Conducting the Swarm

The deep-strike raid of 2026 is no longer a salvo. It is a score — decoy waves to map the defenses and drain the magazines, propeller Shaheds to force the radars on, jet Shaheds against the defended targets, cruise and ballistic missiles timed to the confusion, all sequenced across hundreds of kilometers like instruments entering a symphony. Raid design has become the drone war's mature art, and both Russia and Ukraine now practice it with a conductor's discipline.

The Score

Read a modern Russian mass raid as sheet music. First movement: decoys — Gerbera drones of foam and plywood, launched hours early to light up air-defense radars, map coverage, and force the first expensive shots; their wreckage is reconnaissance data delivered by enemy interceptor. Second movement: propeller Geran-2s on converging tracks, saturating sector defenses and holding radars radiating so strike planners can see them. Third: jet Geran-3s at better than 300 knots against the defended targets, arriving through corridors the earlier waves priced. Coda: Kalibr and Kh-101 cruise missiles and Iskander ballistics, timed to land in the window when magazines are reloading and crews are retasking.

A single night's package can run to 500-plus airframes across these echelons — a number unthinkable in 2023, routine in 2026. The sophistication lives in the sequence, not in any single munition.

The Symmetric Art

Ukraine conducts the same orchestra against Russian refineries and depots, with its own instrumentation: reconnaissance decoys that map the air-defense belt, waves of cheap airframes to drain the Pantsir and Tor magazines around the target, then the strikers — Behemoth-class heavy drones and long-range one-way attackers carrying the warheads that matter — timed to arrive when the defenders' launch tubes are hot and empty. The refinery campaign of 2025–2026 showed the learning curve in fire statistics: hit rates climbed as raid composition matured from single-wave swarms to sequenced echelons.

Both sides studied the same teacher — the other's intercept statistics — and convergence was inevitable. Cheap raids get intercepted. Expensive raids win through arithmetic. Conducted raids win through timing. What distinguishes the mature campaign from the amateur one is repetition with variation: the same score played differently each night, so the defender's lesson from Tuesday is obsolete by Friday.

The Conductor's Software

The command-and-control behind a sequenced raid is the piece nobody photographs. Raid-planning software allocates launch windows across dispersed sites — a 500-airframe night may originate from a dozen locations on different azimuths — and computes time-on-target so echelons arrive in order despite different speeds and ranges. Launch scheduling is logistics: railheads, erector crews, weather windows, moon phase, and the mundane tyranny of how many launch rails a site can cycle per hour. Real-time abort and redirect, where datalinks allow, turns a raid from a firing into a maneuver — and the absence of those links in cheap airframes is a deliberate trade, because autonomy costs less than control.

The planning cell is the weapon. Munitions are inventory; the raid file — which decoys, which corridors, what timing — is combat power, and it improves with every night's telemetry. A raid that loses seventy percent of its strikers still pays the planner in data: which radars radiated, which sectors held fire, how long reloads took. The failed raid is reconnaissance with a warhead.

The Defender's Answer

Defense against the conducted raid is a doctrine problem before it is a hardware one. Raid-pattern recognition — machine learning over radar and acoustic tracks that classifies a wave as decoy, dilution, or strike within its first minutes — lets commanders hold fire against the foam and shoot the warheads. Engagement sequencing doctrine, what to ignore and in what order, is now written discipline in Ukrainian air-defense units, born of nights when shooting everything meant defending nothing. Magazine discipline completes the set: interceptor allocation per sector, reserve policies, and the unglamorous logistics of having missiles left at hour six of a six-hour raid.

The scoreboard is cost exchange, recalculated nightly. A decoy that draws a $100,000 missile wins; a decoy ignored is plywood wasted. The defender's hardest discipline is restraint under fire — crews who watch tracks cross their sector untouched, trusting the pattern-recognition call while the raid is still live. Both sides know the arithmetic. Both sides adjust by dawn.

The 2028 Horizon

The next turn of this spiral is algorithmic on both ends. Raid planning is already software-assisted; by 2027–2028, expect raid files generated by optimization against a live model of the defender — routes, decoy ratios, and timing computed nightly from fresh intercept data. Defense answers with pattern-recognition systems that retrain on the same night's tracks. The loop closes on both sides, and tempo becomes the decisive variable: the side whose planning cycle — raid, telemetry, analysis, new raid — turns faster imposes adaptation cost on the other. Export control, incidentally, cannot touch any of this. Software for conducting raids rides the same open machine-learning stack as logistics optimization, and the models that matter train on data only the combatants have.

The swarm was never the weapon. The score is the weapon — and the conductor who rewrites it fastest owns the night.

That competition will not produce a winner so much as a gradient: raid sophistication and defensive adaptation rising together, cost curves grinding against each other, advantage accruing to whichever industrial base refreshes software faster than the other refreshes airframes.

What to Watch

- Raid composition ratios. The decoy-to-striker fraction in nightly tallies is the attacker's adaptation made countable. - Time-on-target coordination. Simultaneous arrivals from multiple azimuths in strike footage evidence planning software, not luck. - AI raid-planning disclosures. Contractor and service announcements about automated strike planning mark the loop's closure. - Shoot/no-shoot doctrine changes. Defender publications on engagement sequencing admit, formally, that the conducted raid won the argument.

The deep strike has matured from terror weapon to conducting art, and the art's next movement will be written by whichever side closes that decision loop fastest.