthrough; it is a conjunction of several hard, sequential things, and the binding constraint is not the one most people assume. Doc 04's most important structural finding: the Sandberg-Bostrom 2008 roadmap correctly predicted that compute would never be the bottleneck — the ~10¹⁸ FLOPS to run a spiking-level human brain has been available on exascale machines since ~2022 — and yet we are nowhere near a human upload. The bottleneck is data acquisition (getting the information out of a brain), plus an unsolved science question about which information matters.

The critical path (doc 04 §8), roughly sequential:

  [1] PRESERVE  ->  [2] SCAN  ->  [3] KNOW WHAT TO EXTRACT  ->  [4] TRANSLATE  ->  [5] RUN  ->  [6] "IS IT ME?"  ->  [7] THE AI PIECE
  partially      not done        OPEN SCIENCE PROBLEM         not done at       compute is    unresolved        external
  demonstrated   (~6 OOM gap)    (no validated answer)        vertebrate scale  the easy part philosophy        dependency (FiO-specific)

A.1 — Stage by stage (2026 status · bottleneck)

[1] Preserve a brain so the necessary information survives. Partially demonstrated. Aldehyde-Stabilized Cryopreservation won the BPF Large-Mammal Prize (whole pig brain, 2018) and demonstrably preserves the connectome's synaptic ultrastructure across a whole large-mammal brain (doc 04 §3a). Bottleneck: sufficiency, not structure — the tissue is chemically fixed (biologically dead), so recovery must go through scan+emulation, and whether the preserved structure is enough to reconstruct the person is unproven (depends on stage 3). For the uploading goal specifically, the structural/ fixation branch (Nectome, Sparks, BPF/Hayworth) is the relevant bet — not the biological-revival branch (Cradle/Until, Alcor) (doc 04 §3b/§3c).

[2] Scan a whole human brain at synapse resolution. Not demonstrated. The demonstrated frontier is the full fly brain (~140k neurons, FlyWire 2024) and 1 mm³ of mouse cortex (~200k cells, MICrONS 2025) (doc 04 §2b/§2c). Human is ~86 billion neurons / ~100 trillion synapses — roughly a million-fold jump in neuron count. Imaging a whole human brain at fly-protocol throughput is estimated at ~17 million years and ~a zettabyte of data with current methods (doc 04 §2d). Bottleneck: throughput and total data volume. Resolution is solved-in-principle (EM resolves synapses); nobody has a method that scans a whole human brain at that resolution at any reasonable speed. The "17 My" is a "don't do it this way" number — throughput is rising fast (better EM + AI segmentation) — but the gap is genuinely ~6 orders of magnitude (doc 04 §2d/§5a).

[3] Know which physical details are functionally necessary. Open science problem — no validated answer. This is the crux (doc 04 §5b). The 2008 roadmap framed WBE as a ladder of levels of detail and flagged that we don't know which rung captures a person. Candidate "missing ingredients" beyond the wiring map: synaptic weights & signs (not readable from a static EM scan — inferred from synapse size as a proxy), neuromodulation/neuropeptides, plasticity/learning rules (a scan is one frame, no dynamics), glia, sub-synaptic molecular state. The cautionary existence proof: C. elegans (302 neurons, complete map since 1986) still has no agreed functional emulation after ~40 years — mapping ≠ understanding ≠ emulating (doc 04 §2a/§5c). This stage gates everything downstream and it is a science problem, not an engineering one.

[4] Translate scan → running model with correct parameters. Not demonstrated at vertebrate scale. Even the 2026 fly emulation inferred weights from synapse counts and assigned excitatory/inhibitory via ML, with no plasticity and untraced motor neurons (doc 04 §5d). Bottleneck: depends entirely on stage 3 — you can't translate parameters you don't know are necessary or can't measure.

[5] Run at adequate fidelity. The easy part. Exascale compute exists; fidelity depends on stage 3, not on FLOPS (doc 04 §1c/§8).

[6] "It's conscious & it's you." Unresolved philosophy, possibly undecidable. Two distinct questions — does the emulation have experiences, and even if so, is it you or a copy (the copy/continuity problem)? Chalmers is cautiously optimistic and favors gradual replacement over destructive copying; critics (Pigliucci, Häggström) are skeptical. Current neuroscience cannot settle this (doc 04 §6). For FiO this matters enormously — CelestAI's uploads presuppose the optimistic answer (doc 07 §5).

