The primary target is common-mode (CM) current that turns the wiring harness into an antenna. A common-mode choke (CMC) and split termination manage CM return paths so CM energy is contained, while the differential (DM) signal remains within timing and waveform limits.

• Radiated EMI often tracks CM current, not the DM signal itself.
• Any asymmetry converts DM → CM (parasitics, layout, protection mismatch, return-path gaps).
• Split termination provides a controlled CM return via the midpoint-to-ground network.
• CMC raises CM impedance so CM energy is less likely to flow onto the harness.
• Success requires two passes: EMI improves AND DM waveform/timing margin is preserved.

CM current is the in-phase current on both bus lines relative to chassis/return. When CM current is excited, the harness and connector geometry can radiate efficiently and can also act as an entry point for injected RF interference.

• CM symptom patterns: emission peaks that shift with harness length/routing; strong near-field hotspots at the connector/harness exit; RF injection sensitivity that does not correlate with DM amplitude.
• Root mechanism: DM-to-CM conversion from imbalance (component parasitics, placement, skew, reference discontinuities, uneven ESD/TVS capacitance).

• Emission signature: peak bands, margin-to-limit, and sensitivity to harness length/routing changes.
• CM current trend: compare CM current near the connector before/after changes (trend is more important than absolute).
• DM integrity: edge rate, overshoot/ringing, and sampling-point/timing margin under real harness + temperature.
• Error correlation: error counters vs operating states (load/temperature/switching events) to detect “randomization” from saturation or resonance.

• This page covers: CMC behavior (CM impedance, leakage effects), split termination networks (midpoint C/RC), layout/return-path strategy, and a verification workflow that preserves DM integrity.
• This page does NOT cover: full TVS/surge selection catalogs, detailed controller/register behavior, or gateway/diagnostics policy. Those belong to sibling pages and should be treated here only as parasitic inputs or interface constraints.

• Radiated EMI margin is poor and peaks shift with harness length or routing (antenna-like signature).
• Near-field hotspot at connector/harness exit or around the bus interface area (CM current leaving the ECU).
• RF injection sensitivity causes error bursts without a clear DM amplitude collapse (CM entry dominates).
• Harness is long / heavily loaded / branched (stubs and asymmetry increase DM→CM conversion risk).
• Ground offsets or noisy returns exist between modules (CM return paths become uncontrolled).

• Tight timing margin: added network capacitance or CMC leakage can slow edges or alter symmetry, reducing sampling-point headroom.
• Resonance risk: midpoint C with harness/ESD/connector parasitics can create narrowband peaks or ringing.
• Saturation/temperature drift: a poorly chosen CMC can behave nonlinearly across current/temperature, producing intermittent or “random” failures.

Rule: any EMI improvement must be accepted only if DM waveform and error counters remain stable across harness and temperature.

• Protection devices (TVS/ESD arrays): treated here as parasitics that can increase DM→CM conversion (capacitance mismatch, placement asymmetry). Detailed selection belongs to the protection page.
• Isolation decisions: large ground potential differences or HV domains should route to the isolated CAN/CAN-FD page.
• Controller/gateway behavior: filtering, bridging, and time-triggered topics belong to controller/bridge pages.

Differential-mode (DM) is the information-carrying signal. Common-mode (CM) is the EMC amplifier: CM current couples more efficiently to the harness and chassis return. Any imbalance converts DM energy into CM, and the termination/return network determines where CM energy flows.

• DM: currents are equal magnitude and opposite direction on the two lines (signal energy stays within the pair).
• CM: currents flow in the same direction on both lines relative to chassis/return (energy escapes to larger loops).
• Why CM is sensitive: the return loop often becomes physically large (harness + chassis), increasing radiation and injection coupling.

Any imbalance forces part of the DM energy to re-appear as CM. The most common sources cluster into three categories:

• Impedance asymmetry (ΔZ): unequal series/termination impedance, connector or harness mismatch.
• Parasitic asymmetry (ΔC/ΔL): ESD/TVS capacitance mismatch, uneven via/trace geometry, unequal coupling to reference.
• Return-path asymmetry: reference discontinuities, long chassis return, ground splits that enlarge loops.

