measured | Current LeoAware v3.17 FillGap | Crest starlink_v1 ACCEPT, not Current

Full-speed Starlink

Name: OrbitStack
Author: Pitchfork-and-Torch

Built for engineers shipping real traffic over LEO - including X and other high-volume apps on Starlink-class paths. LeoAware is an endpoint congestion controller that invalidates stale RTT/BW samples after reconfigs. ASCENT-D optionally feeds integrity-protected satellite state (erase-on-fail). Optional OrbCC-style pathID signals for hybrid assist.

See measured numbers Open-source code How ASCENT helps Why this matters at X

Current LeoAware v3.17 FillGap | starlink_v1 multi-seed mean 82.45 Mbps p95 76.26 ms | vs BBR 82.44 / 76.66

LeoAware v3.9 Crest | starlink_v1 ACCEPT, not Current 82.07 Mbps p95 76.26 ms | clears 75 / 138.8

BBR approx on the same starlink_v1 path 82.44 Mbps p95 76.66 ms | product-path reference

LeoAware v3.7 OCE | ope_v36, not Current 58.78 Mbps p95 152.09 ms | vs BBR 58.21 / 152.89

Source: open-source discrete-time sim (Pitchfork-and-Torch/leo-aware-transport). Do not mix eras. Current is v3.17 FillGap on starlink_v1 (seeds 13,7,42,99,123; 90s; endpoint-only). Absolute dual-gate: gp >= 75 and p95 <= 138.8. FillGap 82.45 / 76.26 vs BBR 82.44 / 76.66. Terr 79.05. Soft-QIR stays 0.20 / 25 ms. Constructor defaults stay use_fill_gap=False / use_openslot=False. SHA 781bd77 (PR #22). Docs 58b6e28 (PR #24). v3.9 Crest on the same path is ACCEPT, not Current (82.07 / 76.26). v3.7 OCE on ope_v36 is not Current (58.78 / 152.09 vs BBR 58.21 / 152.89). WetLinks v3.11 is a separate measured-CSV era (156.70 / 63.98 vs BBR 161.91 / 64.38). Not Current. v3.13 leocc_v1 is a research era (337.97 / 89.60 vs BBR 379.80 / 89.60). Not Current. Coupled-RNG 75/138.8 numbers from v3.4/v3.5 are historical only. Not dish PHY Mbps. Reproduce: python3 -m experiments.run_starlink.

OrbitStack editorial LEO mesh: constellation path control to edge node — Landing art - LEO mesh + edge path (editorial) full frame | no crop

01 / measured

Actual suite output (not marketing placeholders)

Same path model for all three CCAs. Handovers redraw RTT and capacity and inject non-congestive loss bursts. The stress case is frequent reconfiguration - the failure mode Starlink-class users hit when CUBIC treats mobility loss as queue overflow. Tabs below show algorithm evolution (scientific honesty): Current is v3.17 FillGap on starlink_v1. Crest, OCE, WetLinks, and leocc_v1 are other eras, not Current. Coupled-era absolute Mbps (v3.4 / v3.5) remain as history and are not comparable.

Primary objective: starlink_v1 multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
CUBIC	8.57 Mbps	71.63 ms	baseline
BBRv3approx	82.44 Mbps	76.66 ms	same starlink_v1 orbit
LeoAware v3.17 FillGap	82.45 Mbps	76.26 ms	gp >= 75; p95 <= 138.8; edges BBR
LeoAware v3.9 Crest (prior lock)	82.07 Mbps	76.26 ms	ACCEPT, not Current
Geometry (oracle)	84.03 Mbps	70.79 ms	path ceiling, not a CCA

Per-seed FillGap (means are the gate)

Seed	Goodput	p95
7	75.36 Mbps	67.81 ms
13	96.80 Mbps	72.21 ms
42	81.25 Mbps	97.56 ms
99	73.19 Mbps	64.09 ms
123	85.61 Mbps	79.65 ms

Product lock on starlink_v1. FillGap is a delay-clean 1 MSS cwnd fill when delivery is caught and cwnd is under 0.85x delivery BDP. It does not retune OpenSlot 0.80 and never gates ep:loss_burst. Constructor defaults stay False. Seed 99 is below 75; the published gate is the five-seed mean. Archive: results/archive/20260814-v317-fillgap/. Design: leoaware_v317_fillgap.md | SHA 781bd77 (PR #22) | docs 58b6e28 (PR #24) | eras: harness_eras.md

Terrestrial control

CCA	Goodput mean	p95
LeoAware v3.17 FillGap	79.05 Mbps	46 ms

Terrestrial gp>=77 PASS. Soft-QIR sojourn is in the p95 (path 40 ms + sojourn), not a path-only 40 ms floor. Reproduce: python3 -m experiments.run_starlink.

