Skip to content

SKYPULSE.md

Latest commit

 

History

History
152 lines (115 loc) · 6.05 KB

SKYPULSE.md

File metadata and controls

152 lines (115 loc) · 6.05 KB
Raw

ASCENT SkyPulse

Status: Additive (ascent-wire 2.1.0; Nexus architecture card 3.0.0 Crystal Wire). SPEC 1.0.0-rc1 parse law unchanged.
Unit: PATHHINT / SKYSTATE (0xC5, schema 0x01)
Profiles: usage note ASCENT-E-LEO vs ASCENT-D (not a new P0-P2 grammar)

ASCII hyphens only.

Honesty fence

ASCENT does not raise Starlink physical RF Mbps. SkyPulse is for usable session bandwidth, session continuity, and control overhead under LEO. Never market a bare "bandwidth upgrade."

What it is

SkyPulse is a path foresight unit for LEO IP sessions (Starlink and similar). It carries a compact, fail-closed hint. Normative field semantics:

Field Required Meaning
next_capacity yes Predicted bottleneck bits/s as seen by the sender. Not RF PHY rate.
freeze_until yes Growth-freeze window. v1 wire is relative freeze_ms (FLAG_RELATIVE_FREEZE). Semantic freeze_until = now + freeze_ms.
path_id / epoch yes Path or epoch. Unknown/new id replaces stale hint.
confidence yes [0, 1] blend weight.
ttl_ms yes Hint lifetime. Stale TTL => erase (consumer-side).
obstruction optional [0, 1] if present.
elev_deg optional Degrees if present.

Apps and congestion control can slow down, freeze CWND growth, or QUEUE instead of spraying retransmits across a 15-second LEO reconfig or a tree obstruction. That is usable session bandwidth and session continuity, not a faster radio.

What it is not (honesty fence)

Claim Allowed?
ASCENT increases Starlink physical downlink/uplink Mbps No
Bare "bandwidth upgrade" No - say usable session bandwidth
"Grok offline in space forever" No
PATHHINT is a measured RF throughput No - predicted sender bottleneck
Dual-gate LeoAware product wins No - see OrbitStack bridge
Replace Starlink networking / Dishy firmware No
Invented Starlink telemetry API No

Allowed: usable session bandwidth, fewer wasted retransmits, foresight for CCA/apps, LEO-efficient framing, fail-closed integrity.

Wire (additive, skip-by-length)

Assigned from reserved P2 leads 0xC4-0xCE. Cont class 0xA0-0xBF stays frozen. No P0-P2 parse-law break. SPEC stays 1.0.0-rc1 + this appendix.

0xC5                    # SKYSTATE lead
schema:u8               # 0x01 = PATHHINT v1
len:u16be               # body length (v1: 26 or 30 with CRC)
body

PATHHINT v1 body (26 bytes, +4 IEEE CRC-32 if FLAG_CRC):

flags:u8
path_id:u64be
next_capacity:u32be     # predicted sender bottleneck; bps, or kbps if FLAG_CAP_KBPS
freeze_ms:u32be         # relative freeze_until (FLAG_RELATIVE_FREEZE always set in v1)
confidence:u16be        # 0..10000 => [0.0000, 1.0000]
ttl_ms:u32be
obstruction:u8          # 0..255 => [0,1] if FLAG_HAS_OBSTRUCTION; else 0xFF
elev_deg_x10:i16be      # degrees * 10; 0x7FFF absent
[crc32:u32be]           # IEEE CRC-32 of the 26-byte prefix

Fail-closed (never act on corrupt units):

Condition Action
CRC fail or short-RS fail erase (applied=false)
Stale TTL erase (consumer evaluate_pathhint)
Unknown schema / version erase (skip-by-length)
Missing FLAG_RELATIVE_FREEZE erase (missing_freeze_until)
Reserved flags, bad length, confidence > 10000 erase
Truncation before len is readable hard-fault (SPEC stream rule)

Goldens: tests/skypulse_vectors.json. Canonical 50e6 bps is a hint, not RF.

LEO-IP vs ASCENT-D integrity (when each applies)

Do not stack CRC + RS. Pick one profile.

ASCENT-E-LEO (Starlink IP/QUIC) ASCENT-D
Parse law Same E grammar + PATHHINT Same + mandatory P9
Interactive Starlink IP Yes Poor fit (RS tax + latency)
Integrity Light: PATHHINT CRC (v1 shipped) or short RS(255,239) I=1 as a substitute, never stacked Full RS(255,223) P9 erase-on-fail
Double FEC Forbidden as required P9 is the outer ECC; skip extra PATHHINT CRC
Use Chat/agent turns, PATHHINT, QUEUE Spool files, deep-space, high BER

recommend_integrity("ASCENT-E-LEO") vs recommend_integrity("ASCENT-D") in ref/ascent_skypulse.py. Short RS is an allowed LEO-IP substitute, not a new v1 wire flag.

ASCENT ship success metrics

Separate from LeoAware dual-gate. Not "+X Mbps Starlink."

Metric v1 target
Goldens Py ≡ JS tests/skypulse_vectors.json + tests/run_js_lock.js
PATHHINT round-trip encode/decode fields match; PathHint.encode() == wire
Erase-on-fail CRC flip, unknown schema, missing freeze, stale TTL => applied=false
Overhead bytes/hint 30 plain, 34 with CRC
Apply vs reject lab samples: 2 applied / 2 rejected in test_ship_metrics_not_rf

Ship order

  1. SPEC appendix + goldens + ascent-wire 2.1.0
  2. Wire Lab panel (meters labeled sim)
  3. Publish PATHHINT codec (done: ascent-wire 2.1.0 on PyPI/npm)
  4. LeoAware later (OrbitStack product; not this PR)

Do not claim OrbitStack paid wins. Live site deploy still needs Jon approval.

Sacred-meter

Observational (obs) only when a real dish/API exists. Wire Lab and lab env overrides are sim. TCP :9200 is reachability, not a telemetry API.

Quick start

PYTHONPATH=ref python -c "from ascent_codec import canonical_pathhint_bytes, decode_stream; print(decode_stream(canonical_pathhint_bytes()))"
python -m ascent pathhint --profile ASCENT-E-LEO
python tests/test_skypulse.py
node tests/run_js_lock.js

Client: examples/ascent_starlink_client/daemon.py --pathhint

Implementers

Surface Location
Python wire ref/ascent_codec.py (encode_pathhint, decode_skystate)
Policy ref/ascent_skypulse.py (evaluate_pathhint, recommend_integrity)
JS site/public/ascent_codec.js / packages/ascent-js
SPEC SPEC.md appendix SkyPulse (append-only)
LeoAware docs/ORBITSTACK-LEOAWARE-BRIDGE.md (hybrid fuse)