• DM view: termination targets reflections and differential impedance matching.
• CM view: the termination/return network defines the CM impedance to chassis/return and therefore decides whether CM energy is absorbed locally or radiates on the harness.
• Implication: split termination is primarily a CM-path control element, not a “DM stability” shortcut.

1. Locate hotspots: strong fields near the connector/harness exit often indicate CM leaving the ECU.
2. Check harness sensitivity: peak frequency/margin changing with harness length or routing suggests antenna behavior (CM-dominant).
3. Correlate errors vs DM waveform: if errors burst without a clear DM collapse, CM injection or DM→CM conversion is a prime suspect.

Split termination is primarily a common-mode (CM) path control technique. Splitting the termination resistors and tying the midpoint to chassis/return through a capacitor (or C/RC network) creates a controlled high-frequency CM return, reducing harness radiation and improving immunity—only if the midpoint loop is short and symmetric.

• Standard termination: sets DM loading across the pair, but CM return can remain uncontrolled (energy escapes to harness/chassis loops).
• Split termination: creates a midpoint node that can be tied to chassis/return, providing a shorter CM return for higher frequencies.

The midpoint-to-return element behaves like a frequency-selective CM return. Increasing Cmid makes the CM return effective at lower frequencies, but also increases the risk of edge-shaping and resonance if the loop is inductive or asymmetric.

These estimates set intuition only; the harness, connector, and protection parasitics reshape the real response.

• Loop inductance: a long midpoint-to-return path can nullify high-frequency CM shunting and create narrowband peaks.
• Asymmetry: unequal routing or unequal parasitics to the midpoint strengthens DM→CM conversion.
• Resonance with parasitics: Cmid can resonate with harness/connector/ESD parasitics, increasing ringing or creating new EMI peaks.
• DM integrity impact: excessive Cmid or poor layout can slow edges and alter timing margins.

• Start small: begin with a modest Cmid (e.g., 4.7 nF) to target higher-frequency CM without heavily reshaping DM edges.
• Move with the failure band: increase Cmid only when the problematic emission/immunity band suggests a lower-frequency CM return is needed.
• If narrowband peaks or ringing appear: treat layout/loop inductance first; damping (RC) belongs to the next chapter.

A pure midpoint capacitor provides a high-frequency common-mode (CM) return, but it can also form a high-Q resonance with harness, connector, protection, and loop parasitics. Adding series damping (RC) is used to reduce Q, flatten narrow spikes, and improve repeatability across OEM sweep tests and harness variations.

• Primary effect: creates a shorter high-frequency CM return via the midpoint node.
• Best fit: the EMC issue is broadband or dominated by higher-frequency CM energy, with low harness sensitivity.
• Hidden assumption: the midpoint loop is short, low-inductance, and symmetric to avoid extra DM→CM conversion.

A midpoint capacitor is a strong reactive element. Combined with loop inductance and parasitic capacitances, it can create a narrowband resonance (high Q). High Q concentrates energy, producing sharper peaks in frequency-domain scans and longer ringing in time-domain edges.

• Loop L: long midpoint-to-return path, via stacks, distant chassis tie points.
• Parasitic C: connector + harness + ESD/TVS capacitance and mismatch.
• Low damping: insufficient loss in the CM loop keeps Q high.

• Reduce Q: series R introduces loss, lowering resonance sharpness and ring-down time.
• Flatten spikes: improves sweep-test stability by avoiding “one-point failures” at narrow bands.
• Improve repeatability: less sensitive to harness and chassis return variability.
• Trade-off: some CM shunting strength is sacrificed; verification gates are mandatory.

1. Start with C: use a modest midpoint capacitor to target high-frequency CM first.
2. Observe: check peak shape (broad vs spiky), ring-down behavior, and harness sensitivity.
3. Add RC only if needed: if narrow spikes or long ringing appear, introduce series damping.
4. Lock with two gates: EMI improves without creating new spikes, and DM waveform/timing margin stays intact.

A CMC is selected to present high impedance to common-mode current while minimizing differential-mode impact. The practical behavior is governed by Lcm, leakage/mismatch (Llk), winding resistance (RDC/DCR), and parasitic capacitance (Cp). Saturation and temperature drift can turn an EMC improvement into a corner-case failure if margins are not verified.

