Evidence · public orbital data · signed & verifiable

Independent re-screening of public conjunction data, signed

Vaelon independently re-propagated 25public conjunction events, this week's closest public predictions (SOCRATES, ranked by minimum range; 21 unique object pairs, one close-orbit pair recurs across several passes), from scratch, gated each with the real flight engine, and folded every verdict into one tamper-evident HMAC-SHA256 chain. No permission, no partner, no flight required, just public data and the same code that runs onboard. Because this set is selected closest-first, it demonstrates false-alarm reduction only, see the honesty note below and the unbiased live catalogue screen further down.

Download the Safety Case Summary (PDF)
25
conjunction events (21 unique pairs)
16
Vaelon BLOCK (closest approach < 5 km keep-out)
25
decisions signed, one chain
11/25
reproduced to sub-km
Signed audit head
b03d95e1f44b286af1c58d2f6f49bd9bc816933bb6de0c9ceb1ee430ada1f90f
key id vaelon-conjunction-screen-2026 · verifies the entire decision log; the key never appears in a record

Live catalogue screen: Vaelon finds the conjunctions itself

2026-08-04 13:45 UTC

No third-party conjunction list. Vaelon fetched 120 objects from the live public catalogue (CelesTrak GROUP=last-30-days (live)), propagated each with validated SGP4, screened all 7,140 pairs itself, gated every flagged approach, and signed each verdict into one chain. 1108approaches fell inside the 5 km keep-out and were flagged, predominantly recent launch-mate clusters still in formation (real proximity, correctly caught, not uncontrolled near-collisions).

120
live objects screened
7,140
pairs screened, real SGP4
1108
sub-keep-out, flagged + signed
1243 / 1248
ALLOW / BLOCK
signed head 6808ef79360a0cc023bcf4161fe098ee14661cdb8462dd0f9a2edf7c7ebe7614
Screened from the live public catalogue with the real Vaelon guard, signed and dated above. The screener (make screen-live) and the flight core are shared with evaluators and partners under NDA, where the signed head can be independently re-verified byte-for-byte against a second implementation.

What this set can and cannot show

These 25events are selected as this week's closest public predictions (SOCRATES, ranked by minimum range), so every one is an approach SOCRATES already flagged as near. That means this set can only demonstrate one direction: false-alarm reduction, where our fresh, from-scratch propagation downgrades a flagged conjunction (the ALLOWs). By construction it cannot show the more important opposite direction, catching a real approach a stale source called safe, because a missed detection can never appear in a closest-first sample. That unbiased direction is the live catalogue screen above: Vaelon screens all 7,140 pairs itself, with no third-party list, and finds the conjunctions on its own.

On the reproduction metric, |Δ| < 1 km between our independently-computed miss distance and SOCRATES's published minimum range, 11/25 of 25 events agree. That is a soundness check on the propagation + screening path against an independent authority, not a proof of correctness. Where the horizon is long or a payload has maneuvered, our propagation diverges from the stale snapshot, as expected (right).

How does it compare to SOCRATES?

Stratified by propagation horizon: ≤2 days (n=6) median |Δ| 0.8171 km, within 1 km on 4/6, but even here 2 of 6 diverge (one by ~9 km at a 1.1-day horizon), so agreement is not uniform; >2 days (n=19) median 3.2225 km, with a single outlier at 3963 km, a stale-element / maneuvered-payload case whose old TLE badly mispredicts the current pass, not a propagation error. This is the honest, expected behaviour of multi-day SGP4 conjunction prediction: reliable near-term, degrading with horizon and crossing speed. A soundness check, not a validation claim.