Primary objective: starlink_v1 multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
BBRv3approx	82.44 Mbps	76.66 ms	same starlink_v1 orbit
LeoAware v3.9 Crest	82.07 Mbps	76.26 ms	gp ≥ 75; p95 ≤ 138.8
Geometry (oracle)	84.03 Mbps	70.79 ms	path ceiling, not a CCA

Per-seed Crest (means are the gate)

Seed	Goodput	p95
7	75.01 Mbps	67.81 ms
13	96.65 Mbps	72.21 ms
42	80.67 Mbps	97.56 ms
99	72.83 Mbps	64.09 ms
123	85.17 Mbps	79.65 ms

Product lock on starlink_v1 (cruise RTT 40-75 ms, HO spike in the 0.4s loss window, cap 40-150 Mbps). Crest does not beat BBR on mean goodput (82.07 vs 82.44). It does clear the absolute bars that OPE-fair cannot reach (oracle gp 60.48 on ope_v36). Seed 99 is below 75; the published gate is the five-seed mean. Archive: results/archive/20260812-v39-starlink-v1/. Design: leoaware_v39_starlink_v1.md | eras: harness_eras.md

Terrestrial control

CCA	Goodput mean	p95
LeoAware v3.9	78.62 Mbps	46 ms

Terrestrial gp≥77 PASS. Soft-QIR sojourn is in the p95 (path 40 ms + sojourn), not a path-only 40 ms floor.

LeoAware v3.13 leocc_v1 | research era | not Current | LeoAware 337.97 / 89.60 vs BBR 379.80 / 89.60 vs CUBIC 31.14 / 87.20 | ACCEPT_ERA_REJECT_BBR

leocc_v1 three-CCA means (not Current)

CCA	Goodput mean	p95 mean	Gate
CUBIC	31.14 Mbps	87.20 ms	baseline
BBRv3approx	379.80 Mbps	89.60 ms	ahead on goodput
LeoAware v3.13 leocc	337.97 Mbps	89.60 ms	absolute 75 / 138.8 PASS

Research era. Label ACCEPT_ERA_REJECT_BBR, not Current. Absolute 75 / 138.8 PASS. Terrestrial 78.62 PASS. leocc does not clear BBR (337.97 vs 379.80). Do not mix with WetLinks 156.70 / 63.98 or v3.9 Crest 82.07 / 76.26.

Terrestrial control

CCA	Goodput mean	Gate
LeoAware v3.13	78.62 Mbps	terr PASS

LeoAware v3.11 WetLinks | measured 5x90s CSV | gp 156.70 / p95 63.98 vs BBR 161.91 / 64.38 | ACCEPT era | not Current

Hold-expanded WetLinks windows (net_iperf + net_ping)

CCA	Goodput mean	p95 mean	Notes
CUBIC	38.46 Mbps	63.58 ms	same 5 windows
BBRv3approx	161.91 Mbps	64.38 ms	ahead on goodput
LeoAware Crest replay	156.70 Mbps	63.98 ms	gp ≥ 75; p95 ≤ 138.8
Geometry (oracle)	244.85 Mbps	60.78 ms	UDP iperf mean, not dish PHY

New era. Do not mix with Crest 82.07 / 76.26 or OPE-fair 58.78 / 152.09. Product replay uses dt=0.01 and a 250 KB / 200 Mbps send ceiling (w1/w2 sit on that ceiling). An uncapped-buffer cook reached LeoAware 239.72 vs BBR 240.48 and was REJECT (does not clear BBR). Archive: results/archive/20260813-v311-wetlinks/. Spec: leoaware_v311_wetlinks.md

Primary objective: OPE-fair multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
CUBIC	5.56 Mbps	130.4 ms	baseline
BBRv3approx	58.21 Mbps	152.89 ms	OPE-fair reference
LeoAware v3.7 OCE	58.78 Mbps	152.09 ms	gp ≥ BBR; p95 ≤ BBR (+0.57 / -0.80)
LeoAware v3.6 Keel (prior)	58.27 Mbps	152.09 ms	first OPE-fair dual-gate pass
LeoAware v3.5 Tide (coupled historical)	76.27 Mbps	147.39 ms	coupled-RNG era; not comparable