• Ideal: DM flux cancels (low DM impact), CM flux adds (high CM impedance).
• Real: imperfect coupling introduces leakage and mismatch; winding resistance adds loss; parasitic capacitance creates high-frequency bypass paths.

• Lcm: sets CM impedance vs frequency (main EMC knob).
• Llk / mismatch: appears in DM path as unintended series inductance → edge shaping / ringing risk.
• RDC/DCR: amplitude loss + heating; can shift behavior across temperature.
• Cp: high-frequency bypass path; can weaken CM suppression or introduce new coupling.

• CM impedance in the target band: Zcm(f) must be meaningful where emissions/immunity problems occur.
• Leakage control: keep DM impact small and symmetric to avoid DM→CM conversion.
• DCR / loss: avoid unnecessary amplitude loss and temperature-dependent drift.
• Saturation margin: ensure transient/common-mode currents do not push the core into nonlinearity.
• Variation: part-to-part tolerance and harness variation should not flip pass/fail in sweep tests.

• Core saturation: effective Lcm collapses under certain conditions → suppression weakens and peaks move.
• Temperature drift: DCR and magnetic properties shift, changing the balance between CM suppression and DM impact.
• Result: pass/fail may depend on operating corners unless margins are validated with real harness + temperature.

• Placement: commonly between transceiver and connector/harness to reduce CM current leaving the ECU.
• Relative to split: place the CMC closer to the interface side when the goal is “less CM on the harness”; keep the split midpoint loop short and symmetric near its return reference.
• Verification set: CM current trend (interface), DM waveform impact, and corner robustness (temperature/transients).

CAN and FlexRay use different termination definitions. Split termination must follow the system’s termination “metering” (Rterm per end and topology rules). Treat split termination as a common-mode return strategy on top of an existing termination, not as a re-definition of the termination itself.

• Typical HS CAN bus: two end terminations, often 120Ω per end (across the pair).
• Power-off resistance sanity: the whole bus typically measures ≈60Ω (two ends in parallel).
• What it catches: missing termination, extra termination, open harness, incorrect endpoint placement.
• Scope: this is a topology sanity check; high-frequency behavior still depends on harness and parasitics.

• Rterm is often a range: many FlexRay designs use a per-end termination within ~80–110Ω (system-defined).
• Equivalent load varies: the observed “total load” depends on bus vs star, node count, and whether couplers carry termination.
• Dual-channel A/B: each channel is metered independently; avoid collapsing A and B into one “single number.”
• Sanity approach: verify termination placement and topology assumptions first, then interpret measurements.

• Rule 1: determine the system-defined Rterm per end for CAN or FlexRay.
• Rule 2: split resistors must sum to that same Rterm (per end), then a midpoint network is added for CM control.
• Rule 3: never reuse CAN “60/60” assumptions on FlexRay, and never back-propagate FlexRay ranges onto CAN.
• Rule 4: any termination change must pass both EMI and DM integrity gates (waveform + margin) on real harness.

Use a target-band-driven workflow: define the failure mode first, then apply the minimum change (split C), add damping (RC) only when resonance signatures appear, introduce a CMC when CM still escapes to the harness, and verify every step with dual gates: EMI improvement and differential-mode (DM) integrity.

• Failure mode: radiated emissions vs injected immunity weakness (record the test method and setup).
• Target band: identify the frequency region that fails and the most sensitive harness/attachment condition.
• Output: a single “target” statement that drives every subsequent change.

• Start with split C: begin with a modest Cmid (e.g., 4.7 nF) to target higher-frequency CM.
• Watch for resonance signatures: narrow spikes, long ringing, or strong harness sensitivity indicate high-Q behavior.
• Add RC damping: introduce series damping to reduce Q and stabilize sweep results when those signatures appear.

• Trigger: split/RC improves some bands, but CM-related peaks or sensitivity remain strong at the interface.
• Selection cues: low RDC/DCR, high coupling (low leakage impact), Zcm meaningful in the target band, and robust current/temperature margins.
• Goal: reduce CM current leaving the ECU while keeping DM waveform and timing margins intact.

