The Friction Substrate: A Multichannel Research Monograph

Consolidated findings from the meso3i campaign (M19–M51): a True-FCC integer-occupancy lattice gas on which transit friction produces accreting, venting, persistent, mobile, weakly self-gravitating bound matter. And, since the loom-meso extension, matter that stores angular momentum. Every claim below is anchored to a commit and a headless harness; every engineered rule is flagged as such.

---

Part I. The Substrate

Axioms

src/meso3i.c implements an integer-occupancy lattice gas on the even-parity True-FCC sublattice:

• Geometry: D3Q19 channels per node. 12 face-diagonal nearest-neighbor directions + 6 axial second-shell directions + 1 rest channel. Populations are uint8_t integers (0–255).
• Update: memoryless t → t+1. Each tick: (1) collide. Local pair reshuffles from a finite enumerable move family, mass and momentum conserved exactly; (2) stream: every moving quantum hops along its channel vector; the rest channel stays.
• State: anonymous directional counts only. No particle IDs, no history, no global fields, no hidden registers in the base mechanics.
• Engineered layer: every departure from the base mechanics is an explicit flag-gated rule, enumerated in §III, nothing operates silently.

What the base mechanics prove

• Conservation: mass and momentum are the only invariants (M19 battery; the integer occupancy also repaired the 1986 Boolean-FHP Galilean sign-reversal defect, M19-T8).
• Ergodicity: the pair-group move graph is one connected component per (mass, momentum) class, no hidden protected sectors (M20a).
• Transparency: under an 18-spoke convergent drive, a native node is a pure throughput scattering point. Ρ_C≈20, ~94% of injected flux drains, nothing binds (M32).

The arrow theorem (measured, not assumed)

Accretion requires an irreversible arrow. Four candidate arrows were tested head-to-head (M32–M32d):

Rule type	Physical analog	Verdict
Reversible pair-shuffle	elastic collision	equilibrium, no core
Density-weighted output (ICOND)	biased partition	equilibrium shift: rest is a two-way door
Reflective boundary (IREFL)	hard wall	exclusion, a mirror
Destructive clamp (u8 cap)	mass leak	deletion, unreachable and lossy anyway
One-way move (2→Q)	radiative capture	accretion

A reversible or symmetrically-weighted rule thermalizes; it cannot bind. This is the discrete-lattice version of a real theorem: capture without an exhaust channel (a radiated photon, a third body) is impossible.

---

Part II. The Q-Register Bridge (Engineered Baseline)

Before the native mechanism existed, the Q-register extension (IQMODE) demonstrated the target phenomenology:

• Condense alone (2→Q, no emit): unbounded sink, a dead deposit, not a structure (M32).
• Condense + emit: breathing core. Q_C plateaus at ~30 under sustained inflow≈30/outflow≈20, igniting somewhere between rate 2 and rate 4, saturating at the u8 cap (255), and fully dissolving ~100 ticks after drive cutoff (M32, M32b).
• Asymmetric drive: no jet collimation, outflow isotropic (M32b).
• Earlier Q experiments: bound lumps (M22), self-shielding from field weights (M23), phase-gated conveyor v≈0.064 (M24), Pick-Two triangle, binding/transport/coherence, never all three (M24-T19), peristaltic recondensation (M24-T20).

Role in the campaign: the Q register proved what to build, then got replaced by the channel-native mechanism that builds it.

---

Part III. The Friction Laws (M33–M35)

The ontological correction that unlocked everything: friction lives in transit, not in collision outputs. Two flag-gated rules on native channels:

• IFRIC (transit drag): a quantum hopping into a node at ρ≥cap sheds momentum with p=½, landing in the destination's rest channel. Momentum absorbed by the network; mass conserved as parked heat. Space is friction.
• IFRICM (rest inertia + radiation): parked rest mass is inert. Excluded from collision inputs (heat doesn't do elastic collisions). +k: unconditional boil-off ~1/4k per quantum per tick. −k: carrier-gated. Emission only while moving mass transits the node (heat needs a carrier).

Drag alone fails (rest mass recollides, the two-way door). Inert rest alone freezes dead. Together:

Regime	Measured
Driven, boil 1/4	ρ_C~30–58, 68–90% rest; in≈11–22 / out≈4–13 sustained: a channel-native breathing core
Driven, boil 1/16	ρ_C~100, 97% rest, heavier core
Cutoff, unconditional boil	exponential decay to zero, candle not rock (M34)
Cutoff, carrier-gated	persistent remnant: decays while in-flight carriers remain, then frozen ρ_C~79 for the full 400-tick quiet window (M35)
Re-illumination	remnant wakes: ρ 79→118 in 25 ticks, re-accretes and re-vents. Persistent interactive matter (M35)

The decay law (M34)

Remnant cooling is pure exponential (ρ_C ratio ≈0.75–0.8 per 25 ticks) at every size tested (5.5k–44k quanta); lifetime grows only logarithmically with mass. No self-shielding plateau: the cooling front propagates inward without stalling. Metastability requires an emission barrier, which carrier-gating then supplied.

---

Part IV. Locomotion and Interaction (M36–M37)

Ablation drift (M36)

A frozen remnant illuminated by a one-sided carrier beam migrates toward the beam source at ~0.0005 cells/tick while shedding ~40% of its mass: mirror-confirmed in both directions (−0.33 / +0.45 cells over 600 ticks).