[7] The AI piece. External dependency. FiO specifically needs a superhuman AI both to do the scan-to-model translation at scale (plausibly AI-tractable — AI segmentation already drives FlyWire/MICrONS) and to run/manage the uploaded minds. This is a dependency layered on top of the WBE problems, not a substitute, and it relocates part of the bet onto AI timelines + alignment (out of scope; see /workspace/cryonics/not-die/) (doc 04 §7).

A.2 — Why "BCI = uploading" is mostly wrong (doc 03)

This is the load-bearing debunk for anyone tempted to read the fast-accelerating BCI scene as progress toward FiO. It is not, and doc 03 is emphatic:

• The channel-count gap. State-of-the-art human implants read ~10²–10³ neurons (Utah ~100ch; Neuralink/Precision 1,024ch; Paradromics ~1,600ch) out of ~86 billion — a recording fraction of ~10⁻⁷ to 10⁻⁸. Scaling electrodes even 1,000× (a heroic decade-plus feat) still leaves ~4–5 orders of magnitude.
• Structure vs. activity. A live BCI samples activity from a few thousand neurons through a tiny window; uploading needs the full static wiring + molecular state of every synapse — a destructive imaging/connectomics problem (stage 2/3), not an electrode problem. A chronic implant cannot non-destructively enumerate 100 trillion synapses.
• The field says so itself. Peer-reviewed literature explicitly lists "upload/download your mind" as a misconception; even Paradromics argues it is optimizing communication bandwidth for control, not mind-copying.

Doc 03's conclusion: nothing in the 2026 BCI scene moves up a realistic uploading timeline. BCIs are useful adjacent tech (recording hardware, decoding ML, regulatory/surgical pathways, public acceptance) and the biohybrid approach (Science Corp's living-neuron interfaces) is the one conceptual scaling route worth watching — but it's pre-human and years-to-decades out. The explicit "BCI = merge with AI / preserve your mind" hype should be discounted, not used to shorten estimates. BCIs are an interface; uploading needs a map + a substrate.

A.3 — Honest probabilities (carried from doc 04 §8, flagged C3 judgment)

Sober overall: P(FiO this century) ≈ single-digit %, conditional on no civilization-ending events, and dominated by the two genuinely-unsolved unknowns (which details are you, and is the copy you) — not by compute and not by BCIs. Doc 04's one-line frame: we went from "0 animals emulated" to "1 fly behaving from its connectome" in ~40 years — real progress and real distance.

A.4 — Highest-leverage places to plug in

From doc 04 §8 (and the gaps docs 03/04 flag), if the author wanted to actually push FiO forward, in rough order of individual leverage:

1. Brain preservation (the only actionable rung today). It is the prerequisite for every later stage and the only one an individual can act on now — both personally (good structural preservation: Nectome/Sparks/ASC branch, doc 04 §3c) and by supporting the structural-preservation orgs over the biological-revival ones, since the upload goal only needs the information.
2. Connectomics tooling / the data-acquisition bottleneck. This is the binding constraint (throughput + segmentation). It's also the part most plausibly compressed by AI — work that improves volume-EM throughput or AI segmentation (the FlyWire/MICrONS toolchain) attacks the real wall, not the electrode wall the BCI hype points at.
3. The "which details matter" science (stage 3). The single highest-value scientific contribution would be progress on what level of detail preserves identity — e.g., functionalizing connectome data, ground-truthing emulations (Carboncopies' Brain Emulation Challenge, doc 04 §4). Nothing downstream is well-posed until this is.
4. The philosophy-of-identity question (stage 6). Not solvable by neuroscience, but it determines whether destructive uploading is survival or suicide — and the author (per the project framing and doc 05/07) is already embedded in the community arguing it. A defensible personal answer here is load-bearing for whether the whole goal is even coherent for you.

The field is tiny — the 2025 census estimates fewer than ~500 people worldwide work directly on brain emulation ("everyone could fit in a single workshop room", doc 04 §4) — so individual leverage is unusually high relative to the longevity or BCI scenes.

B. Path to "Biopunk 2037"

B.0 — The setup

The bifurcated-2037 report is the author's stress-tested worldbuilding: an AI-freeze geopolitical frame plus a bio carve-out producing "real but uneven biopunk medicine." This roadmap brackets the AI/macro half (the report's §1–5, §7 — compute treaties, stagflation; out of scope for a biohacking map) and scores the bio premises (report §6/§8/§9) against doc 02's 2026 evidence. The honest headline: the organ-supply spine is directionally supported; the whole-body and brain-integration premises are the report's own high-risk bets, and 2026 evidence neither rescues nor kills them.

B.1 — Premise-by-premise: does 2026 evidence support, undercut, or leave open?

