This is a sustained collaboration between Brian Sheppard (project owner — direction, scoping, slice-scope choices, trust grading, gating decisions, compute budget, domain reading) and a series of frontier AI coding agents — primarily Anthropic Claude (Sonnet, Opus, Opus 4.7) accessed via GitHub Copilot, 2026-03 onward. The agents executed essentially all of the symbolic derivations, notebook code, sweep modules, HF-Jobs orchestration, and prose drafting; the human direction set the questions, the stop conditions, and the honest-accounting discipline.
It started as a capability probe — “what can a current coding agent do with a speculative GR construction?” — and evolved into a genuine landscape map as the human side built working familiarity with the literature. Wins and losses are shared. See the about page for the full dual-layer framing (research-finding layer + capability-probe layer + cross-portfolio data-scientist methodology through-line) and speculation/epistemological_style_guide.md for the framing rules.
Fuchs et al. 2024 (arXiv:2405.02709) show a constant-velocity Alcubierre shift can be embedded in a Tolman-Oppenheimer-Volkoff matter shell that satisfies all four energy conditions (NEC, WEC, DEC, SEC) in the bubble interior and on the wall.
Anchor point: R₁=10 m, R₂=20 m, β=0.02 c, M = R₂c²/(6G). Reproduced independently in Warp Factory v1.0 at all four ECs = 1.0000.
The shell cannot be arbitrarily thin. Across a 27-cell (C, R₂, β) sweep at 6 inner Δ values each, the smallest DEC-respecting thickness obeys
with κ ≈ 5 typical (median 6, mean 5.3, std 1.0; range [4.0, 6.0] for 15 bracketed cells of 27).
Form universal across the surface; numerical κ is a slice-value, not a constant. Decision-gate-B per Session 19.
ADM 4-momentum conservation forbids self-acceleration of the shell in vacuum (Schuster–Santiago–Visser 2023 Theorem 3). Three escape routes catalogued; only ordinary mass ejection is viable.
Quantitative ceiling under the most favourable Fuchs-compatible parameters (HF-Jobs sweep + Varma 2022 NR rescaling): GW recoil ≲ 0.25% of vwarp. Not enough.
Krasnikov-tube networks: symbolic Einstein tensor reproduces Everett–Roman 1997 Eq. 14 to 14 decimals; universal scaling ρpmin ∝ −η/ε². WEC pass-rate 0.0000 across 300 (η, ε, n) points.
Closes the static-infrastructure-network branch independently of QFT loopholes.
Within the canonical static slice — Alcubierre βxx̂ shift × spherical Fuchs-class shell × asymptotically flat vacuum exterior × steady-state metric or its Lorentz boost × 4D General Relativity — the answer is "a positive-energy bubble exists, but no classical mechanism can accelerate it to useful speeds and no static infrastructure walls can be built."
This is not the closure of the classical question. Six adjacent slices (alternate shifts, hybrid walls, time-dependent acceleration, Krasnikov-2003 QI loopholes, cosmological exterior, modified gravity) are mapped in Phase 2C and all return null; the unenumerated landscape is logged with reopening criteria as Phase 2E. The three published positive-energy candidate constructions have all now been adjudicated in-pipeline (Sessions 42–46): Fell–Heisenberg's strict-pass is a box artifact, Lentz fails the full energy conditions at class level, and Garattini–Zatrimaylov reproduces exactly but is a comoving vacuum void with zero transport content.
The audit's one reversal converted into a real lever: a two-component configuration drops the certified canonical floor 13.3% to Mmin = 2.226 × 10²⁷ kg (RES_CONF-certified; −17.8% at the R₂=15 cell). A cross-builder discriminator then corrected the mechanism: at identical density, the per-shell pressure ansatz (each component with its own P=0 surface) sits 12% below the single-shell floor while a contiguous graded wall sits 14% above it — the lever is the two-body pressure structure, not density grading. The continuous graded-wall family is closed negative (0/80 members below the floor).
Kill-tests: the mask-blind-spot hypothesis refuted exactly; every f=0 baseline reproduces the Session-32 map bit-for-bit.
Time dependence (2E.1): rigid-profile spin-up is an exact spacetime whose Einstein tensor carries no v̈ — the whole inflate-coast-deflate lifecycle reduces to a margin surface, and it is EC-clean for any spin-up slower than ~one light-crossing of the wall (τ* ≈ 0.4–0.8 R₂/c). The static-slice restriction was not load-bearing for this family.
Topology + joint vorticity (2E.4): closed in full — the exact far-field coefficient −V√(σπ) is parameter-independent, so compact quotients admit no member of the FH family and interior surgeries can't touch the exterior violation; the joint vortical+Π exponent space is negative at 180/180 dual-box cells.
