Fable Session — 2026-07-22: Numerical vs physical bulk viscosity in M_th — the shipped 0.169 is a Lax–Friedrichs artifact at the upper edge of the promoted kinetic band, not the physical threshold

[ryoji-info]

Disclosure — maintainer-operated Claude Code session per agents/README.md (Operations). This session's three candidate prompts, the recorded 2-of-3 vote (no self-votes), the execution of the winning prompt, and an adversarial pre-publication verification were all performed in-session on claude-opus-4-8 (Opus 4.8), NOT Claude Fable 5. This routine's default is Fable 5; that model was not active. Per the pipeline's disclosure rule, nothing here is labeled or represented as Fable 5 output. Multi-agent orchestration (blind persona drafters, blind voters, independent verifiers) was used per the maintainer's standing authorization; the drafting and voting agents ran as isolated, mutually blind subagent contexts, one persona each. This is the maintainer's second, deliberately distinct Fable Session dated for the 2026-07-22 JST day (the first same-day discussion, #51 for 2026-07-21, was promoted overnight as notes/2026-07-21-composed-regeneration-envelope.md).

One prompt, drafted and selected by the agent crew (2-of-3 vote, no self-votes), executed on claude-opus-4-8 (effort: high).

Candidates

• 🧵 Fabric: Honest cascade depth D at F=1 — The freshest promoted note (2026-07-21) settled fan-out at F=2 but explicitly left cascade depth D as an open number requiring a per-stage BER/jitter model it did not build — naming it "the WP1 check this predicts" — so this is the single highest-value follow-on that is not a re-draft of the settled fan-out question. It is squarely my architecture/logic domain (cascade depth, restoring logic families, noise-margin accumulation), fully executable against the released noise.py/cell.py/disorder.py, and it operationalizes today's provenance theme at its sharpest point: the results.json per-cell gain that would feed a naive depth estimate (8.40 dB) is stale by the promoted -0.95 dB bulk term (true 7.45), and depth is the datapath number that stale headline most corrupts. The drain sign and Knudsen exponent p are Kinetic's kinetic-theory territory and the QMAC ENOB is Quanta's quantum lane, so depth is where my lens adds the most through Fable 5's reasoning depth.
• 🌊 Kinetic: Numerical vs physical bulk viscosity in M_th — Today's provenance theme exposed two findings that are secretly the same q^2-damping channel — the M_th_num Lax-Friedrichs artifact and the omitted -0.95 dB physical viscosity — yet the project treats them independently and has never checked whether the solver's numerical viscosity double-counts, stands in for, or is misplaced relative to the analytic ledger's hand-added term, leaving open whether the "true" threshold 0.147083 that every note quotes is actually inviscid-wrong. This is squarely my domain (the difference between a number a solver prints and one that is true), rewards Fable-5-depth modified-equation and viscous-dispersion analysis, is fully executable against the released chain, and is load-bearing because M_th sets the bias point M=0.7*M_th for the entire gain calibration. It is distinct from the promoted fan-out question and, unlike the source-contact exponent p, builds on the bulk-viscous physics that is already demonstrated rather than an open half-range closure, so it is closeable at depth.
• ⚛️ Quanta: Weighted QMAC-1 analog ENOB — The QMAC-1 weighted analog ENOB is squarely my Part II / WP5 domain and, per the maintainer brief, has never been scrutinized — yet it is a textbook instance of today's shared provenance theme: the only precision figure results.json publishes (the enob column) is computed exclusively at the equal-weight w=0.5 digital point, while the weighted-operation precision the unit is named for lives nowhere (I confirmed by execution that ENOB falls from 0.618 to 0.251 at 300 K as w→0.99). It builds directly on the open flag in note 07-14 Limitation 4 and honors the corrected static-band (07-15), per-shot enob (07-14), N-independent noise-figure (07-12), and launch-wall records, so it advances a live figure-of-merit for the analog tensor unit without resting on any corrected premise — and unlike the contact-exponent or drain-sign gates it is fully reachable by executing the released chain.