Novel: OCE arms a ~3 RTT echo after SER - chase delivery into bw_est / 1.42×BDP, roll back if delay_ratio > 1.30. SER-lite: ack_ia+loss_burst without rtt_mad/loss_rtt keeps min_rtt (cut 0.80). Builds on v3.6 Soft-QIR, Keel, 2PC TBPR, SER. Hybrid fuse rails unchanged (endpoint suite default). Archive: results/archive/20260812-v37-oce/. Integrity green. Design: leoaware_v37_oce.md

Terrestrial control

CCA	Goodput mean	p95
LeoAware v3.7	~78.65 Mbps	46 ms

Terrestrial gp≥77 PASS. Delay includes soft-QIR sojourn term (not path-base-only).

LeoAware v3.6 Keel | OPE + Soft-QIR + SER + 2PC | first OPE-fair dual-gate | gp 58.27 / p95 152.09 vs BBR 58.21 / 152.89

OPE-fair multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
CUBIC	5.56 Mbps	130.4 ms	baseline
BBRv3approx	58.21 Mbps	152.89 ms	OPE-fair reference
LeoAware v3.6 Keel	58.27 Mbps	152.09 ms	gp ≥ BBR; p95 ≤ BBR (thin margin)

Integrity fix: separate loss_rng so path identity is CCA-independent (OPE). Soft-QIR exposes queue without letting a full buffer dominate p95. Keel anchors delay across epochs; SER keeps min_rtt on pure ep:loss_burst; 2PC TBPR rolls back on delay spikes. Coupled-era absolute bars retired as primary gate. Design: leoaware_v36_keel.md

LeoAware v3.4-p95 | coupled-RNG era HISTORICAL | multi-seed endpoint gp 73.57 (beats BBR 70.88) | p95 138.37 under BBR 138.8 | terrestrial 78.20 @ 40 ms

Historical (coupled-RNG): multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
CUBIC	5.47 Mbps	124.8 ms	baseline
BBRv3approx	70.88 Mbps	138.8 ms	coupled-era reference
LeoAware v3.4-p95	73.57 Mbps	138.37 ms	coupled-era: gp > BBR; p95 under BBR
LeoAware v3.3-A (prior)	78.06 Mbps	149.7 ms	higher gp; p95 residual
LeoAware v3.1 (prior)	68.98 Mbps	133.6 ms	historical p95 under BBR

Not comparable to OPE-fair numbers. Coupled-era mobility-loss draws advanced the same RNG as path dynamics, so each CCA saw a different orbit. Absolute bars (gp≥75 / p95≤138.8) are retired. v3.5 Tide later closed gp≥75 on that same coupled sim (76.27 / 147.39) - also historical. Levers: earlier cruise delay_yield, soft max-filter under delay, lower REPROBE fill ceiling. Archive: results/archive/20260812-p95-reclaim/.

Terrestrial control (coupled-era)

CCA	Goodput mean	p95
BBRv3approx	78.81 Mbps	40.0 ms
LeoAware	78.20 Mbps	40.0 ms

LeoAware v3.3-A | hybrid fuse + ASCENT-D integrity | multi-seed endpoint gp 78.06 (beats BBR 70.88) | p95 149.7 residual | hybrid ablation p95 109.6

Historical: multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

CCA	Goodput mean	p95 mean	Gate
CUBIC	5.47 Mbps	124.8 ms	baseline
BBRv3approx	70.88 Mbps	138.8 ms	reference
LeoAware v3.3-A	78.06 Mbps	149.7 ms	gp beats BBR; p95 residual

Hybrid fuse ablation (fast, seeds 13+7) - fuse accept record

Variant	leo_fast_ho gp	p95	Note
endpoint	84.04 Mbps	123.9 ms	no assist
hybrid (D+Orb fuse)	75.57 Mbps	109.6 ms	no double-cut; p95 under endpoint
orb	72.99 Mbps	116.9 ms	pathID only
ASCENT-D noisy	= endpoint	= endpoint	erase-on-fail proven

Hybrid fuse: Orb pathID REPROBE suppressed during ASCENT assist/freeze/REPROBE (2.0s); never util-MD when path hints on. Integrity: python -m experiments.test_ascent_d_integrity. Design: leoaware_v33_hybrid_fuse.md

Terrestrial control

CCA	Goodput mean	p95
BBRv3approx	78.81 Mbps	40.0 ms
LeoAware	77.43 Mbps	40.0 ms

LeoAware v3.1 | path-aware BDP + max-filter reclaim | multi-seed mean gp 68.98 / p95 133.6 (under BBR 138.8)

Endpoint multi-seed gate accepted in v3.1; v3.2 added ASCENT-D / Orb control-plane rails (no standalone tab - see Multi-seed history). Superseded by later reclaim / OPE work.