Modified gravity (2E.2): closed negative — see the theorem below.
Within 4D semiclassical GR the project has no known candidate for a vacuum + DEC + dynamical warp realisation — and the f(R) escape hatch is now shut with a theorem-shaped result: Jordan-frame NEC is linear in (f′, f″, f‴) at each value of R, and on the Alcubierre wall 23.4% of sampled points individually admit no coefficient triple at all — no f(R) with a ghost-free graviton, of any functional form, yields NEC-respecting Jordan-frame matter there (161-point α-map uniformly negative alongside). Phase 2E stands fully dispositioned: 2E.1 positive-within-slice, 2E.2 and 2E.4 closed negative, 2E.3/2E.5 externally gated. None of the computationally accessible relaxations opens a useful-warp loophole.
Sessions 53–55 then retired the last optional residues: the two-body standoff gap is a plateau (floor unchanged); the Session-27 oblate "+3.09% improvement" — the last number carried on the demoted thin-slab convention — was reversed (a sign error in the lost scratch driver; spherical local optimality now holds in both deformation axes); and the Lentz ℓ¹ kink sheets were characterized as finite, strictly EC-violating surface layers — a third independent failure of that class. The board is now entirely externally-gated items awaiting publications, plus the deep-dynamical legs (real ADM evolution, superluminal transitions) deferred with explicit criteria.
Closed: 0 (Foundation), 1 (Linearization), 2A (Matter-shell, slice-scoped), 2B (Casimir route — closed negative as a physical mechanism, Session 34), 2C (six adjacent slices, all null), 2D (Fell–Heisenberg follow-up — strict-pass re-scoped as an evaluation-box artifact, Session 42), 3 (numerical verification through the certified minimal-mass maps: single-shell Session 32, two-body re-base Sessions 47–48), 2E (fully dispositioned, Sessions 49–52 — 2E.1 first leg positive-within-slice, 2E.2 and 2E.4 closed negative, 2E.3/2E.5 externally gated).
Open: externally-gated 2E items (quantum sources; higher-D Krasnikov — awaiting publications) · deferred 2E.1 dynamical legs (ADM evolution; superluminal transitions). Not started: 4, 5 (informal coverage exists via 2A.10 / 2D.7 / 2D.9 and the Session-46 ANEC probe).
Trust grades are maintained as dated addenda, including downgrades and withdrawals. Recent moves: the Session-47 "graded wall" reading of the new mass floor corrected in place one session later (numbers stood; mechanism reattributed); the Session-27 oblate +3.09% reversed as a sign error after four reconstructions of its lost scratch driver — the reproducibility complaint the project levels at published papers, turned on its own record (Session 54); and a Session-55 a-priori hypothesis (dipole-layer kink sheets) refuted by its own instrument and recorded as such.
Corrections are recorded, not erased: the Session-36 frame-projection bug, the Session-42 box-artifact re-scoping, the Session-48 mechanism correction, and the Session-54 sign-error reversal are all documented where they happened.
Full table: TRUST_AUDIT.md.
The trusted EC oracle for sharp shell profiles is the exact-symbolic radial evaluator (warp_factory_py/solvers/axisymmetric_ec.py), certified against a closed-form ground truth — now extended with an exact time-dependent spin-up evaluator (Session 49) and a Jordan-frame f(R) matter evaluator with a code-generated exact-cancelled correction tensor (Session 51); Cartesian FD is demoted to a smooth-only cross-check.
25+ parameter sweeps dispatched (full grids moved from HF Jobs to a free LAN compute node in Session 47 — 100-point certified maps in ~10–40 min); 25+ tracked verification harnesses in verification/ — validation suites, ground-truth adjudicators, adversarial kill-test batteries, and the Session-52 designer-f feasibility LPs.
Outputs: bshepp/alcubierre-sweeps.
Phase 2A · the κβ/C scaling, the acceleration obstruction, the Krasnikov no-go. Anchor result of the project.
Phase 2C · six independent assumption-relaxations of the static-slice no-go. All six return null. Side-by-side.
Phase 2D · multi-mode static shift, 1404 strict-pass points, then dismantled by four structural tests.
Phase 3 · independent MATLAB reproduction of the Fuchs anchor and the κ-surface across 162 metric builds.
Phase tracker: 0 through 5, with status, decision gates, and reopening criteria for every closed item.
13 Jupyter notebooks, the MATLAB Warp Factory reproduction kit, and the HF-Jobs sweep dispatcher.