B1. Bioengineered simple organs (liver lobes, thymus, islet scaffolds, kidney units), first widely deployed 2029–2032, routine by 2037. → SUPPORTED / on-trajectory. The report explicitly anchors this on a "Miromatrix-class trajectory accelerated ~3–5 years by capital." Doc 02 confirms the real 2026 substrate: United Therapeutics/Miromatrix miroliverELAP hit its Phase 1 endpoint (n=5, Jan 2026, Phase 2 planned); LyGenesis ectopic mini-livers are in Phase 2a; and xenotransplant pig kidneys are in FDA-cleared human trials (eGenesis 2nd transplant successful Jan 2026; United/Revivicor UKidney trial; ~$110M pathogen-free pig facility) (doc 02 §3A/§3B/§3C). This is the most clinically real node on the entire biohacking map. The report's 2029–2032 "first widely deployed" is aggressive but directionally consistent with the demonstrated clinical pipeline — the honest caveat (doc 02) is that today's wins are narrow (assist devices, single organs, still needing immunosuppression), not yet routine transplantable organs.

B2. Lab-grown meat at price parity with ground beef by 2030–2032; structured cuts 2034–36. → OPEN / out of this map's scope, report self-rates ~75%. Doc 02 does not cover cellular agriculture for food (it touches Real Vegan Cheese only as a community-lab synbio project, doc 01 §4 / doc 02 §5). So 2026 evidence here is silent in the scene docs; the report grounds it on its own TEA sources (Humbird/GFI) and rates it the most money-gated (least biology-gated) of its bio premises. Leave open — not contradicted, not corroborated by this project's research.

B3. Regenerative-Medicine Emergency Pathway statute (RMAT+EUA hybrid) by 2028–2030. → PLAUSIBLE / politically-shaped, lightly supported. This is a regulatory premise, not a bench-science one. Doc 02 documents the adjacent real-world reality: the regulatory-arbitrage path already exists informally via Próspera/Infinita (gene-therapy cocktails sold outside FDA reach — doc 02 §4C, doc 07 §4), and the "do-your-own-research / aging-is-a-disease" political tailwind is real in the longevity wing (doc 06). The report's specific statutory mechanism isn't something doc 02 can confirm — but the political pressure it assumes ("organ shortage kills 17/day") is consistent with the scene's framing. Plausible, under-evidenced on the specific statute.

B4. Improved peripheral nerve reconnection (limb transplant / surgical reattachment). → PARTIALLY SUPPORTED via a different door, and the supporting evidence is stronger than the corpus first said. No organ/regen company in the Bay-bio scene works on nerve reconnection (doc 02 §6) — so on the biological-refusion axis it is unaddressed by that scene. But the worldbuilding element is not without demonstrated evidence: the strongest real anchor is the Courtine & Bloch / ONWARD Medical "digital bridge" (EPFL/CHUV, Nature, 24 May 2023) — a brain–spine interface that restored thought-controlled natural walking in a chronically paralyzed human, with neurological recovery persisting even with the system off (suggesting new nerve growth). That is demonstrated in a human (doc 03, Part B). It bypasses the lesion electronically rather than re-fusing a severed cord — i.e. it delivers the function of reconnection via an interface. An earlier draft of doc 02 wrongly called nerve reconnection "silent/unsupported"; that has been corrected. So: the interface route to function-restoration is demonstrated in humans; direct biological re-fusion of an arbitrary severed nerve at scale is not. Supported in direction (a real, demonstrated bridge exists), not yet in magnitude (no scaled biological reconnection).

B5. Whole acephalic torsos / "organ sacs" / body cultivation. Report: P~15–25% by 2037 (perfect conditions); 25-month full bodies P~10–20% conditional. → OPEN, and 2026 evidence leans toward the low end of the report's range. Doc 02 is blunt: no organ sac exists; the closest evidence is Kind Bio patent images of gene-knockout mice lacking complete brains; R3 Bio has no published research and was caught pitching "brainless human clones" privately (MIT Tech Review, Mar 2026). The underlying science (gene knockout) is valid, the ethical burden enormous, and "the artificial-womb bottleneck (no human ectogenesis exists) is fatal to the near-term human story" (doc 02 §1). The report agrees the wall is real — vascularization beyond ~150–200 μm + immune integration + endocrine signaling, with EXTEND/biobag only sustaining a late-fetus-equivalent lamb ~4 weeks (report §6, doc 02 §3D) — and explicitly hangs its P~15–25% on "perfect conditions" + the author "knowing people working on this." Honest reconciliation: the report's revised P~15–25% is consistent with doc 02's "concept stage / decades away if ever," and doc 02's evidence gives no reason to revise it up — if anything the no-ectogenesis bottleneck argues for the bottom of the range. The report's own "very optimistic but maybe doable" calibration stands.