CCA	Goodput mean	p95 mean
BBRv3approx	70.88 Mbps	138.8 ms
LeoAware v3.1	68.98 Mbps	133.6 ms

LeoAware v3 | prior-BW seed + dual-signal gate + fair_mode + freeze | multi-seed mean gp 66.4 / p95 133

Multi-seed leo_fast_ho (accepted v3)

CCA	Goodput mean	p95 mean
BBRv3approx	70.88 Mbps	138.8 ms
LeoAware v3	66.44 Mbps	133.2 ms

Superseded by v3.1 (+2.5 Mbps multi-seed mean under same p95 gate).

v2 | multi-signal reconfig + two-phase REPROBE | suite seed 13

Fast handovers

CCA	Goodput	p95 RTT
BBRv3approx	65.38 Mbps	188.6 ms
LeoAware	70.36 Mbps	123.2 ms

Strong single-seed result; multi-seed mean goodput 64.8 with p95 146 (above BBR p95 mean).

v0 launch baseline | soft re-probe + mobility taxonomy (pre multi-signal v2) | suite seed 13

Fast handovers (`leo_fast_ho`)

CCA	Goodput	Avg RTT	p95 RTT	Loss	HOs
CUBIC	6.44 Mbps	65.4 ms	134.3 ms	0.14%	7
BBRv3approx	65.38 Mbps	116.2 ms	188.6 ms	1.18%	7
LeoAware	68.31 Mbps	87.3 ms	164.0 ms	1.26%	7

Single long flow (`leo_single`)

CCA	Goodput	p95 RTT	Loss
CUBIC	9.04 Mbps	111.2 ms	0.10%
BBRv3approx	83.99 Mbps	161.8 ms	1.27%
LeoAware	64.67 Mbps	129.1 ms	1.32%

Terrestrial control

CCA	Goodput	p95 RTT
BBRv3approx	78.81 Mbps	40.0 ms
LeoAware	77.98 Mbps	40.0 ms

Scientific record | multi-seed means (seeds 13,7,42,99,123) | coupled-era vs OPE-fair eras marked

leo_fast_ho multi-seed mean (primary objective)

Version	Goodput mean	p95 mean	Era / vs BBR	Notes
BBRv3approx (coupled)	70.88 Mbps	138.8 ms	coupled reference	pre-OPE
BBRv3approx (OPE-fair)	58.21 Mbps	152.89 ms	OPE + soft-QIR reference	identical orbit timeline
LeoAware v2	64.82 Mbps	146.5 ms	coupled; above BBR p95	prior best single-seed peaky
LeoAware v3	66.44 Mbps	133.2 ms	coupled; under BBR p95	+1.6 Mbps vs v2
LeoAware v3.1	68.98 Mbps	133.6 ms	coupled; under BBR p95	+2.5 Mbps vs v3
LeoAware v3.2 rails	(assist mixed)	(assist mixed)	integrity gate	ASCENT-D + Orb modules (no standalone tab)
LeoAware v3.3-A	78.06 Mbps	149.7 ms	coupled; gp beats BBR	hybrid fuse; p95 residual
LeoAware v3.4-p95	73.57 Mbps	138.37 ms	coupled; under BBR p95	gp > BBR; delay reclaim
LeoAware v3.5 Tide	76.27 Mbps	147.39 ms	coupled HISTORICAL	gp≥75 stretch; p95 residual
LeoAware v3.6 Keel	58.27 Mbps	152.09 ms	OPE-fair; dual-gate pass	first fair gp≥BBR and p95≤BBR
LeoAware v3.7 OCE	58.78 Mbps	152.09 ms	OPE-fair, not Current	relative-BBR dual-gate (+0.57 / -0.80)
LeoAware v3.9 Crest	82.07 Mbps	76.26 ms	starlink_v1 ACCEPT, not Current	absolute 75 / 138.8 ACCEPT
LeoAware v3.17 FillGap	82.45 Mbps	76.26 ms	starlink_v1 Current	edges BBR 82.44; absolute 75 / 138.8 PASS
LeoAware v3.13 leocc	337.97 Mbps	89.60 ms	leocc_v1 research, not Current	ACCEPT_ERA_REJECT_BBR
LeoAware v3.11 WetLinks	156.70 Mbps	63.98 ms	wetlinks_v1 CSV	measured batch; not Current