#PrimarySecondarySOCRATES (km)Vaelon DCA (km)Δ (km)horizon (d)Vaelonsig
1STARLINK-33894 [+] (63697)OBJECT F [-] (47937)0.01256.415+56.4036.32ALLOWef1ddb6d
2KUIPER-00174 [+] (65787)EGYPTSAT 1 [-] (31117)0.0150.818+0.8030.68BLOCKd0a47cb9
3IRIDIUM 160 [+] (43569)COSMOS 1680 [?] (16011)0.0202.712+2.6926.39BLOCK9ca559fc
4STARLINK-3983 [+] (52618)OBJECT B [+] (60013)0.025130.705+130.6802.65ALLOW21e64bd2
5YAOGAN-36 01A [+] (53943)STARLINK-37871 [+] (69549)0.0291686.645+1686.6163.54ALLOWd9a602e1
6YAOGAN-39 02C [+] (57890)TOMORROW-S11 [+] (67399)0.0298.912+8.8837.14ALLOW6aa5d4f4
7GOSAT 2 (IBUKI 1) [+] (43672)SL-14 R/B [-] (14552)0.0314.724+4.6936.22BLOCK6f8b0d5a
8SHIYAN-5 (SY-5) [+] (39455)FENGYUN 1C DEB [-] (31566)0.0330.190+0.1570.93BLOCK3815368f
9OBJECT K [+] (61903)OBJECT C [+] (62727)0.04212.829+12.7876.24ALLOW1af7ecf3
10STARLINK-34157 [+] (64084)COSMOS 2251 DEB [-] (35443)0.0430.079+0.0357.28BLOCKa8655088
11YAOGAN-28 [+] (41026)STARLINK-5124 [+] (54011)0.0470.560+0.5130.53BLOCKfad504fe
12SWARM A [+] (39452)YAOGAN-35 03C [+] (53320)0.05067.337+67.2875.03ALLOWf512ba29
13HULIANWANG JISHU SHIYAN 4A [+] (58691)DELTA 1 DEB [-] (7130)0.0503.272+3.2234.50BLOCK99d1f142
14STARLINK-30933 [+] (58410)BANDWAGON-4 OBJECT M [+] (66304)0.0526.668+6.6162.96ALLOW437d50d9
15OBJECT D [+] (66056)FLOCK 4H-8 [+] (66711)0.0532.951+2.8983.31BLOCKedc3000a
16GLOBAL-2 [+] (43812)SL-8 DEB [-] (1392)0.0540.269+0.2152.15BLOCK89f42605
17STARLINK-1308 [P] (45419)STARLINK-11084 [+] (59717)0.0549.360+9.3061.12ALLOW3c11c15f
18SUPERVIEW NEO-2 03 [+] (62079)SUPERVIEW NEO-2 04 [+] (62080)0.0540.152+0.0985.09BLOCKac6a89d8
19SUPERVIEW NEO-2 03 [+] (62079)SUPERVIEW NEO-2 04 [+] (62080)0.0550.161+0.1065.02BLOCK41ba599a
20STARLINK-30774 [+] (58625)FLOCK 4G-29 [+] (62642)0.0561.522+1.4660.31BLOCK2809b810
21SUPERVIEW NEO-2 03 [+] (62079)SUPERVIEW NEO-2 04 [+] (62080)0.0560.143+0.0875.15BLOCKedd2bcc1
22SUPERVIEW NEO-2 03 [+] (62079)SUPERVIEW NEO-2 04 [+] (62080)0.0590.171+0.1124.95BLOCK4fc333a1
23LEMUR-2-FURIAUS [+] (42840)STARLINK-1572 [+] (46060)0.0600.877+0.8170.29BLOCK7533c9b2
24SUPERVIEW NEO-2 03 [+] (62079)SUPERVIEW NEO-2 04 [+] (62080)0.0610.135+0.0745.22BLOCK56c9e20b
25STARLINK-3998 [+] (52610)TEVEL2-8 [+] (63239)0.0623962.847+3962.7856.55ALLOWb03d95e1

Object-name suffixes are CelesTrak operational-status codes: [+] operational · [−] non-operational · [P] partially operational · [?] status unknown.

Source: CelesTrak SOCRATES Plus, Top 25 by Minimum Range · this is a fixed snapshot dated 2026 Jul 24 18:19:29 UTC, not a live feed (the SOCRATES candidate list is a manual pull). It is here as an independent-propagation soundness check against SOCRATES; the current, self-refreshing evidence is the live catalogue screen above, which depends on no third-party list.

Reproduce:both objects' orbital elements are fetched from CelesTrak and re-propagated with the reference SGP4 library; the closest approach is recomputed from scratch and gated by vaelon-core (the real flight engine), which signs each decision. The signed head above verifies the whole log. Decision rule: for a ground re-screen the gate is a keep-out check on the computed closest approach, BLOCK when it is inside 5 km. So a fast crossing whose closest approach is well outside the keep-out is correctly ALLOWed, even where the source catalogue flagged it. Download the signed pack (JSON) →

horizon = days from the later element-set epoch to closest approach. Agreement with SOCRATES decays with horizon and crossing speed, as expected for multi-day SGP4 conjunction prediction.