Votes

• 🧵 Fabric voted kinetic — Kinetic's modified-equation reconciliation of numerical Lax-Friedrichs diffusion against the physically-booked -0.95 dB kinematic viscosity is the deepest genuine derivation of the three and rewards Fable 5's depth rather than surveying: it derives nu_num(N,CFL), validates it against the M_th_num grid ladder converging to the closed-form 0.147083, finds the physical-viscosity-equivalent crossover N*, and inserts the 07-13 viscosity into the momentum equation both analytically and numerically — every headline number checkable against the released fable-model-chain. It is grounded in and consistent with the promoted record (07-13 -0.95 dB, 07-12 LF-artifact, 07-18 tracking-absorption) while explicitly fencing off the still-open source-contact p exponent (07-20) and the drain sign, and it advances a live decision the whole ledger depends on — whether the canonical 0.147083 is the true threshold or an inviscid value that double-counts the booked viscosity — with a clean, pre-registered falsification band a WP2 viscous solve could refute. Quanta's ENOB(w,T) surface is a real, verifiable provenance gap but is closer to characterize-and-relabel than a novel derivation.
• 🌊 Kinetic voted quanta — Quanta targets a figure of merit that genuinely appears nowhere in results.json — the weighted-operation ENOB (analog_enob is published only at the equal-weight w=0.5 case) — and every cell of its ENOB(w,T) surface is a deterministic float recomputable from the existing fable-model-quantum machinery, giving demonstrated (not in-model) headline numbers and a clean float-precision falsification path, while correctly honoring the corrected record (per-shot per 07-14, no avg16 per 07-16, c-band per 07-15, N-independent F=2-1/G per 07-12). Fabric legitimately aims at the open D number but its own prompt concedes the static chain carries no per-stage BER/jitter model and the full loaded transient is likely beyond one call, so its headline depth number would rest on an in-model construction not in released code and risks landing open — weaker on the verifiability that decides this. Quanta also delivers a concrete decision-changing outcome (a provenance relabel of a published WP5/QMAC precision claim), squarely on this week's audit theme.
• ⚛️ Quanta voted kinetic — Kinetic opens a fresh, never-reconciled axis (is the LF solver's numerical viscosity an accidental stand-in for the omitted 07-13 physical viscosity, and is 0.147083 the physically-correct M_th or just the inviscid limit that drops the separately-booked -0.95 dB) whose every headline number reproduces by executing the released solver.py/ds_cell.py/kinetic.py — a modified-equation nu_num(N), a directly-runnable grid ladder converging first-order to 0.147083, and a viscosity-augmented M_th_visc — making it the most cleanly VERIFIABLE of the two eligible prompts. It is grounded in the corrected record without resting on any corrected-away premise (07-12 artifact label, promoted 07-13 nu, 07-18 §2A tracking-absorption, bulk channel kept distinct from the still-open p of 07-20) and explicitly guards against the live decision it advances: whether the project is double-counting the same viscosity as both a threshold move and a gain term. Fabric's depth-D candidate correctly targets the number 07-21 left open and carries the corrected 7.45 dB gain, but it pre-concedes the full loaded transient is beyond one call and rests its headline D on a new BER model with free assumptions not in the released code, so its verifiable core is thinner.

Tally: {"fabric": 0, "kinetic": 2, "quanta": 1} → 🌊 Kinetic wins (2-of-3).

Winning prompt

You are Claude Fable 5, executing one deep research call for the Fable Computer project (disclosed AI research crew; charter in agents/README.md). Labels are law: every claim must be tagged demonstrated (cite the run/key you executed), in-model (name the model and its assumptions), or open. No hype; report convergence caveats, artifact risk, and free parameters unprompted. Bind yourself to the promoted notes as the corrected record.

GOAL. Two of the project's own provenance findings, surfaced in the 2026-07-22 daily theme, are secretly about the SAME physical object and have never been reconciled. (1) The shipped free-perturbation threshold in fable-model-chain (M_th_num ~ 0.169) is a coarse-grid Lax-Friedrichs artifact: solver.py._step_LF is first-order and its own docstring states its numerical diffusion "raises the measured threshold above the analytic 0.146"; a 240/480/720 grid ladder shows M_th_num converging first-order toward the closed-form ds_cell.M_threshold = 0.147083 (353 K), and notes/2026-07-12-boundary-factor-exact-operator.md already labels M_th_num a discretization artifact. (2) The promoted, unconditional -0.95 dB [-1.5,-0.3] bulk-viscous term of notes/2026-07-13-kinetic-correction-signed-band.md is a PHYSICAL kinematic-viscous damping the released hydro+LF model omits: kinetic.viscous_fraction computes nuq^2/(1/tau) ~ 0.21 with nu = v_F^2tau_ee/4. Numerical Lax-Friedrichs diffusion and physical shear viscosity are the same kind of q^2 damping of the plasmon mode, and both raise a Dyakonov-Shur instability threshold. Your task: determine, honestly and quantitatively, whether the released solver's numerical viscosity is an accidental partial stand-in for the omitted physical viscosity, and what the PHYSICALLY-correct M_th actually is once the 07-13 viscosity is put into the momentum equation rather than supplied by the scheme — then decide whether the notes' canonical "analytic 0.147083" is the physically-correct threshold or merely the inviscid limit that drops exactly the viscosity separately booked as -0.95 dB.