Capabilities (v3 through v3.17)

CSV trace replay - Starlink-like t_s,rtt_ms,capacity_mbps,reconfig paths.
fair_mode - multi-flow delay AIMD friendliness.
ASCENT freeze windows - hold + honest next_capacity peek + REPROBE.
v3.1 path-aware BDP - soft min_rtt age + max-filter reclaim (multi-seed win).
v3.2 ASCENT-D - RS(255,223) erase-on-fail path hints; never act on corrupted control.
v3.2 Orb hybrid - optional pathID assist surface.
v3.3-A hybrid fuse - single REPROBE ownership; no stacked ASCENT+Orb soft cuts.
v3.4-p95 reclaim - delay-aware cruise/fill; coupled-era p95 under BBR while gp still beats BBR.
v3.5 Tide - time-bounded post-hop reclaim; coupled-era gp≥75 stretch (historical only).
v3.6 Keel / OPE - Orthogonal Path Entropy + Soft-QIR + Keel + SER + 2PC TBPR; fair dual gate.
v3.7 OCE - Orbit Capacity Echo + SER-lite; ope_v36 relative dual-gate vs BBR, not Current.
v3.9 Crest - starlink_v1 ACCEPT, not Current; first absolute 75 / 138.8 ACCEPT.
v3.11 WetLinks - 5 measured CSV windows; 156.7 Mbps batch (era separate, not Current).
v3.13 leocc - leocc_v1 research era; ACCEPT_ERA_REJECT_BBR, not Current.
v3.17 FillGap - delay-clean 1 MSS cwnd fill; Current starlink_v1 dual-gate vs BBR (constructor default False).

Log: experiment_log.md | leoaware_v37_oce.md | traces_and_ascent.md

02 / problem

What breaks on Starlink-class paths

Peak bandwidth is fine. The path is not. Handovers, beam switches, ISL re-routes, and non-congestive loss look like congestion to loss-based CCAs and leave model-based CCAs with stale min-RTT and BDP. Interactive and bulk both suffer - latency variance and self-limited goodput.

What most stacks assume	What OrbitStack does
Stable path, stable RTT Mobility loss -> cut cwnd Keep min-RTT across hops Recover slowly after every reconfig	Reconfiguration-aware control Detect path change at the endpoint Discard stale samples; soft re-probe Taxonomy: mobility loss != congestion

03 / ASCENT

Wire state the CCA can actually trust

Endpoint detection alone is deployable today. ASCENT is the upgrade path: a greppable, integrity-protected stream for telemetry, agent roles, and multimodal refs - so Starlink-class terminals or edge agents can announce reconfigs and capacity instead of forcing the sender to discover them after the RTT spike.

ASCENT/1.0 # approach vector locked alt=120km status=nominal role:pilot role:untrusted:external-telemetry cid:sha256:deck-camera-03

Sacred ASCII stays greppable. ASCENT-D can wrap critical feedback with RS(255,223) erase-on-fail so the CCA never acts on corrupted control.

state

Predict, do not only react

Altitude, status, vector into the bottleneck model. Freeze/ramp around known windows before p95 blows out.

control

Protected feedback

Capacity estimates and reconfig notices in ASCENT-D. Erase on parity failure. No thrash on garbage state.

efficiency

Lower offered load

Content-addressed REFs and lightweight sacred control free headroom for goodput on the same window.

04 / design

Endpoint-first, quiche-shaped

No network cooperation required to start. Interface maps cleanly onto a QUIC congestion controller (on_ack, on_loss(congestive?), optional on_path_hint / ASCENT-D ingest, optional on_orb_signal). That is the integration surface for Cloudflare-class edge stacks and client-side mobile/desktop transports.