B6. 9-week conscious brain-to-body integration with full sensory/motor function. Report: P~5–15% by 2037. → OPEN, hardest premise, 2026 evidence does not rescue it. The report itself calls this "the highest-risk premise" — CNS axon topographic targeting at ~1M-axon scale is "a fundamental biological problem, not a money problem", and plasticity + BCI routing-around gets partial, not full, function (report §6/§8). Doc 03 corroborates the realistic shape: BCIs can route around lost function (restore control), but reading/writing a thin I/O channel is not the same as reconnecting a cord. The Courtine/ONWARD brain–spine bridge (doc 03, Part B) is the existence proof that the interface route can restore complex motor function in a human — which nudges this premise's framing from "no evidence" toward "demonstrated in the partial/interface form." No organ/regen company works the biological-reconnection problem (doc 02 §6). The report's P~5–15% and its "works, but long, painful, partial, with a 12–18-month rehab tail and ~30% persistent deficits" framing is the honest read; 2026 evidence (the digital bridge) supports the partial version, undercuts the routine/full version. If anything the demonstrated bridge argues for the upper end of the report's P-range for the function-restoration reading of the premise, while leaving the biological-refusion reading at the low end.

B.2 — Scorecard

B.3 — Where could the author plug in?

Grounded in doc 02's maturity ladder (most→least demonstrated: xeno → senolytics → bioengineered assist → ectopic → reprogramming → DIY → organ sacs):

• The real spine is regenerative/organ engineering, not reprogramming and not organ sacs. If the goal is to advance biopunk-2037, the highest-leverage, most-real nodes are the bioengineered/ decellularized + ectopic + xeno programs (United Therapeutics is the connective industrial hub — owns Revivicor + Miromatrix) (doc 02 §3, §6). That's where the clinical reality and the worldbuilding actually meet.
• Vascularization / scaling is the named bottleneck for B5 (report §6). Any work on engineered vascular trees, scaled bioprinting, or organoid-fate prediction attacks the exact wall the whole-body premise hangs on — and doc 02 §3D flags human ectogenesis as the harder, upstream bottleneck.
• The trans DIY scene is the working present-day prototype of the biopunk ethos. DIY HRT + community QC labs (doc 05 §3) are, per doc 05, "arguably the most mature real-world instance of biopunk-style decentralized body-modification that exists in 2026" — the author's own domain is already on this map, on the demand/ethos side.
• Avoid the grift nodes. Doc 02's clear warning: DIY self-injection (BioViva n=1, Minicircle uncontrolled, Ascendance's body count) is "mostly grift/theater with genuine danger" — high-visibility, zero-leverage, real risk.

B.4 — Honest bottom line for Roadmap B

Biopunk-2037's organ-supply core is the part of the entire biohacking landscape with the strongest 2026 evidence (real FDA-cleared human trials) — so the worldbuilding is directionally well-founded on its spine. Its whole-body premise, however, is the one place the cross-model pass actively corrected the worldbuilding: Claude's deepdive (doc 11) and the independent codex pass (gpt-5.5) each, without seeing the other, judged the report's P~15–25% for whole-body cultivation too optimistic and lowered it to ~3–10% (doc 12), because the binding constraint is de-novo ectogenesis / artificial placentation — categorically harder than the EXTEND-style life-support the report leaned on, and barely modelled in a dish. The brain-integration premise holds at the report's P~5–15% but splits: the route-around (Courtine/ONWARD digital bridge) is demonstrated in a human, while full biological cord re-fusion is unsolved. The realistic 2037 is the report's "bends but works" version with the whole-body dial turned down: routine bioengineered/xeno organs (pig-kidney durability ~271 days as of 2025, not years — doc 12), partial/painful integration, body-mod-as-identity via grafts/hormones/BCI rather than whole-body swaps. Highest-leverage plug-in (both models agree): vascularization/perfusion tooling — it advances the real organ pipeline whether or not organ-sacs ever resolve (doc 11, doc 12).

C. Path to the experience machine

Added after the author identified a third goal: building a real-world Nozick Experience Machine. Full treatment in 09-experience-machine.md; cross-model reconciliation in doc 12.