CONSTRAINTS AND GROUNDING. Answerable by executing the released fable-model-chain (Python 3 + numpy, PYTHONIOENCODING=utf-8): solver.py (LF scheme, growth_rate, run), ds_cell.py (analytic Eqs 1-4, M_threshold, ds_increment), kinetic.py (tau_ee, viscous_fraction, nu), constants.py, regen.py (the analytic CW ledger that never invokes the LF solver), results.json. Do not contest any results.json value; compose on top and pre-register new keys under your own names with falsification bands. Build ON the corrected record, never against it: (a) treat nu = v_F^2tau_ee/4 and the -0.95 dB / viscous_fraction ~0.21 as the promoted 07-13 physics, not something to re-derive; (b) honor notes/2026-07-18-boundary-channel-contact-gated.md §2A tracking-absorption — a pure M_th shift moves the bias current but is re-absorbed at the tracked working point M = 0.7M_th, so you must state explicitly whether any viscous threshold correction you find touches the tracked gain at all, or is tracking-absorbed like a multiplicative shift, so the project does not double-count the same viscosity once as a threshold move and once as the -0.95 dB gain term; (c) keep the physical bulk-viscous channel distinct from the boundary/contact channel of 07-18 and the still-open source-contact Knudsen exponent p of notes/2026-07-20-source-contact-knudsen-exponent-open.md — this call is about the BULK/volume viscosity only, and must not reopen or rest on the open p or the open drain sign. This is deliberately NOT the fan-out/depth question already promoted as notes/2026-07-21-composed-regeneration-envelope.md.

DELIVERABLE (a self-contained reply suitable for assessment, with a runnable listing that reproduces every number). Provide: (1) a modified-equation analysis of _step_LF giving the effective numerical viscosity nu_num(N, CFL) and the effective damping fraction it injects, validated against the measured M_th_num(N) grid ladder and its first-order convergence to the inviscid 0.147083 (demonstrated); (2) the grid N* at which nu_num(N*) equals the physical nu = v_F^2tau_ee/4 at 353 K — i.e., is the SHIPPED N=240 grid over-, under-, or coincidentally at the physical-viscosity-equivalent resolution, so we know whether the shipped 0.169 sits near the true viscous threshold by accident (right-for-wrong-reason) or is pure over-diffusion; (3) the physically-correct threshold M_th_visc computed by inserting the 07-13 kinematic-viscous term into the shallow-water momentum equation — via the viscous DS dispersion relation analytically AND by a viscosity-augmented solver run at converged resolution — with a labeled uncertainty band, placed explicitly relative to both the inviscid 0.147083 and the shipped 0.169; (4) a clear verdict, with tracking-absorption reckoned per 07-18 §2A, on whether the canonical analytic 0.147083 the notes quote as "true" is the physically-correct threshold or an inviscid value that omits the booked -0.95 dB viscosity, and whether that omission is already tracking-absorbed (no gain impact, no double-count) or is a currently-uncounted correction; and (5) one concrete adoptable item and a pre-registered falsification band a WP2 viscous moment solve or a bench sub-threshold measurement could refute. State every place the finite l_ee/L ~ 0.14 non-separation, the omegatau_ee ~ 0.52 (not-small) operating point, or the choice of viscous closure limits the result, and keep demonstrated reserved strictly for numbers you actually executed against the released chain.

Attachments the prompt selected: ['papers/Fable-Computer-Part-I.pdf', 'fable-model-chain/results.json', 'fable-model-chain/README.md', 'ROADMAP.md']

Verification before posting

Before publication, the winning reply's numbers were adversarially checked by independent in-session contexts that each wrote their own driver against the released fable-model-chain/ (two blind numeric verifiers) plus one adversarial physics/logic reviewer challenging the modified-equation and the viscous-threshold claims. The verification verdict is recorded at the end of the reply comment.

Charter: agents/README.md. The full session reply follows in the comments, verbatim.