01

LeoAware detection

RTT jump outliers, ACK inter-arrival gaps, loss bursts without RTT inflation. Soft re-probe - not CUBIC collapse. SER / SER-lite keep min_rtt on RTT-stable mobility.
02

Optional ASCENT-D side channel

Subscribe to pilot / dish / gateway agent frames wrapped in ASCENT-D (erase-on-fail). Cross-check untrusted roles. Optional Orb pathID when in-network telemetry exists.
03

Adaptive profiles

Nominal -> lower ECC, higher probe headroom. Degraded geometry -> stronger protection + QUEUE coordination. OCE chases capacity only inside a short echo window.
04

Measure on real ASNs

A/B CUBIC / BBR / LeoAware on Starlink-attached clients or edge PoPs. Endpoint-only first; path hints when available.

05 / audience

For X engineers (and anyone on LEO)

If you run interactive product, media, or bulk sync over Starlink-class last miles, classic CCA assumptions leak into user-visible latency and goodput. OrbitStack is a concrete brief plus measured sim evidence for two complementary pieces:

LeoAware - ship endpoint logic that does not self-limit on mobility loss (v3.17 FillGap Current on starlink_v1 82.45 / 76.26 vs BBR 82.44 / 76.66; v3.9 Crest 82.07 / 76.26 is ACCEPT, not Current; v3.7 OCE 58.78 is ope_v36, not Current).
ASCENT-D - integrity-protected structured state so reconfigs are signals, not mysteries (never act on garbage).
VELA - the language that builds onto LeoAware. hint.ascent is Option (erase-on-fail). Missing hint is None, not a hop. Observe-only Reach is the no-regress wrap. Not a rival CCA.
Orb hybrid - optional pathID / queue assist when programmable telemetry exists; degrades to endpoint otherwise.
Honest scope - research sim and design, not a claim of production BBRv3 parity or official Starlink affiliation. Coupled-era absolute Mbps are historical only. Not dish PHY Mbps.

Natural pilot: Starlink-attached clients or edge terminations A/B'ing rate controllers, with ASCENT frames as a later assist. Related: ascent.jonbailey.xyz (wire) and skycache.jonbailey.xyz (mesh / DTN).

GitHub: leo-aware-transport Explore ASCENT hello@jonbailey.xyz @suddenlyjon

06 / path

Cloudflare x Starlink collaboration shape

Cloudflare: high-volume QUIC and global edge. Starlink: LEO topology and potential path-change signals. OrbitStack is the bridge brief - endpoint measurement first, authenticated reconfiguration hints later. Multipath and AI-scale bulk are phase-two traffic classes once the single-path rate loop is honest about handovers.

phase 0

This brief + sim

Shared vocabulary, metrics, and LeoAware behavior under synthetic LEO dynamics.

phase 1

Live A/B

Instrument Starlink ASNs / PoPs. Validate which endpoint signals fire around real hops.

phase 2

ASCENT assists

Optional epoch markers and capacity advice into on_path_hint without forcing cooperation.

Full-speed Starlink

Actual suite output (not marketing placeholders)

Primary objective: starlink_v1 multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Per-seed FillGap (means are the gate)

Terrestrial control

Primary objective: starlink_v1 multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Per-seed Crest (means are the gate)

Terrestrial control

leocc_v1 three-CCA means (not Current)

Terrestrial control

Hold-expanded WetLinks windows (net_iperf + net_ping)

Primary objective: OPE-fair multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Terrestrial control

OPE-fair multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Historical (coupled-RNG): multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Terrestrial control (coupled-era)

Historical: multi-seed leo_fast_ho endpoint (seeds 13,7,42,99,123)

Hybrid fuse ablation (fast, seeds 13+7) - fuse accept record

Terrestrial control

Multi-seed leo_fast_ho (accepted v3)

Fast handovers

Fast handovers (leo_fast_ho)

Single long flow (leo_single)

Terrestrial control

leo_fast_ho multi-seed mean (primary objective)

Capabilities (v3 through v3.17)

What breaks on Starlink-class paths

Wire state the CCA can actually trust

Predict, do not only react

Protected feedback

Lower offered load

Endpoint-first, quiche-shaped

LeoAware detection

Optional ASCENT-D side channel

Adaptive profiles

Measure on real ASNs

For X engineers (and anyone on LEO)

Cloudflare x Starlink collaboration shape

This brief + sim

Live A/B

ASCENT assists

Primary objective: starlink_v1 multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

Primary objective: starlink_v1 multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

Primary objective: OPE-fair multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

OPE-fair multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

Historical (coupled-RNG): multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

Historical: multi-seed `leo_fast_ho` endpoint (seeds 13,7,42,99,123)

Fast handovers (`leo_fast_ho`)

Single long flow (`leo_single`)