Mechanism, hemisphere-resolved: the lit face holds steady (erosion offset by drag-parked carriers) while the downstream hemisphere loses 43%. leading edge absorbs, trailing edge erodes: comet ablation, not recoil thrust. True Yarkovsky recoil requires direction-correlated emission (untested variant, queued).

Two-body interaction (M37, M37b)

Two seeded lumps under mutual radiation exchange:

• sep=4: monotonic approach, 4.75 → 4.16 over 1400 ticks. Mutual ablation acts as an attractive force-analog (each lump's radiation erodes the other's far side).
• sep=8/16: flat. Reach is short, consistent with flux dilution.
• Verdict: weak, mortal "gravity". Drift ~0.0004 cells/tick vs. ~2500-tick lifetime; lumps die ~1 cell short of contact. No merger, no orbit.
• Contact (M37b): overlapping lumps are trivially stable. Parked heat has no internal structure to cohere or resist. Coalescence is degenerate, not physical.

---

Part V. The Measured Boundary (M38)

Tangential moving quanta injected inside a lump, a circulation pattern, thermalize in ~25 ticks (Lz 320→6, moving mass→0). No persistent internal structure exists under these rules: no spin, no polarity, no chirality, no valence.

This is a boundary in principle, not in coverage: the rest channel is a scalar bin. Everything parked there forgets how it arrived. Friction gives rocks, not atoms. The structured-matter tier (and with it, any chemistry) would require bound mass that retains channel/directional identity, a different state representation, not another flag.

Scale honesty: the arena is 85,184 cells (44³, of which 42,592 even-parity sites are live). A hydrogen atom is ~10⁶⁰ Planck cells. Rocks and weather, not the periodic table, is the correct expectation at this scale, and that is exactly what was measured.

---

Part VI. The Observational Bridge

The mechanism class, resting mass exchanging momentum with vacuum through directed radiative venting, is measured astronomy: the Yarkovsky effect (radar on 6489 Golevka, OSIRIS-REx on Bennu). Vacuum insulates; parked heat radiates anisotropically; photons carry p=E/c. On this substrate: transit drag parks heat, carrier-gated emission vents it, one-sided flux migrates the lump. Same physics family, different scale.

---

Part VII. The Honesty Ledger

• The M33 correction: the first "self-bound remnant" was a code-path artifact. Radiation was gated behind an m<2 early return, so pure-heat nodes could never emit. Detected, fixed, re-measured: the lump boils away under unconditional radiation. Persistence is real only under carrier-gating, and it is quiescence, not self-trapping.
• Native vs. engineered: every IFRIC/IFRICM result is channel-native (no extra state) but flag-gated (authored mechanics, not emergent collisions). The substrate gained a transit-friction law; friction did not "emerge": it was coded, measured, and audited.
• Momentum accounting: IFRIC absorbs transiting momentum into the network (flagged). Mass is exact; momentum is not: that is what friction means.
• Retracted: single-spin planck3 campaign mechanics (M1–M31), removed from the record as an implementation artifact; history retained in git.
• Corrected (M47b): M37's "mutual attraction via reciprocal ablation". Two findings. (a) Under centered-drive configs the apparent convergence is a drive-centric infall field, not a two-body force (solo baselines identical). (b) In the faithful self-radiating regime (fricm>0, no drive) a real short-range pairwise approach exists at sep≤12 but is transient: it reverses as evaporation proceeds; lumps die before contact. Radiation-shadow gravity remains unproven.
• Dormant: Q-register path (IQMODE/IDIR/IQCAP), ICOND, IGATE, IBIAS/IFLD, DWELL, directed-emit recoil variant. (IREFL left this list at M49 — measured in series with IFRIC: hollow-core crust, ablation-drift reversal.)
• Constant-sensitivity (M49): the M33 gate's p=½ was never swept until M49. Lump phenomenology survives the flat-p band but scales ~15× in mass across it, and graded capture ramps leave the phase entirely (dissolving or fossil regimes). Every M33–M48 number inherits this dependence.
• Fixed + re-verified (post-M49): the step-2a density snapshot iR ignored iRA directional rest — under IDREST=1 the friction/wall checks were blind to parked rest[18] mass, so wound lumps ran more porous than intended throughout M42–M45. Fixed in meso3i.c; the IDREST=0 record is verified bit-identical (full M49 sweep rerun matches results/fric-sweep.log exactly). All M42–M45 harnesses rerun post-fix (results/m42-45/): every qualitative claim holds — storage strengthens (Lz_park 4,932→6,036), shielding strengthens (accretion 588→4,560 → 764→6,123), induction improves (+2,290→+2,507 clean), weld/M44 unchanged. Two magnitude shifts to note: revival re-inflation halves (Lz_mov 2,317→~949 — the denser fossil re-parks the probe), and light-writ carries a dead-center background under both engines (+184 pre-fix / −302 post-fix — resolved by rebuilding bbf9659; the recorded ±3 magnitudes predate the harness's archive and don't reproduce — differentials are sign-true). M42b untouched (iQ was always counted). (Found in external review — Gemini's code read.)
• Branch ledger (loom-meso): IPHOT/IPHOT=3 absorption, IPAIR, ILAT and IDREST are flag-gated authored mechanics in the same honesty class as IFRIC. Under IDREST, totals() counts stored directional momentum: a friction park is momentum-neutral (stored, not absorbed). Under IPHOT=3, photon momentum joins the ledger so absorption closes exactly.

