Validation / Evidence Ledger

Evidence before claims.

Planetary Defences separates discovery benchmarks, research candidates, proxy validation, posture replay and external outcome validation. Results are not promoted across categories they have not tested.

Frozen benchmark ≠ future guaranteeProxy truth ≠ asset outcome truthEvidence class remains explicit
Current Evidence

One ledger.
Different evidence classes.

Each result below is stated with the status it actually earned. Open the methodology record for scope, exclusions and limitations.

Frozen benchmarkS1 / GEOMAGNETIC

Historical discovery

1,943 / 1,943

Availability-qualified events · 12 h median discovery lead.

ReplayS1 / 2026

Time-ordered replay

15 / 15

Stored events through the verified cutoff.

Research candidateS2 / LEO DRAG

Elevated-drag warning

187 / 193

96.89% · 240-minute median lead · research candidate.

Proxy validationS3 / RADIATION

Defined proxy truth

2,467 / 2,467

0 proxy false alarms. Asset-specific outcome validation is separate.

Outcome pendingS4 / GNSS · NAVIC

Historical replay

3.53M+

Historical snapshots replayed · receiver-level outcome validation remains pending.

Outcome pendingS5 / INFRASTRUCTURE

Historical replay

3.53M+

Historical snapshots replayed · sector-specific outcome validation remains pending.

S1Geomagnetic discovery benchmark+

The frozen retrospective discovery benchmark covers 1981–2026 and uses an availability-qualified truth set.

  • 1,943 availability-qualified events detected.
  • 12-hour median discovery lead.
  • Source-availability gating is explicit in the truth definition.
  • Unavailable evidence is not silently converted into either a detector success or a detector miss.
  • A separate action lane is evaluated independently from the discovery lane.
Discovery and action evidence remain separate categories. Protocol-level exclusions, event ledgers and action-lane details are available to qualified evaluators rather than exposed as implementation guidance on the public site.
S2LEO drag research candidate+
  • Candidate evaluated across 2001–2025.
  • 187/193 defined drag-risk events detected — 96.89%.
  • Approximately four-hour median warning in the stated research evaluation.
  • The candidate was selected using the broader historical dataset.
This is an all-data-selected research candidate, not a pure model-selection-blind result. Independent untouched validation is desired before stronger operational claims.
S3Radiation & charging proxy validation+
  • 2,467/2,467 defined proxy events detected.
  • 0 proxy false alarms.
  • Independent environmental proxy cross-checks were also passed within their stated truth definitions.
These results validate defined environmental proxy truth. They do not establish spacecraft-specific anomaly, charging or dose outcomes. External spacecraft telemetry and anomaly truth are the next validation stage.
S4GNSS / NavIC integrity replay+

Software is implemented and more than 3.53 million historical snapshots have been replayed.

Independent receiver-level outcome truth is still required, including signal-quality, lock, positioning and timing outcomes.

No receiver-event detection percentage is claimed.
S5Infrastructure & mission-risk replay+

Software is implemented and more than 3.53 million historical snapshots have been replayed.

Sector-specific outcome validation remains pending across relevant infrastructure and mission contexts.

No infrastructure-event detection percentage is claimed.
Controlled Evaluator Detail

Public proof should be inspectable.
Implementation detail should remain protected.

The public ledger states the result, evidence class, qualification and current validation status. Exact thresholds, state distributions, source-selection rules, event-by-event exclusion ledgers and internal algorithmic logic are reserved for controlled evaluation or NDA review.

Request evaluator detail →
Evidence Discipline

What this page is
designed to prevent.

Strong numbers become misleading when evidence classes are collapsed. The validation architecture keeps those boundaries visible.

01 / SCOPE

Truth stays named.

Proxy, event, replay and object-level outcome truth are not interchangeable.

02 / EXCLUSIONS

Missing data stays visible.

Availability exclusions are documented rather than silently counted as observed outcomes.

03 / TRANSFER

Promotion requires transfer evidence.

Research candidates remain clearly labelled until the declared external validation stage is completed.

04 / FUTURE

Benchmarks are not guarantees.

Historical evidence supports evaluation, not promises of identical future-event performance.

← Return to the operating system