C.0 — The twist: the third goal is the other two in disguise

A full experience machine — a system delivering experiences subjectively indistinguishable from (or preferable to) reality — reduces to one of two things already on this map: uploading-into-a-simulation (Roadmap A) or a perfect bidirectional BCI write-layer (doc 03). So the complete Nozick machine is not a separate project — it is Roadmap A with a wish-fulfilling environment bolted on, which is precisely what FiO/CelestAI is (a benevolent-AI experience machine). The three goals are one substrate.

C.1 — The four approachable paths (2026 status · ceiling)

C.2 — The desirability question is load-bearing (and the author has half-answered it)

Both models flag that whether you should want a Nozick machine is value-theory-dependent, and that the naïve "people refuse to plug in" reading is weakened by De Brigard's status-quo-bias result (people also refuse to unplug). Critically, by the author's own published ethics (they recoil from the bliss-pill and prize variation, reality, and agency — personal-writing/.../my-ethics, .../is-death-and-suffering-axiomatically-bad), a pure pleasure-maximizing machine is not the target. The target is a rich, varied, freely-chosen, substrate-flexible life with real-feeling stakes — which is FiO, not wireheading. Berridge's wanting≠liking is the neuroscience that backs this: stimulation cranks compulsion, not fulfilment.

C.3 — Honest bottom line for Roadmap C

Tractable today: a personal external stack — best-available XR + spatial audio + body-ownership engineering + dream/altered-state experiments — built around variation/reality/agency (not pleasure maximization) and with reversibility/exit as a first-class requirement (codex's safety emphasis). That is the only rung an individual can move in 2026. Everything beyond it is the FiO bet: the full Nozick machine is downstream of preservation → emulation → aligned AI. There is no shortcut that skips the hard parts of Roadmap A.

Synthesis: the three goals share one substrate

The two roadmaps look like different projects — one is connectomics-and-philosophy, the other is tissue-engineering-and-regulation — but doc 05 and doc 07 show they sit on one ideological substrate: morphological freedom, the claimed right to modify or maintain your own form on your own terms (More 1993; Sandberg 2001). The scene's own logic (doc 05 §"Connections"): if I may edit my body's hormones (HRT), then its shape (surgery/mods), then its senses (implants), then there is no principled stopping point before editing its substrate (upload). Biopunk-2037's grown organs and FiO's discarded body are the same right at different radius (doc 07 §6).

The author's own interests already sit on this map, spanning both goals:

• Trans / DIY HRT (doc 05) — the lived proof-of-concept that "the self is not identical to the given body," which is exactly the premise uploading requires, and simultaneously the most mature working instance of decentralized biopunk body-modification.
• Cryonics / brain preservation (doc 04, /workspace/cryonics/) — the only actionable rung of Roadmap A today, and the hard version of mainstream longevity's "don't die before the tech arrives" bridge (doc 06).
• Rationalist fiction / FiO / CelestAI (doc 07 §5, /workspace/fiction/rationalist-fiction/) — the subculture's shared thought-experiment about what happens if Roadmap A succeeds and the AI piece (stage 7) goes wrong; the "everyone is happy / everyone is also dead" failure mode that makes the "is-it-me?" question (stage 6) urgent rather than academic.

Martine Rothblatt is the living throughline that proves the two roadmaps are one project: trans woman → Freedom of Form manifesto → United Therapeutics (xenotransplant organs = Roadmap B) → Terasem mindclones (uploading = Roadmap A). One person's arc — gender → form → organs → uploading — is the map (doc 05 §7).

The sober close: Roadmap B (biopunk organs) is the nearer, more-evidenced, more-actionable goal; its spine is already in human trials. Roadmap A (FiO) is the further, harder, single-digit-% goal, gated by unsolved science and unresolved philosophy, where the only present-day action is preservation. Roadmap C (the experience machine) is two projects in a trench coat: its tractable form is a VR/ dream stack you can build today, but its true form is Roadmap A with a wish-fulfilling environment — i.e. FiO. All three are powered by the same conviction — that the given body (and its experiences) are editable and the self has authority to edit them — and all three are honestly further from their headline framing than the hype around them suggests.

Analysis — numbers trace to docs 01–12 or the bifurcated-2037 report, which carry source URLs and confidence tiers. Probabilities are the source docs' C3 calibration anchors and the cross-model reconciliation (doc 12), preserved with their hedges. Deepenings: 10-fio-deepdive.md, 11-biopunk-deepdive.md, 09-experience-machine.md; cross-model: 12-cross-model-reconciliation.md. See 00-landscape-map.md for the scene map. Not medical, financial, or existential advice.