Detailed verification tactics are handled in the verification chapter; this workflow defines when each knob is allowed to move.

Layout determines whether the network suppresses common-mode (CM) energy or converts differential-mode (DM) energy into CM. Symmetry and a controlled return path are first-order variables; component values are second-order tuning only after the return path is correct.

• Matched environment: route the pair together with consistent spacing and the same reference plane.
• Matched discontinuities: keep via count, stubs, and neck-downs symmetric across the two lines.
• Interface symmetry: connector pads, TVS/ESD placement, and copper clearances must not favor one line.
• Plane continuity: avoid reference splits and uncontrolled return detours near the bus entry/exit region.

• Shortest loop wins: midpoint-to-return must be the smallest loop area on the page.
• Do not cross splits: a split plane or long detour adds inductance and weakens the capacitor’s effect.
• Resonance risk: large loop area increases Q and can create narrow spikes instead of suppression.
• Placement priority: place the midpoint network as a “local CM return” rather than a remote accessory.

• Same surroundings: avoid routing one line near a slot, via wall, chassis stitch edge, or copper boundary.
• No one-sided coupling: asymmetry around the CMC increases mismatch/leakage impact and can generate CM.
• Connector side discipline: keep the pair tightly controlled between CMC and connector to reduce harness-launch CM.

Verification must prove two outcomes at the same time: EMI improves in the target band and communication margin is not reduced. EMI-only wins are incomplete. Use a gated acceptance flow: EMI gate + signal-margin gate, both validated on real harness and across temperature corners.

• Near-field scan: locate the dominant source region and observe how it moves after each change.
• Shape matters: verify the target-band peak reduces and no new narrow spike becomes dominant.
• Harness sensitivity: repeat with representative harness routing/attachment, not only a bench layout.

• DM waveform integrity: overshoot, ringing, edge shape, and symmetry must not degrade.
• Timing/sampling window: confirm sample-point margin on the real harness and with worst-case temperature.
• Operational trend: error counters, retransmits, or dropouts must not increase after EMC fixes.

• Real harness: validate with representative length, branches, routing, and attachment points.
• Temperature corners: confirm the same configuration passes across automotive temperature range.
• Repeatability: verify results are stable across sample-to-sample variation (resonance-induced “random” behavior is a red flag).

• Standby current: confirm split/CMC networks do not introduce unintended bias/leakage paths.
• Wake behavior: ensure wake/sleep logic remains robust; deeper strategy is handled in wake/partial-networking pages.

A common failure pattern is “fix one EMI peak, create three new reliability problems.” Treat EMC changes as gated engineering: preserve differential-mode (DM) margin and repeatability across harness and temperature corners.

Note: concrete MPNs appear in H2-12 as reference BOM examples. Always verify value/tolerance/package/suffix/automotive grade and availability.

This section provides system patterns and selection logic, not a product catalog. Use scenario constraints (harness, nodes, immunity, ground offset, temperature) to drive choices.

Material numbers below are reference examples. Always verify value, tolerance, voltage rating, package, suffix, AEC-Q200 grade, and availability.

• Primary constraints: long/branched harness, many nodes, robust immunity, repeatability across build variants.
• Preferred order: layout gate (symmetry + midpoint loop) → split C minimum-change → add RC only if resonance is proven → add CMC if CM still escapes to harness.

• Primary constraints: larger CM disturbances, strong RF immunity requirements, corner repeatability (load/temperature).
• System stance: treat return-path planning as the first design object; consider isolation where ground potential differences dominate (handled in the isolated-bus page boundary).
• CMC focus: avoid saturation and mismatch-driven DM distortion; enforce “both lines see the same world” in layout.

• Primary constraints: termination definition depends on topology and channel; do not reuse CAN split values blindly.
• Selection principle: set Rterm by system definition first; then split the resistor values to match that definition.
• CMC decision: add only after proving CM escape remains dominant; verify DM timing/margin across corners.