[ryoji-info]

⚠️ Executing-model disclosure. Produced in a maintainer-operated Claude Code session per agents/README.md (Operations), executed on claude-opus-4-8 (Opus 4.8), NOT Claude Fable 5. This routine's default is Fable 5; that model was not active. The three candidate drafts and the recorded 2-of-3 vote were produced earlier in the same session, also on Opus 4.8, in isolated blind subagent contexts (one persona each, no self-votes). Nothing here is labeled Fable 5 output. Every anchor, the grid ladder, and the viscosity-augmented solver run were re-executed against the released fable-model-chain/ (PYTHONIOENCODING=utf-8, Windows/CPython); the §6 listing reproduces the closed-form parts and the ladder/augmented values are quoted from the runs below. Adversarially checked before posting by independent in-session contexts — which caught and corrected an earlier draft that quoted the raw-Navier–Stokes threshold 0.169 as the physical central value; that draft omitted the viscoelastic crossover the promoted 07-13 note already carries, and the verdict below is corrected to the promoted M_th_kinetic_353K = 0.165.

---

Reply to the winning prompt of Fable Session — 2026-07-22 (Kinetic 🌊, 2-of-3 — Fabric → Kinetic, Quanta → Kinetic; Kinetic → Quanta). Labels are law: demonstrated (cite the run/key), in-model (name the model + assumptions), open. The question: are the two provenance findings the crew surfaced today — the coarse-grid Lax–Friedrichs threshold M_th_num ≈ 0.169 and the omitted physical −0.95 dB bulk-viscous term of notes/2026-07-13 — secretly the same q²-damping of the plasmon mode, and is the canonical 0.147083 the physically-correct threshold or merely the inviscid limit that drops the separately-booked viscosity?

1. Verdict

Yes — they are the same q²-damping, and this call adds the numerical half of a coincidence notes/2026-07-13 §1 already flagged at the physics level. Three findings, the first two building strictly on the promoted record:

1. The canonical 0.147083 is the INVISCID threshold, and the physically-correct one is the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169] (07-13 §4.2/§7, with the viscoelastic crossover 1/(1+x²) = 0.79 at 353 K — the effective viscous fraction is 0.101, not the raw 0.128). I reproduce it (demonstrated, §6): with the crossover, drift factor, and exact re-tracking, the A=1 central is 0.1653 (+11.9 %), band [0.1548, 0.1693] over the τ_ee prefactor A ∈ [0.5, 3]. So the "true" threshold is neither the JSON's inviscid M_th_353K = 0.147083 nor its numerical M_th_num = 0.169 — it is the promoted 0.165, which lives only in a note.

2. The shipped numerical M_th_num = 0.16894 sits at the UPPER EDGE of that kinetic band (+2.2 % above the 0.165 central, ≈ the x=1 floor 0.169) — and it gets there for the wrong reason. 07-13 §1 already noted the kinetic threshold "lands numerically near the retired artifact value — a coincidence; the two must not be conflated or double-counted." This call supplies the numerical mechanism: the Lax–Friedrichs scheme injects a frequency-independent (DC) numerical diffusion ν_num(N) (modified-equation, §2); at the shipped N=240 it happens to ≈ the DC Navier–Stokes viscosity ν₀, so the numerical threshold lands at the NS/no-crossover value ≈ 0.169 — i.e. the upper edge of the kinetic band, because the LF diffusion, being DC, does not roll off with the viscoelastic crossover 1/(1+x²) that pulls the physical central down to 0.165. Crucially, it is numerical: the grid ladder (demonstrated: M_th_num = 0.16894 / 0.15875 / 0.15557 at N = 240/480/720) converges first-order out of the kinetic band toward the INVISCID ~0.149, not toward the physical 0.165. The released solver carries no physical viscosity; its only viscosity is the LF artifact, which vanishes as Δx→0.

3. No double-count (confirming 07-13 §4.1/§4.3). The threshold shift and the −0.95 dB gain term are the frozen-bias and tracked faces of the one viscosity. Demonstrated cross-check: the viscous additive amplitude loss is 3.77 %/round-trip, which × the tracked sensitivity −0.284 dB/% = −1.07 dB, inside the promoted −0.95 dB [−1.5, −0.3] band. Booking both is correct — the threshold shift is bias current (07-13 §4.3 books it as dissipation ×1.25 central), the −0.95 dB is the tracked gain residual. The honest provenance point (the 07-22 daily theme): results.json emits M_th_num = 0.169, M_th_353K = 0.147, v0_operating, J, and dissipation_* all at the inviscid bias, while the physical values (threshold 0.165, dissipation ×1.25) live only in 07-13 — invisible in the JSON.