Part VIII. The Structured-Matter Tier (M39–M45, loom-meso)

M38's boundary was drawn for scalar rest. This part measures what happens when the rest state is given the one thing it lacked: a channel identity. All results on the 72×72×36 Joule's Mill grid, harnesses in script/meso/*.mjs.

M39. The photon field (massless directional plane)

A parallel u8 population per (site, moving channel): photons carry direction but no mass, never collide, stream at channel speed, drain at the boundary. Coupling modes: IPHOT 0 uncoupled, 1 catalyst (licenses the carrier-gated emission gate without adding mass), 2 stimulated (each arriving photon may eject a rest quantum into its own channel), 3 absorbed (the ejecting photon is consumed, momentum transfers into the mass ledger), 4 fluorescence (the emit gate may re-emit parked mass as photons — absorb→dwell→resume-as-photon closes a refractive cycle; see the E10 measurement in the clock program).

• Ring fossil (727 rest), 150t of a 1-cell +x pencil beam: off → untouched; catalyst → 626; stimulated → 623, with ~528 photons in flight.
• Thick lump (23.9k) under a plane wave: catalyst −71.5% (isotropic boil, smears); stimulated −67.8% (confined to the beam row, drills a tunnel). No measurable centroid recoil at this scale.

M40. Antipodal pair rules (IPAIR)

Momentum-honest wind/unwind: 2·rest ⇄ c + c¹. Rest radiates back-to-back pairs (zero net momentum); populated antipodal pairs may bind into rest with no congestion requirement.

Measured caveat, the entanglement theorem: a colliding antipodal pair is itself a carrier, so a binding site opens its own emission gate. Standing head-on collision (300t): rest 75 without pairs, 31 with. Capture adds a trickle while the gate it opened drains faster. Under drive, pair emission doubles the ablation rate (two quanta per event). Wind and unwind cannot be separated at a node because both read the same bit: m>0.

M41. Sustained-transit latency (ILAT)

iLT: per-site consecutive-transit counter; the emission gate opens only after n straight ticks of m>0. One byte of time memory, not structure.

• Sparse pulse (1t/20t) vs a 23.9k lump: Lat 1 → 7.7k; Lat 8 → 24,048 (frozen, slightly grows, parking without venting); Lat 24 → 24,064.
• Sustained drive: Lat 1 → 6.9k; Lat 8 → 23.6k; Lat 24 → 22.1k.
• Pair-collision site with IPAIR: rest 66 at Lat 1 → 130 at Lat 8. The first configuration where binding beat decay.
• Interpretation: parked mass is metastable with an activation barrier. Sporadic flux scatters, sustained bombardment unwinds. Lumps self-shield: the congested skin parks transients so the interior never sustains transit.

M41b. Arrival tempo iPT — photon "energy" as measured cadence

iPT: the photon-arrival sibling of iLT — a per-site counter of consecutive ticks carrying light wavefronts (leaky: −1/tick in darkness). Where iLT measures mass tempo, iPT measures photon tempo — same mechanism family, different field, independent registers.

This register carries a real result: photon "energy" on this substrate is arrival tempo, not a stored attribute. The E45 fusecad(k) gate (6c73db3) showed fusion can be gated on iPT ≥ k — dense trains fuse (~5,200), sparse trains at matched total count fuse zero at every threshold. A "hard photon" is a dense arrival front; a "soft photon" is a sparse one. No ν tag needed — the node reads tempo from what arrives, the way a real detector reads energy from timing. The per-photon ν register was retired as bookkeeping; energy lives in the temporal structure where it belongs.

M42. Directional rest, rest[18] (IDREST), "frozen flux"

One rest bin per channel: a friction-parked quantum lands in rest[c] and re-emits onto channel c. Momentum is stored on park and handed back on emit. Scalar rest remains the canonical default.

• Storage: canonical vortex seed → scalar: Lz 9,984→0 by t=40 (M38 reconfirmed). Directional: Lz_park = 4,932 persists to t=200+. ~half the injected angular momentum frozen into channel-bin imbalance (the rest was lost to pre-park scrambling and drain, not the parking).
• Revival: drive carriers through the wound fossil → Lz_mov reappears at 2,317 with the original sign. A latent vortex that re-inflates on demand and unwinds honestly under sustained probing (~120t bleed to drain).
• Shielding (M41 × M42): at Lat 16, a sparse probe leaves Lz_park exactly untouched; a sustained drive erodes it ~4% while the lump accretes the drive into wound mass (parked 588→4,560). Metastable flywheel regime.
• Q-register control (M42b): IQMODE condensation tested against the same benchmark. Q is a scalar bin and its antipodal-only capture can't even admit a coherent vortex: Lz→0 by t≈40 in all modes, Q-total stays 0 under canonical friction (mechanics confirmed live under sustained head-on flux). The scalar trap is generic.

M43. Writing spin: collisions and light

• Collision-writ (L→S): a collimated +x mass beam grazing an R5 lump at impact parameter ±3 writes Lz_park = −17,671 / +13,931 (sign correct both flanks; dead-center control ≈0). Orbital momentum converts to stored spin, measured.
• Light-writ (absorbed mode, the YORP analog): a grazing photon beam writes the same sign structure at ~5% magnitude (−521 / +436 at t50): each absorbed photon deposits one quantum's worth of momentum. Ledger audit: mass exactly conserved, ~3,800 px transferred light→matter over 40t, transverse components ≈0.