• RDC gate: set an RDC upper bound to avoid DM amplitude loss and thermal drift penalties.
• Zcm effectiveness: require meaningful CM impedance in the failing band (from H2-8 “target band” definition).
• Non-linearity gate: avoid saturation in expected operating corners; unstable EMI behavior is a red flag.
• Coupling/consistency: high coupling and controlled leakage reduce DM distortion risk.
• Layout feasibility: only select packages that can be routed with symmetric environment (H2-9 layout gate).

• Start small: begin with 4.7nF midpoint C and verify both gates (EMI + DM margin).
• Tune by evidence: increase C only when the target-band peak remains dominant and DM margin is preserved.
• RC trigger: add damping only when resonance evidence exists (narrow spikes, long ringing, high harness sensitivity).
• Structure before values: loop area and return path quality are prerequisites for meaningful tuning.

These FAQs only close long-tail troubleshooting loops. Each answer is constrained to four lines with measurable pass criteria.

Data placeholders (X/Y/Z/f1–f2/N) are intentional—fill them with program-specific limits and test conditions.

EMI improves after adding split cap, but error frames increase — suspect C too big or loop too long first?

Likely cause Midpoint loop inductance dominates (C becomes ineffective or resonant), or C is over-loading the channel and shrinking DM timing margin.

Quick check Keep layout fixed; step C down (e.g., 4.7nF → 2.2nF) and compare edge symmetry + error-counter trend on the same harness.

Fix Reduce loop area first; then tune C from the minimum-change value. Add RC damping only if resonance evidence exists.

Pass criteria EMI gate: peak reduced by ≥ X dB in [f1–f2] MHz with no new dominant spike. DM gate: error frames ≤ Y per Z frames at [Temp corner] + [Harness variant].

Same 4.7nF works on Vehicle A but worsens Vehicle B — harness resonance shift or midpoint return difference?

Likely cause Harness impedance/branches move the resonance band, or the midpoint return path (signal GND vs chassis path quality) changes the effective CM loop.

Quick check Compare near-field peak band and time-domain ringing across A/B under the same ECU build; log the failing band and whether ringing duration changes.

Fix Keep C at minimum-change; improve midpoint return and symmetry first. If a narrow-band peak dominates, add controlled damping (RC) rather than blindly increasing C.

Pass criteria Repeatability gate: same fix holds across N harness routings/build variants. EMI gate: peak reduced by ≥ X dB in the recorded band. DM gate: errors ≤ Y per Z frames at corners.

EMI rebounds in certain temperature/current corners after adding a CMC — saturation or bias dependence?

Likely cause CMC non-linearity (approaching saturation) or mismatch/leakage converting DM→CM under corner conditions.

Quick check Correlate peak amplitude with load/current/temperature; if EMI shifts abruptly with operating point, non-linearity is likely. Compare with a higher-headroom CMC family if available.

Fix Select CMC with more headroom (avoid saturation) and better consistency; enforce symmetric routing/placement so both lines see the same environment.

Pass criteria EMI gate: improvement maintained across [Temp min–max] and [Load range] with drift ≤ X dB. DM gate: error frames ≤ Y per Z frames across the same corners.

Near-field at the connector improves, but whole-vehicle radiated emission barely changes — is the DM→CM conversion point not at the interface?

Likely cause The dominant conversion occurs elsewhere (harness branching, ground discontinuity, asymmetric protection, or a distant return-path break), so interface-only fixes cannot remove the main radiator.

Quick check Scan along harness/branch points to locate the peak hotspot band; compare “hotspot moves” vs “hotspot remains” after interface changes.

Fix Treat symmetry and CM return continuity as a system problem: eliminate the dominant conversion point first, then re-tune split/CMC locally if needed.

Pass criteria Hotspot gate: dominant radiator location reduced by ≥ X dB in [f1–f2] MHz. Vehicle EMI gate: test limit margin ≥ Y dB. DM gate: errors ≤ Z per corner test.

RC damping flattens the peak, but edges become slower — how to set a “not too slow” threshold?

Likely cause Damping resistance is consuming too much DM energy, reducing edge rate and shrinking sampling margin.

Quick check Sweep Rdamp in small steps and record: (1) EMI peak height, (2) DM rise/fall time, (3) error-counter trend on the same harness.

