Transition posture
What is built, what readiness that actually earns, and what it would take to field.
A demonstrator that will not state its readiness level is asking to be assessed generously. This page states it, including where the number is lower than the demonstration makes it look. Everything here is a planning framework offered for discussion — no program of record, no contract vehicle, and no acquisition decision is represented or implied.
Honest headline
Readiness at a glance
TRL 4Component validated in lab
TRL 6Target, relevant environment
BlockedValidation gate
0Real sensor feeds
The demonstration looks like TRL 5–6 because it runs live, at speed, on realistic
geometry. It is TRL 4. The difference is that every corpus is synthetic and no
component has met measured truth data. That gap is a data-access problem, not an
engineering one, and it cannot be closed by writing more software.
Capability metrics
Before this build, and after it
Measured against the site as it stood before this increment. Each "after" claim is demonstrable on a page of this site, not asserted.
| Metric | Before | After | Where to verify |
|---|---|---|---|
| Impact-point prediction | None — corridors terminated at synthetic endpoints | Reachable set with 50% / 90% containment and a modal cell | Warning Console |
| Flight model | Gaussian likelihood over heading and speed | 3-DOF point mass, RK4, US Standard Atmosphere 1976 | Warning Console · adl-physics.js |
| Concurrent tracks | 1 | 24 | Warning Console |
| Trajectory complexity | Single static evaluation, no time evolution | Multi-turn maneuvering with bank-angle budgets | Warning Console · glide archetypes |
| Hypersonic classes | Absent — 14 subclasses, none hypersonic | Glide and cruise archetypes as first-class families | Warning Console · Trajectory surface |
| Temporal behaviour | Historical / static | T+ clock with footprint collapse across the track's life | Warning Console scrubber |
| Out-of-distribution handling | None — model answered everything confidently | Novelty scoring; classifier narrowing suppressed on novel tracks | Warning Console · novel-only filter |
| Probability calibration | Never scored | Brier with Murphy decomposition, per-band reliability | Assurance |
| False-alarm budget | Undefined | Interactive operating point with modelled trust decay | Assurance |
| Analytic transparency | Bare numbers | ICD 203 estimative language, confidence scored separately, drivers, alternatives, falsifiers, lineage | Warning Console · assessment record |
| Decision rules | None | Bound to asset, action, named authority, and execution feasibility | Warning Console |
| Model lifecycle | None | Registry, stated limits, retraining triggers | Assurance |
| Reproducibility | Not seeded | Every track and every footprint seeded and reproducible | Warning Console · seed shown per track |
| Baseline payload | 38 MB parsed on two pages | Reduced by removing blocks no client code read | Wargame Console · M&S Dashboard |
Readiness assessment
Component by component
| Component | TRL | Basis |
|---|---|---|
| Atmosphere model | 6 | Reproduces the published standard to 3–4 significant figures. Mature, verifiable, uncontroversial. |
| 3-DOF propagator | 4 | Verified internally consistent and deterministic. Never compared against measured flight data. |
| Reachable-set / IPP engine | 4 | Produces coherent, collapsing footprints on synthetic tracks. No validation corpus exists. |
| Archetype classifier | 3 | Represented in the registry and drives band narrowing, but trained only on synthetic tracks generated by the same physics. Circular by construction. |
| Novelty detector | 3 | Detects parameter novelty. Untested against a genuine adversarial profile. |
| Calibration harness | 5 | The machinery is real and would work unchanged on real outcomes. Only the corpus is synthetic. |
| Decision-rule layer | 3 | Structure is sound. Thresholds, actions, and authorities are notional and would be set by an operational owner, not a vendor. |
| Operator interface | 4 | Runs at speed on realistic geometry. Never placed in front of an actual watch crew. |
The system TRL is the minimum of its load-bearing components, not the average. That puts this build at 3–4 regardless of how the demonstration presents.
Path to the next level
What each gate actually costs
TRL 4 → 5 · relevant-environment data
Requires: access to instrumented truth data, one archetype minimum
Gated on data access, not engineering effort
6–9 mo
TRL 5 → 6 · relevant-environment demonstration
Requires: live sensor feed, a real track picture, an operational sponsor
Gated on sponsorship and enclave access
9–15 mo
Classifier de-circularisation
Requires: training corpus not generated by the same physics it is used to constrain
Highest-value single fix — current arrangement flatters the model
3–6 mo
Independent test & evaluation
Requires: an evaluator who did not build it, with authority to fail it
Cheap, routinely skipped, decisive in a bake-off
2–4 mo
Watch-floor comprehension trial
Requires: real operators, timed tasks, and a willingness to publish a poor result
Lowest cost, highest information per dollar on this list
1–2 mo
Integration
A warning that reaches nobody is not a warning
The most common way capability demonstrators die is that they were never designed to emit into anything. Below is what this would have to plug into and what stands in the way.
| Integration surface | What ADL would emit | Principal obstacle |
|---|---|---|
| Common operational picture overlays | Footprint polygon, containment rings, modal cell, confidence, seed | Polygon-valued warnings do not fit displays built for point tracks. The display change is harder than the data feed. |
| Existing tactical data links | Reduced point plus uncertainty radius, degraded from the real product | Fixed-format messages cannot carry a footprint. Any link emission is lossy by construction and must be labelled as such. |
| Air and missile defence command systems | Advisory overlay alongside the organic impact-point solution | Two disagreeing impact predictions on one screen is an operator problem before it is a technical one. Deconfliction policy has to exist before the interface does. |
| Watch-floor collaboration tools | The assessment record verbatim, as a postable artifact | Lowest-friction and highest-value first integration. Text and an image, no certification burden. |
| Coalition picture | Tier 0 or Tier 3 model output only | Requires a separately trained releasable model, not a filter applied to a national one. |
| Post-event scoring archive | Frozen forecast plus observed outcome | Nothing currently exists to receive it, which is why nobody's impact-point model has a public calibration record. |
Sequence matters more than ambition. The collaboration-tool row is the
one to build first. It requires no certification, delivers the full assessment record
without loss, and puts the product in front of real operators months before any
system-of-record integration could clear. Everything above it is easier to justify once
somebody on a watch floor has already asked where the footprint went.
Scope discipline
What this is not, stated plainly
Not a targeting system. No aimpoints, no weapon pairing, no engagement recommendation. Target and Engage are removed from the kill chain on the console itself.
Not validated. No component has been compared against measured flight data. Every corpus on this site is synthetic and seeded.
Not connected. No live sensor, no classified input, no runtime backend. A static site with no data at rest.
Not 6-DOF. Point-mass only. Any claim requiring attitude dynamics, control surfaces, or structural response is outside this model.
Not a program of record. No contract vehicle, no sponsor, no acquisition decision represented or implied.
Not a substitute for the organic solution. Positioned as an advisory overlay that shows uncertainty, not as a replacement for a fielded impact-point capability.
Release
Version, and what the number means
Version policy is loaded from a single manifest that every page in the site stamps into its footer, so no surface can quietly drift out of date with the others.
—Capability version
—Build
—Site release
—System TRL
| Component | Version | Role | TRL |
|---|