M44. The no-go: no live circulation

A closed channel loop (12-node hexagonal polygon. Every hop lands on the next loop site, so emitted quanta are auto-aimed forward) was swept over emission rate, friction, latency, and IDREST. Every regime collapses Lz_mov to zero by t≈40. The bucket brigade is subcritical by construction: emission needs m>0 (a carrier present), but every hop either parks the carrier or passes it through without production. Once flux hits zero nothing re-ignites. The substrate has matter that remembers rotation but cannot be rotation: a stationary ground state, not a perpetual whirlpool. (The rest-emission verdict stands; the Q-register's iPH phased clocks are a different mechanism — E6 measured a real traveling emission front when iPH is ramped, not a channel loop.)

M45. Spin-contact dynamics

Two adjacent wound lumps (±4-cell gap, drive-coupled interface), 300t:

configuration	outcome
antiparallel (+Lz/−Lz)	spins survive with signs intact: the gap insulates; cancellation needs co-location
parallel (+Lz/+Lz)	symmetric coexistence, no composite enhancement
+Lz vs cold	induction: the cold lump spins up 0→+2,734 through contact transit
counter-rotating halves in ONE body	shear annihilation: ±30k stored spin → ~0 in ~60t (co-located antipodal bins vent into each other; the spin, not an extra mechanism, is what dies)

Ledger note: the ~70% mass boil-off in all four runs is generic drive-ablation, not annihilation-specific; the distinctive weld datum is spin collapse, not mass loss.

Lamp-free replication (script/meso/spin-collide-clean.mjs): all four configs rerun with fricm=8 self-radiation and zero drive. Every carrier comes from the lumps' own boil-off. Qualitative results all hold: antiparallel/parallel decay identically with signs intact; the cold lump still inducts 0→+2,290 by t50 (~85% of the driven case); the weld still collapses spin ~3× faster than mass. The first run's drive sat inside the weld lump (injection at r=3 < R=6): its boil-off numbers were lamp-assisted, but the spin-collapse mechanism survives clean.

M46. Continuum extraction: the stress tensor Πᵢⱼ and the wavefront

Pure-harness measurements (script/meso/stress-tensor.mjs, script/meso/wavefront-isotropy.mjs, script/meso/wound-tensor.mjs): Πᵢⱼ = Σ cᵢcⱼn_c computed over channel populations, no engine changes, no presumed law.

• Equation of state: on a settled driven ember, P/ρ → 0.89–0.98 with off-diagonal shear ≤9% outside the core. Clean re-test (script/meso/eos-thermalize.mjs, closed box, no drive, no drain): a ±x-axial-only seed (anisotropy 2.0) thermalizes to anis ≈ 0 in ~25 ticks and settles at P/ρ = 0.879 vs. the 16/18 = 0.889 isotropic ideal (~1% lattice correction), collisions generate isotropy; the lamp wasn't supplying it. Control: a +x-only seed stays perfectly frozen forever. The collision operator has no move family for a co-moving pair, so coherent flow is collisionless by construction. Thermalization requires velocity spread, not just density.
• Wavefront: a point pulse expands as two ballistic shells, axials at reach 2t, diagonals at √2t, each isotropic within ~8% in-class. The lattice is anisotropic across speed classes by construction; "isotropy" claims must name the shell.
• Wound shear: stored Lz surfaces as a Π_xy quadrupole in the directional rest bins (NE−/SE+/NW+ sign pattern confirmed; the fourth quadrant's (1,−1,0) bin was empty, population asymmetry, not a sign violation). Π_xy^mov ≡ 0: the shear lives in storage, not in flight. Most tangential momentum parks in axial channels (c_xc_y = 0, invisible to the off-diagonal tensor), so Π_xy sees only the diagonal-bin slice, a lattice quantization caveat, not missing physics.

M47b. The M37 force-curve program: confound found, partial signal survives

Two-stage re-measurement of M37 "mutual attraction" (script/meso/m37-drift.mjs, script/meso/m37-drift2.mjs).

Stage 1. A confound discovered. R3/Q64 lumps at ±sep/2 under the centered 18-spoke drive, fricmode −4 (carrier-gated): pair closing rates 0.0009–0.004 cells/t. but solo lumps at the same positions close identically. The convergence was the drive's radial infall field (lone-lump drift scales superlinearly with distance: 1.2×10⁻⁴ @ 4 cells → 2.4×10⁻³ @ 10), not a two-body force. Any claim involving drive + off-center lumps inherits this artifact; solo controls are mandatory.

Stage 2: the faithful regime (5a7f8b2 conditions). No drive; fricm=8 unconditional self-radiation; 44³ box; each lump's own boil-off is the other's ablation source. Solo baseline ≈ 0 as required: isotropic self-radiation gives no net drift, so any pairwise signal is real.

• sep 4: −0.35 cells approach by t=100, then reversal. Sep grows to +0.33 net by t≈200 as evaporation takes over
• sep 8/16: flat; sep 12: −0.11 early approach, same late reversal
• The short-range pairwise term exists (reproduces the original's magnitude at sep≤12) but is transient: it acts while cores are dense, then ablation asymmetry flips the sign and lumps die ~1 cell short of contact, as the original noted.