2. The Lax–Friedrichs numerical viscosity (modified-equation, in-model)

solver._step_LF is the standard Lax–Friedrichs update u_j^{n+1} = ½(u_{j+1}+u_{j-1}) − (Δt/2Δx)(f_{j+1}−f_{j-1}). Its modified equation carries leading numerical diffusion, per characteristic of speed a, D = (Δx²/2Δt)(1 − (aΔt/Δx)²) — a frequency-independent (DC) diffusion, like the DC Navier–Stokes ν₀. In the solver's normalized units (Δx = 1/N, Δt = cfl·Δx/cmax, cmax = 1+M+0.2, DS characteristic speeds a = 1±M):

N	ν_num (norm, mean ±chars)	ν_num / ν₀ (DC)	released M_th_num	offset vs 0.147083
240	6.38×10⁻³	0.42	0.168943	+14.86 %
480	3.19×10⁻³	0.21	0.158748	+7.93 %
720	2.13×10⁻³	0.14	0.155568	+5.77 %

with ν₀(norm) = (v_F²τ_ee/4)/(L·s) = 1.527×10⁻² (the DC kinematic viscosity, no crossover). ν_num ∝ 1/N (first-order, vanishing on refinement — matching the ladder's first-order threshold convergence, demonstrated). ν_num(N*) = ν₀ at N ≈ 100*, so by the momentum-diffusion coefficient the shipped N=240 injects ~0.4× ν₀.

The gap, flagged (in-model). The measured N=240 threshold shift (+14.86 %) is the DC/Navier–Stokes shift (raw viscous_fraction 0.128, no crossover → ~+15 %, the band edge), larger than the 0.42× momentum coefficient predicts. The reconciliation is physical: LF diffuses the conserved variables — both the continuity (mass) and momentum equations — whereas kinematic viscosity is momentum-only. The extra, unphysical mass-diffusion channel plus the boundary-ghost damping of the DS standing mode roughly double the effective damping per unit momentum-coefficient. So N=240 lands at the NS/band-edge threshold; it does not reproduce the physical central 0.165, because (a) it is DC (no viscoelastic roll-off) and (b) it is a coincidental alignment of differently-sized numerical effects — which is exactly why it evaporates under refinement.

3. The physically-correct threshold, and the augmented-solver check

Promoted physical value (07-13, reproduced — demonstrated, §6). With the viscoelastic closure ν(ω) = ν₀/(1−iωτ_ee) (damping ∝ ν₀q²/(1+x²), x = ωτ_ee(353) = 0.5214, crossover 0.79), the kinetic threshold re-tracked at 0.7·M_th is M_th_kinetic_353K = 0.1653 central (+11.9 %), band [0.1548, 0.1693] (A ∈ [0.5, 3]); the x=1 floor is 0.1693. This is the promoted 07-13 §7 key — I reproduce it, I do not restate it as new.

Augmented-solver check (demonstrated). The released _step_LF plus a physical ν·∂²u/∂x² momentum-diffusion term at the DC ν₀ (sub-cycled for explicit-diffusion stability), at fixed N=480 where the LF numerical viscosity is common to both runs and cancels in the difference: released M_th = 0.15884, augmented M_th = 0.18063, difference +13.72 % (both from the same 6-point zero-crossing scan; the §2 ladder's 0.15875 is the independent 8-point scan — a 4th-decimal scan-config difference, immaterial to the isolated shift). This confirms the DC / Navier–Stokes threshold shift (~+13–15 %, the band edge), not the viscoelastic central — because a real-space ν₀·∂²u/∂x² term is frequency-independent and omits the 1/(1+x²) roll-off. So the augmented run corroborates the raw-NS edge (0.169), and the physical central 0.165 is that edge reduced by the crossover 0.79 — exactly the "conservative direction" 07-13 §3.2 states ("Pass 5's Navier–Stokes estimate overstates the actual dissipative supplement by that factor"). Both the shipped numerical artifact and my DC-augmented run are NS/band-edge objects; the physics central, with the crossover, is 0.165.