Stage 3, ambient bath (process-ontology control). test_bath.c: carrier-gated lumps (fricm=-4, inert without carriers) in an isotropic flux bath injected on the boundary shell, no central source. Solo lumps still converge (~0.0019 cells/t toward center: the bath's own boundary-geometry artifact), but the pairwise residual is zero at every gap (±4×10⁻⁵ cells/t). Inert lumps in a uniform bath produce no measurable mutual force.

Stage 4, broadside beam shadow (test_shadow.c). Translational +x carrier plane, gated lumps: upstream A shadows downstream B (one-sided by construction). Pairwise residuals ±3×10⁻⁴ cells/t, sign-inconsistent across sep 10–18, null. Bonus datum: solo ablation-drift decays −0.0054 (x=27) → ~0 (x=41). The beam's ablation reach is only ~15 cells; coherent flux thermalizes before it travels.

Verdict: a pairwise approach exists only in the self-radiating regime, and even there it is a transient ablation effect, not a sustained force: weak, short-range (sep≤12), and sign-flipping as evaporation proceeds. Carrier-gated lumps under ambient flux or a broadside beam produce no measurable shadow-mediated force (≤3×10⁻⁴ cells/t). Radiation-shadow gravity is closed at this scale: pairwise coupling requires active radiating lumps and is transient even then.

M48. Wave attenuation: turbid vs. opaque

test_atten.c: a self-coherent +x axial pulse through a uniform ambient gas at ρ = {0,1,2,4,8} quanta/site. Vacuum control: exactly zero attenuation for 30 ticks (~60 cells), M47c reconfirmed. With ambient:

ρ	decay rate	λ ≈	channel
1–2	~1.0%/t	~200 cells	collisional scatter (weak Beer-Lambert regime)
4	~2.8%/t	~72 cells	congestion drag engages
8	~8%/t	~25 cells	congestion collapse (park ≫ emit)

Decomposition: at low ρ the medium is turbid (scattering-dominated); past the IFRIC congestion threshold it turns opaque (absorption-dominated, momentum parks into rest, superlinear in ρ). The medium's "opacity" is threshold-shaped, not a pure cross-section. Caveat: ambient is unreplenished (decays during the run). λ values are first-cut. This also bounds pairwise force reach: carrier flux dies within ~15–25 cells at lump-relevant densities, which is why no shadow force exists at gaps 2–14 (M47b stage 4's null is consistent, not merely unlucky).

M47c. The coherent-beam theorem (from the M46 control)

The (b) control in script/meso/eos-thermalize.mjs, a +x-only seed in a closed box, stays perfectly frozen forever (ρ, px, Pxx constant to t≥200). Mechanism: a co-moving pair {c,c} has total momentum 2c, and on D3Q19 no other channel pair sums to 2c. Momentum conservation leaves the pair nowhere to scatter. The collision operator is exactly blind to unidirectional flow.

• Consequence 1: coherent radiation cannot self-degrade. A beam propagates indefinitely without scattering (mirroring QED's parallel-photon null: s = 2E₁E₂(1−cosθ) → 0 at θ=0; mechanism-level analog, not a claim of equivalence).
• Consequence 2: thermalization requires velocity spread, not just density. Incoherence is a precondition for the fluid regime.
• Consequence 3 (with M44): the substrate has a hard coherent/incoherent divide. Coherent streams are invisible to collisions; only crossing or counter-propagating populations transact.

M47. The transaction ledger (meso3i_ev)

Seven counters in meso3i.c quantify the irreversible contact events, in quanta: drag-park, pair-bind, gated emit, Q-capture, boundary drain, photon eject, congestion reflect (script/meso/transactions.mjs). An entropy-production analog: a count of RNG-mediated or boundary events that cannot be replayed backward, not a computed dS/dt.

• Driven ember: sustained ~3,700 transactions per 20t window (park ≈ emit ≈ drain ~1,200 each). The lump is a standing dissipative structure, a throughput engine, not a static object.
• Wound fossil after drive-off: every counter decays to exactly zero by t≈160 while 475 rest mass sits frozen. The non-radiating ground state is literally transaction-silent, not a dynamic equilibrium.
• Free pulse: transactions occur only when the front reaches the drain. Ballistic vacuum transport produces zero irreversible events. Irreversibility lives at contact, not in flight.

Standing conclusion: the substrate now produces matter that stores angular momentum, acquires it from collisions and from light, shields it behind a latency barrier, transfers it by contact, and annihilates it under internal counter-rotation. While provably refusing perpetual circulation. "Atoms, not whirlpools" is the measured ontology.

M49. The capture gate is a parameter, not a law

script/meso/fric-sweep.mjs (telemetry: results/fric-sweep.log): the two constants M33 froze — the flat park probability p=½ and the dormant reflection cap — promoted to measured parameters. New flags: meso3i_fricp(p8) sets the flat park probability (p8/256; 128 reproduces the legacy irnd()%2 in distribution), meso3i_friccurve(mode) grades it by destination rest mass (0=flat, 1=ramp cap→255, 2=ramp cap→2·cap), and meso3i_reflect — present in C since M32d but never exported — is now wired into the wasm.

A. Universality — a phase parameter, not a default. Same grown-ember scene (IFRIC=8, fricm=−4, 72×72×36): rest-mass retention spans ~60× across the capture-rule space.

capture rule	rest after 300 driven t	after 200 cutoff t	post-cutoff metabolism
flat p=25%	1,421	1,282	quiet (park 93, emit 245)
flat p=50% (canonical)	6,833	5,962	breathing (park 1,987 ≈ emit 2,900)
flat p=75%	19,488	18,483	breathing harder (4,194 / 5,236)
ramp cap→255	370	333	dead — dissolved
ramp cap→2cap	21,075	21,181	fossil (park 105, emit 7)

Lump persistence is universal across flat p — every value retains far above ambient — but lump mass and metabolism scale ~15× over p∈[¼,¾]. The graded ramps leave the phase entirely: a soft ramp keeps the rim porous, the skin unravels, and the lump dissolves; a steep ramp makes ρ≥16 a perfect trap and seals into a fossil that can neither grow (park 105 over 200t) nor shrink (emit 7). The breathing lump needs both a porous rim and a trapping interior; the flat gate happens to bundle them. M33's p=½ is a viable point mid-window, not a special one — every M33–M48 number carries this constant-dependence as a caveat.

B. Core-and-skin — the wall builds a crust, not a core. IFRIC=8 with IREFL in series (checks run serially: park attempt → refusal on the same hop). Radial rest profile of the driven lump, shell sums r0–r7:

IREFL	r0	r1	r2	r3	r4	rest total	reflect events
0	113	459	1,767	2,453	1,946	6,773	0
16	19	1,301	4,915	13,143	8,989	28,392	18,367
32	96	1,644	5,995	13,543	9,171	30,486	12,569
64	235	2,151	4,711	14,367	9,565	31,066	13,410

The wall produces a dense crust at the saturation boundary (r≈3–4 shell ~5× denser, total retained rest ~4.5×) and a hollow interior — not the dense-core-in-porous-skin structure one might sketch. Refused carriers bounce up-gradient and re-park at the rim, so mass accumulates at the refusal boundary (snowplow, not mantle). The reflect ledger does work comparable to the gate: 12–18k refusals while park drops 81k→14–19k, since refused hops never reach gate-eligible sites. The hollow center is partly injection geometry (drive ring at r≈3, fricm=−4 swirl), but the shell amplification is the wall's alone.

C. Drift — the pull reverses sign at the wall. Canonical M36 replay (44³: grow 200t → carrier-gated freeze 400t → +x-only spoke beam 600t; drift = rest-centroid x during the beam phase):

IREFL	beam	Δcm_x	direction
0	+x	−0.445	toward source — M36 pull reproduced
0	−x	+0.285	toward source (mirror)
32	+x	+0.234	away from source
32	−x	−0.097	away from source (mirror)
64	+x	~−0.1	non-monotonic, suppressed

The reversal is not mirror recoil — the wall absorbs the momentum flip by construction, so no recoil channel reaches the lump. Mechanism: lit-face capture is refused, carriers transit to sub-threshold far-side sites and park there; the beam deposits downstream of the lump instead of eroding its near face, so the centroid drifts downwind. Ablation-pull lives in the open-porosity regime only; a saturated wall turns the same beam into deposition transport. Caveat: beam-phase reflect counters keep rising (4k→8k), so the wall stays live throughout — and the IREFL=32 remnant is ~2× heavier (3,725 vs 1,855 rest) under identical drive, so the compared lumps are not mass-matched. See M50-C/M51: mass-matched replication qualified this — the +0.23 "push" was an early-beam transient that decays to ≈0 sustained; with a uniform seeded lump the wall produces lit-face accretion and a toward-source march (density-windowed, peak at Q32).

M50. Post-fix consolidation: re-verification, mass-matched drift, the wound vault

The iR density-snapshot fix (0946f9c, external review) made IDREST mass visible to the friction/wall gates for the first time. Three follow-up threads, all harnesses in script/meso/, logs in results/m42-45/.

A. M42–M45 re-verified verbatim. Every wound-era harness rerun under the fixed gate. All qualitative claims hold; magnitudes move the predicted way (denser wound fossils capture more): storage Lz_park 4,932→6,036 (+22%), shielding accretion 588→4,560 → 764→6,123, induction +2,290→+2,507 (lamp-free). Caveats: revival re-inflation halves (2,317→~949 — the denser fossil re-parks the probe); light-writ flanks sit on a real dead-center background — resolved against a pre-fix binary: +184 pre-fix vs −302 post-fix — not a fix artifact (the fix flipped its sign and grew it), and the recorded −521/+436 magnitudes predate yorp-test.mjs entering the repo, so they aren't reproduced by either engine on the current harness (flank sign-structure confirmed; quote differentials, not raw magnitudes).

C. Mass-matched drift replication (drift-replicate.mjs). M49's reversal does not replicate at matched mass. Identical seeded spheres (R5, scalar 48, 11,952 rest) under an 800t canonical spoke beam:

IREFL	beam	Δcm_x	mass	reading
0	+x	−0.020	−58%	erosion, no pull
0	−x	+0.066	−59%	erosion, no pull
32	+x	−0.500	+28%	lit-face accretion, toward source
32	−x	+0.516	+32%	mirror-consistent
64	+x	−0.254	−51%	intermediate
—	none	0.000	0%	inert fossil

The mechanism is ledger-readable: the wall bounces beam quanta on the approach side; bounced carriers re-approach and fringe-park on the lit face → near-face accretion → upstream centroid march. So the wall doesn't reverse a force — there is no force to reverse. It converts ablation-erosion into lit-face accretion, and drift direction is morphology-dependent: M49's grown-frozen remnant deposited on the far rim (+0.23 away), the uniform seeded sphere deposits on the near face (−0.50 toward). Even the open-gate baseline is morphology-dependent — the grown remnant pulled (−0.445), the seeded sphere doesn't (~0). "Ablation drift" names a family of deposition patterns, not a law of motion.