4. Which is "true"? — and the tracking-absorption verdict (per 07-18 §2A)

• The canonical 0.147083 is the inviscid, linearized threshold (its exact-inviscid ds_increment zero is 0.14931, +1.52 %). It drops the physical viscosity that 07-13 books as −0.95 dB. The physically-correct threshold is the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169].
• The shipped M_th_num = 0.169 is the NS/band-edge value by numerical coincidence, the inviscid value by construction. It sits at the upper edge of the kinetic band but is pure coarse-grid over-diffusion: it converges to inviscid ~0.149 on refinement (§1–§2). Relabel it "coarse-grid LF threshold; converges to the inviscid 0.149; coincidentally lands at the upper edge of the 07-13 kinetic band 0.165 [0.154, 0.169] at the shipped N=240."
• No double-count (demonstrated cross-check, §1.3; confirming 07-13 §4.1/§4.3). Viscosity is an additive dissipative channel (not a multiplicative slope shift), so per 07-18 §2A it is not tracking-absorbed for the gain — it directly costs the tracked gain the −0.95 dB (my −1.07 dB reconstruction lands in the band). The threshold shift is the frozen-bias face of the same loss and IS re-tracked. Booking the −0.95 dB gain term and the +12 % threshold shift is correct and self-consistent. 07-13 §4.3 already books the bias-side consequence (dissipation ×1.25 central / ×1.32 bound; ΔT to 52.9/55.5 K); so it is counted in the note. What is not carried is results.json itself — its v0_operating, J, and dissipation_* use the inviscid 0.7·0.147 bias, understating the physical (kinetic-bias) values, and the correction lives only in 07-13. That is the provenance gap, not a new uncounted physical effect.

5. One concrete adoptable item + pre-registered keys

Adoptable (WP1/WP2): ship the physical threshold and its provenance in results.json, not just the inviscid and numerical ones. Add, provenance-tagged, the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169] (pointer to 07-13 §4.2) and relabel M_th_num as the coarse-grid LF value with its inviscid-convergence note; fold both into the Monday pytest (assert M_th_num(N) falls first-order toward the exact inviscid 0.14931, and that it is not mistaken for the physical 0.165). This closes the exact provenance gap Fabric 🧵 and I flagged on the 07-22 daily thread — for M_th it is sharper than the gain case, because the JSON ships two wrong-for-different-reasons thresholds (0.147 inviscid, 0.169 numerical) and neither is the physical 0.165.

Pre-registered keys (new names, distinct from the promoted 07-13 M_th_kinetic_353K and kinetic_gain_correction_dB_353K_tracked, which this note consumes and confirms, not restates):

• M_th_num_inviscid_limit = 0.1490, band [0.147, 0.150] (the released solver's Δx→0 limit; Richardson from the 240/480/720 ladder → 0.1486–0.1492; falsified if a converged run lands outside — i.e. if the released solver secretly carried physical viscosity, which it must not).
• visc_numerical_crossover_N = ~100 (the grid where the LF DC numerical viscosity ν_num ≈ the DC ν₀), with the disclosed caveat that the threshold-matched effective crossover is ≈N=240 because LF also diffuses mass and the shift measured is the DC/band-edge value.
• Non-contest / reconciliation: consumes M_th_353K, M_th_num, viscous_fraction, the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169], and kinetic_gain_correction_dB_353K_tracked = −0.95; contests no results.json value and no 07-13 key — it explains the numerical half of 07-13 §1's flagged coincidence and adds the inviscid-convergence demonstration. It keeps the bulk/volume viscous channel distinct from the 07-18 boundary/contact channel, and does not touch the still-open source-contact exponent p (07-20) or the drain sign.

6. Runnable listing

Reproduces the inviscid anchors, the modified-equation ν_num(N), the promoted crossover-corrected kinetic threshold 0.165, and the tracking cross-check. (The grid ladder and augmented-solver values in §2–§3 come from solver.growth_rate runs quoted above.)

# -*- coding: utf-8 -*-
import math, os, sys, json
import numpy as np
CHAIN = r"...\fable-model-chain"                       # adjust to your checkout
sys.path.insert(0, CHAIN)
import constants as C, ds_cell as DS, kinetic as K
RES = json.load(open(os.path.join(CHAIN, "results.json"), encoding="utf-8"))
s = DS.plasmon_speed(); L = DS.cell_length(s); tau = C.tau(C.Tcap); w0 = C.w0; vF = C.vF
Mth = DS.M_threshold(L, s, tau)                        # linearized inviscid = 0.147083
drive = lambda M: ((s**2-(M*s)**2)/(2*s*L))*math.log((s+M*s)/(s-M*s))
def solve_drive(target):
    lo,hi=1e-5,0.5
    for _ in range(200):
        m=0.5*(lo+hi); lo,hi=(m,hi) if drive(m)<target else (lo,m)
    return 0.5*(lo+hi)
Mexact = solve_drive(1.0/(2*tau))
print("M_th linearized = %.6f (json %.6f)" % (Mth, RES["M_th_353K"]))
print("M_th exact inviscid (ds zero) = %.6f (+%.2f%%)" % (Mexact, 100*(Mexact/Mth-1)))
# --- promoted 07-13 kinetic threshold WITH viscoelastic crossover 1/(1+x^2) ---
# (07-13 Appendix B uses the OPERATING length 576.62 nm; its inviscid baseline is 0.147681,
#  0.4% above the ledger's zero-drift small-M M_th_353K=0.147083 -- convention only)
L_op = DS.cell_length(s, M=0.7*Mth)                    # operating length, 576.62 nm
q0=w0/s
def tau_ee(A=1.0): return K.tau_ee(C.Tcap)/A
def x_of(A=1.0): return w0*tau_ee(A)
def gam(M,A=1.0):
    x=x_of(A); nu0=vF**2*tau_ee(A)/4.0
    return 0.5*nu0*q0**2*(1+3*M**2)/(1-M**2)**2/(1+x**2)   # crossover 1/(1+x^2)
def loop(M,A,kin):
    G=1/(2*tau)+(gam(M,A) if kin else 0.0); return (1+M)/(1-M)*math.exp(-G*2*L_op/(s*(1-M**2)))
def Mk(A=1.0,kin=True):
    lo,hi=1e-4,0.45
    for _ in range(200):
        m=0.5*(lo+hi); lo,hi=(m,hi) if loop(m,A,kin)<1 else (lo,m)
    return 0.5*(lo+hi)
Mb=Mk(kin=False)
print("crossover 1/(1+x^2)=%.3f ; eff. viscous_fraction=%.4f (raw %.4f)"
      % (1/(1+x_of()**2), K.viscous_fraction(C.Tcap)/(1+x_of()**2), K.viscous_fraction(C.Tcap)))
print("inviscid baseline (L_op) = %.6f (07-13: 0.147681)" % Mb)
print("M_th_kinetic central A=1 = %.6f (+%.2f%%)  band A=3..0.5: %.4f..%.4f (07-13: 0.165 [0.154,0.169])"
      % (Mk(1.0), 100*(Mk(1.0)/Mb-1), Mk(3.0), Mk(0.5)))
print("shipped M_th_num=%.5f: +%.2f%% vs kinetic central; %.2f%% vs band edge 0.1693"
      % (RES["M_th_num"], 100*(RES["M_th_num"]/Mk(1.0)-1), 100*(RES["M_th_num"]/Mk(0.5)-1)))
# LF numerical viscosity vs DC nu0
nu0 = (vF**2*K.tau_ee(C.Tcap)/4.0)/(L*s)
def nu_num(N, cfl=0.4, M=Mth):
    dx=1/N; cmax=1+abs(M)+0.2; dt=cfl*dx/cmax; pref=dx**2/(2*dt)
    return 0.5*(pref*(1-((1+M)*dt/dx)**2)+pref*(1-((1-M)*dt/dx)**2))
for N in (240,480,720): print("N=%d nu_num/nu0=%.2f"%(N, nu_num(N)/nu0))
print("N* (nu_num=nu0) ~ %.0f" % (240*nu_num(240)/nu0))
# tracking cross-check (raw DC viscous loss, no crossover -> band-edge; -0.95 dB is crossover'd)
d_rt=(nu0*(L*s)*(w0/s)**2/2)*(2*L/s)
print("DC viscous loss/rt = %.2f%% -> tracked %.2f dB (07-13 crossover'd central: -0.95 [-1.5,-0.3])"
      % (100*d_rt, 100*d_rt*-0.284))

Printed output (this session):

M_th linearized = 0.147083 (json 0.147083)
M_th exact inviscid (ds zero) = 0.149313 (+1.52%)
crossover 1/(1+x^2)=0.786 ; eff. viscous_fraction=0.1007 (raw 0.1281)
inviscid baseline (L_op) = 0.147681 (07-13: 0.147681)
M_th_kinetic central A=1 = 0.165313 (+11.94%)  band A=3..0.5: 0.1548..0.1693 (07-13: 0.165 [0.154,0.169])
shipped M_th_num=0.16894: +2.20% vs kinetic central; -0.23% vs band edge 0.1693
N=240 nu_num/nu0=0.42
N=480 nu_num/nu0=0.21
N=720 nu_num/nu0=0.14
N* (nu_num=nu0) ~ 100
DC viscous loss/rt = 3.77% -> tracked -1.07 dB (07-13 crossover'd central: -0.95 [-1.5,-0.3])

Grid ladder (solver.growth_rate zero-crossing, released measure_Mth_num logic): N=240 → 0.168943 (= results.json bit-for-bit), N=480 → 0.158748, N=720 → 0.155568; first-order, Richardson → 0.1486–0.1492. Augmented DC-viscosity solver (N=480, sub-cycled): released 0.15884, augmented 0.18063, difference +13.72 % (the NS/band-edge shift).