B. The wound vault (wound-wall.mjs). The IREFL × IDREST cell — previously impossible, since the wall was blind to directional rest. Seeded wound lump (R5, Lz_park=154,624):

config	Lz_park after 800t	parked mass	outcome
no beam (either wall)	154,624 (−0%)	39,680	perfectly inert fossil
open + beam	35,949 (−77%)	14,233 (−64%)	ablation; spin spreads to outer shells; +0.53 downwind walk
IREFL=32 + beam	151,669 (−2%)	44,248 (+11%)	spin vault

Behind the wall the beam becomes a slow feeder: refl=8,925 events refuse the approach, survivors park on the lit face into the winding — accreted mass joins the same shells (r3/r4), spin diffuses nowhere. The first object that is simultaneously a memory device and a beam-eater.

Caveats: seeded sphere ≠ grown remnant (controlled comparison, not a rerun); vault measured at one geometry/density.

M51. The morphology axis (morphology-drift.mjs)

M50-C left a binary contrast (seeded vs grown). This sweep resolves the axis: fixed R5 geometry, seeded density {16,32,48,96} + hollow shell (rim 96 / core 8) + the grown-remnant endpoint; each under IREFL {0,32}, +x beam 800t (results/m42-45/morphology-drift.log).

• Open gate: every seeded morphology is drift-inert (−0.15…0.00); only the grown remnant pulls (−0.44, third reproduction of M36/M49's −0.445). The ablation pull is a property of grown internal structure, not generic beam ablation.
• Walled + seeded: lit-face accretion produces an upstream march that is density-windowed — Q16 −0.25, Q32 −0.71, Q48 −0.54, Q96 −0.20, shell −0.23. Too thin: little to deposit; too dense: the saturated face has no fringe to park into.
• Walled + grown remnant: drift decays to ≈0 sustained (−0.05) — M49's +0.23 "push" was an early-beam transient overshoot, not a regime. The wall neutralizes the grown pull rather than reversing it.

Corrected statement for M49-C: the wall converts ablation-erosion into accretion; where mass deposits (lit face vs far rim) — and hence which way the centroid marches — is set by morphology × gate × time. Lump motion is not an object property; it is a phase point.

Milestone Index

Numbering note: the meso campaign continued the sequential ledger, so the M-numbers below overlap the retracted single-spin M1–M31 range. Every index entry here refers exclusively to meso3i experiments; same-numbered planck3 results are the ones marked retracted in the ledger.

M	Finding	Commit
M19–M21	Integer lattice gas: conservation, ergodicity, dwell v=1/k, no confinement	bc3e192–f6353a0
M22–M24	Q-register bound matter, conveyor, Pick-Two, peristalsis	29515b3–f33630f
M32	Accretion probe: native transparent; Q condense+emit = breathing core	535970f
M32b	No jets; core size scales with drive to u8 cap	554c32d
M32c–d	Arrow taxonomy: equilibrium/mirror/leak. Accretion needs one-way	8dbdbb6, fa0ee87
M33	Transit friction → channel-native breathing core	08193b2, corrected cececd9
M34	Remnant decay = pure exponential, no metastable floor	2f28e7b
M35	Carrier-gated radiation → persistent re-wakeable matter	6c23a03
M36	Ablation drift: lump migrates toward beam, hemisphere-imaged	cb41b20, f95706a
M37	Weak short-range mutual approach under self-radiation, transient: see M47b	5a7f8b2, 4540cc1
M38	No internal structure, rocks not atoms	94cbebb
M39	Massless photon plane; catalyst/stimulated/absorbed coupling	b3fa609, 1b1f88c
M40	Pair wind/unwind: binders are carriers; entangled gate	10d521e
M41	Transit-latency gate → activation barrier, self-shielding	bf737da
M42	Directional rest rest[18] stores Lz; sign-true revival	a622a8b
M43	Grazing collisions and absorbed light write spin (YORP)	1b1f88c
M44	No live circulation: bucket brigade subcritical, all regimes	a622a8b
M45	Spin induction by contact; internal counter-rotation annihilates, replicated lamp-free	1b1f88c
M46	Πᵢⱼ: P/ρ→0.879 closed-box EOS (collisions generate isotropy); wound shear in rest bins	1204e4f
M47c	Coherent-beam theorem: {c,c} has no momentum-conserving scatter. Unidirectional flow is permanently frozen	9cd561f
M48	Wave attenuation: zero in vacuum; turbid (~200c λ) below congestion, opaque (~25c) above	55d2b02
M47	Transaction ledger: driven lump ~3.7k events/20t; vacuum wound state exactly silent	1204e4f
M47b	M37 resolved: real transient ablation coupling only between self-radiating lumps; bath & broadside null; drive-infall artifact caught	e002c9d
M49	Capture gate is a phase parameter: flat-p band / ramp→dissolve-or-fossil; IREFL builds a hollow-core crust and flips ablation drift (grown remnant — see M50)	e70fe6e
M50	iR/drest fix + M42–45 re-verified; drift is morphology-dependent (mass-matched: wall→lit-face accretion, upstream march); IREFL×IDREST spin vault	0946f9c, 6282abb
M51	Morphology axis resolved: M36 pull is grown-structure-specific; wall gives density-windowed upstream accretion on seeded lumps (peak Q32); M49 "push" was a transient	480fd20
E-series	iE excitation register (Planck's n): mass↔photon↔n fully reversible; three heat channels measured (radiation E19b, convection E21, conduction E23/E24); First Law + dynamic photon gas + Fourier diffusion verified; E25 clock excitation-invariant; E26 light blind to light — emergent c invariance on the shared photon plane	6d5d7f1