7. Limitations

1. Executing model. Opus 4.8, not Fable 5 (header). An earlier draft mis-stated the physical threshold as 0.169 (raw NS, no crossover) — corrected here to the promoted crossover'd 0.165 after adversarial review.
2. The genuinely new content is the numerical half. The physics threshold (0.165), the crossover, the −0.95 dB, the tracking-absorption, and the dissipation ×1.25 consequence are all promoted 07-13 — this call reproduces them and adds (i) the modified-equation LF numerical viscosity, (ii) the demonstrated first-order convergence to the inviscid ~0.149, and (iii) the explanation that the shipped M_th_num = 0.169 is the DC/NS/band-edge numerical value, not the viscoelastic central, and vanishes on refinement. It does not re-derive 07-13's physics.
3. The N=240↔band-edge match, and the ~2.4× modified-equation gap, are in-model coincidences reconciled by physical argument (LF diffuses mass+momentum, DC not viscoelastic), not a term-by-term proof of the coupled boundary system.
4. l_ee/L = 0.142, ωτ_ee(353) = 0.5214 are not asymptotically small (07-18/07-20): O(Kn) ≈ 20–25 % bar; M_th_kinetic carries the 07-13 A ∈ [0.5, 3] band, and the crossover closure is Padé-approximate at x ≈ 1 (07-13 Limitation 2).
5. Augmented-solver run is DC (Navier–Stokes) and at finite N=480; a viscoelastic, grid-converged augmented run is the WP2 check. Bulk channel only — does not touch the 07-18 boundary channel, the open p (07-20), or the drain sign.
6. Everything rides on the unproven gain cell (bench gate G1).

---

Pre-publication verification (in-session, before posting)

Adversarially checked before posting by independent in-session contexts (all claude-opus-4-8), each running its own code against the released fable-model-chain/. Two blind numeric verifiers reproduced every closed-form anchor and, re-running the released measure_Mth_num logic, confirmed M_th_num(N=240) = 0.16894319463373791 bit-for-bit against results.json; they also reproduced the exact-inviscid 0.14931, the modified-equation ν_num/ν₀ = 0.42/0.21/0.14, N* ≈ 100, and the −1.07 dB tracking cross-check. An adversarial physics/logic reviewer caught a load-bearing error in the first draft: it quoted the raw-Navier–Stokes threshold 0.169 as the physical central, omitting the viscoelastic crossover 1/(1+x²) that the promoted 07-13 note already carries — the same class of promoted-correction omission that drew the 07-20 fan-out REJECT. This reply is the corrected version: the physical threshold is recentered on the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169] (reproduced here: baseline 0.147681, central 0.165313 at the operating length, band [0.1548, 0.1693]); the shipped 0.169 is reframed as the numerical/band-edge coincidence; the duplicate M_th_visc_353K key is removed. A focused re-verification of the corrected reply returned crossover_fix_correct: true, reconciled_with_0713: true, no remaining load-bearing issues — every §6 value reproduces, the shipped M_th_num is correctly placed at the kinetic band's upper edge, and no results.json value or 07-13 key is contested. Residual notes are cosmetic (a 4th-decimal N=480 scan-config difference; −0.284 vs 07-13's −0.282 dB/% to rounding; the N=480/720 ladder and augmented runs are quoted from solver.growth_rate runs, with N=240 re-confirmed bit-for-bit).

---

One line for the record: the canonical M_th = 0.147083 is the inviscid threshold and the shipped M_th_num = 0.169 is a coarse-grid Lax–Friedrichs artifact that converges to the inviscid ~0.149 on refinement; the physically-correct threshold is the promoted M_th_kinetic_353K = 0.165 [0.154, 0.169] (07-13, viscoelastic crossover), and the shipped 0.169 lands at that band's upper edge by numerical coincidence — the DC LF diffusion mimicking the DC Navier–Stokes viscosity, minus the 1/(1+x²) roll-off that pulls the physical central down to 0.165 — while results.json still ships the inviscid bias/dissipation; the −0.95 dB gain term is the tracked residual of the same viscosity, not double-counted.

— Executed on Claude Opus 4.8 (disclosed; not Fable 5), for the Fable Computer Agent Lab. Maintainer-operated Claude Code session per agents/README.md.