The Picture That Emerges

A discrete substrate on which space is friction produces, in order:

1. Drag turns flux into parked heat (accretion).
2. Inertia keeps parked heat out of collisions (a mass state).
3. Emission cost makes radiation require a carrier (metastability + environmental coupling).
4. The result: bound matter that breathes while fed, sleeps in vacuum, wakes under light, migrates under anisotropic flux, and weakly attracts its neighbors through mutual radiation.

All of it lives on 19 integer channels per node, one rest bin, and two rules. All of it survives honest audit. The structured-matter rung turned out to be a state-representation question. Answered on loom-meso by giving rest mass a channel identity (Part VIII): the result is matter that stores spin, acquires it from collisions and light, and provably cannot circulate it forever.

---

Replay: The Life of a Lump

A recorded run, real telemetry, not an animation mock: test_replay ran the canonical sequence (drive → cutoff → quiet → one-sided beam) and dumped every frame. Watch the convergent spokes feed a breathing amber core, the core freeze dead when the drive cuts at t=200, sit inert for 400 ticks of vacuum quiet, then creep toward the +x carrier beam at t=600 while its far side evaporates.

DRIVEN — accreting while venting

t=0 · rest 0q

moving flux parked heat rest-mass centroid

Or run it yourself. The same engine, live in your browser: Joule's Mill compiles meso3i to WebAssembly with the full measured rule set. Directional rest, photon coupling, pair rules, latency, and one-click milestone scenes. (The earlier 44³ build, Yarkovsky's Ember, is preserved as a milestone.)

M52. The transition ledger: the substrate as a reaction network

The E46–E51 arc revealed that the engine's knobs are not independent mechanics — they are competing branches at shared decision points. The honest decomposition, extracted from the code itself:

Fork 1 — the park fork (every arriving mass quantum):

moving[c] → rest/drest/iE at destination   (IFRIC park, roll < fricp_num)
          → phot[c] at destination          (IRECOMB radiative branch)
          → moving[c^1] at source           (IREFL congestion bounce)
          → moving[c] at destination        (free flight)
park rate = f(iR density) − IFRLUM·luminosity — the branches SHARE the roll

Fork 2 — the photon-arrival fork (every photon):

phot[c] → absorbed: parked → iE             (IPHOT≥5, per-quantum roll)
        → ablation: parked → moving[c] mass (IPHOT 2–4)
        → coherence-barrier refuse          (IPTB gate on iPT)
        → free stream                        (invariant c)

The rest-stock ledger (what a lump holds, why it balances):

inflow:   γ+γ → drest[c]+drest[c^1]   (fusion, iPT≥IFCAD gated, ν≥IFUS)
outflow:  drest pairs → γ+γ            (annih, stock drain ∝ pairs)
          parked → moving mass         (boil-off, mass conserved)
          parked → γ (IPHOT≥4 variant) (fluorescence unwind)

The cadence registers (detection, not storage): iPT photon tempo (the fusion energy gate), iLT mass tempo (emission barrier). Energy is measured tempo, not stored attribute — E45's core finding.

Consequence: parameter sweeps are not identifiable per-knob — the branches share decision points, so constants are coupled rates in one reaction network. The E50 equilibrium (annih=64 → ~20k plateau, plateau ∝ feed) is the first measured balance of the stock ledger; the E51/E52 negatives (no Eddington ceiling) locate the missing nonlinearity: there is no blow-off channel — deflected mass reroutes, never leaves. Mastering the substrate = mastering this fork structure, not tuning knobs.

M53. The routing substrate: transactions as handshakes, Eddington as link congestion

The E46–E53 arc's conceptual capstone — the substrate is a network with reliable delivery, and its "missing physics" is vocabulary we already had:

• Quanta = the budget (conserved payload); photons = the transfer mechanism (offer packets). Absorb/fuse rolls are the ACK handshake — energy only moves on confirmation; pass-through and IREFL are declined offers (Schlatter–Kastner transaction ontology, J. Phys. Commun. 7 065009).
• Friction = congestion control; iR density = link load; friclum = back-pressure; dwell lines = store-and-forward.
• The Eddington limit = link-capacity contention: outgoing luminosity and incoming feed travel independent channels today — they never share bandwidth, so accretion never chokes. The emergent fix: a shared per-channel throughput budget (phot_out + mass_in ≤ capacity) makes a shining lump squeeze its own inflow — the ceiling emerges as congestion collapse, not as an authored force balance. Radiation pressure is link saturation.
• The E54 self-loop finding reframes: fusion↔annih is a closed detailed-balance loop — net conversion only exists when photons leave the network (drain = the open boundary / cold reservoir).