Fix Choose the smallest Rdamp that removes the narrow-band resonance signature while preserving DM edge and symmetry; do not compensate with “bigger C” blindly.

Pass criteria DM gate: rise/fall time change ≤ X% and sampling margin ≥ Y% of baseline; errors ≤ Z per corner test. EMI gate: peak reduced by ≥ W dB in [f1–f2] MHz.

CAN is fine, but FlexRay fails after copying the same split/CMC scheme — is termination definition being mixed?

Likely cause Split resistor values and/or midpoint network are not aligned with FlexRay termination definition (Rterm range/topology), causing incorrect loading and timing distortion.

Quick check Confirm the system’s required Rterm per channel/topology; measure bus resistance per definition (not the CAN “60Ω sanity check”).

Fix Re-derive split resistor values from the FlexRay termination definition, then re-tune midpoint C/RC using the same dual-gate verification.

Pass criteria Definition gate: measured loading matches expected range. EMI gate: ≥ X dB improvement in target band. DM gate: errors ≤ Y per Z frames at corner conditions.

Measured bus resistance is not expected (CAN not ~60Ω) — what termination misconfigurations to check first?

Likely cause Missing end termination, extra termination(s) enabled, incorrect R values, or measurement taken with nodes powered/biased or through unintended parallel paths.

Quick check Verify end nodes and their termination enable states; measure with power removed and isolate segments if the topology includes stubs/branches.

Fix Restore exactly two end terminations per CAN segment (or per the system spec); then re-validate split/CMC tuning only after the base loading is correct.

Pass criteria Loading gate: resistance within expected range (placeholder: ~60Ω for standard two-end CAN). DM gate: waveform symmetry restored. EMI gate: improvements remain after fixing base termination.

Only one node has split termination and the benefit is limited — how do termination location and topology constrain results?

Likely cause Split termination is most effective where the segment’s CM current returns; applying it away from the true end or on a stub cannot control the dominant CM loop.

Quick check Confirm which nodes are true ends; compare hotspot and peak band with split placed at the end vs at a mid-node (keep all else constant).

Fix Place split termination at the segment end(s) per definition; keep midpoint return short and low-inductance. Use CMC only if CM still escapes to the harness.

Pass criteria Topology gate: termination placement matches definition. EMI gate: ≥ X dB reduction at dominant radiator location. DM gate: no increase in errors across corners.

CMC placement (connector side vs transceiver side) changes results a lot — what is the first decision criterion?

Likely cause The dominant CM energy is either escaping through the harness side (interface radiation) or being created by on-board asymmetry/return-path breaks; placement controls which loop is suppressed.

Quick check Near-field scan: if the connector/harness region dominates, place CMC nearer the connector; if the board region dominates, fix symmetry/return first and avoid masking the root cause.

Fix Use placement that blocks the dominant CM loop while keeping both lines symmetric. Re-check DM waveform after placement changes.

Pass criteria Hotspot gate: connector-region CM reduced by ≥ X dB (if that was dominant). DM gate: waveform symmetry maintained and errors ≤ Y per Z frames at corners.

Switching TVS supplier makes EMI worse — how to quickly decide if parasitic mismatch causes DM→CM conversion?

Likely cause Unequal line-to-ground capacitance/ESL between CANH/CANL (or A/B) increases DM→CM conversion near the interface.

Quick check Swap back to the previous TVS or temporarily remove/replace with a matched dummy to see if the peak band follows the protection change (keep routing identical).

Fix Use matched, symmetric protection placement and routing; ensure both lines see equal parasitics. Detailed protection selection belongs to the protection page boundary.

Pass criteria Symmetry gate: measured/estimated parasitics within X% mismatch. EMI gate: peak restored/improved by ≥ Y dB in [f1–f2] MHz without DM margin loss.

Should the midpoint capacitor return to signal ground or chassis ground — which return path to prioritize?

Likely cause The “best” return is the one that forms the shortest, lowest-inductance CM loop without injecting CM current into sensitive internal references.

Quick check Compare two return options with identical placement: measure (1) dominant EMI band peak, (2) near-field hotspot shift, (3) DM waveform symmetry and error counters.

Fix Choose the return that minimizes loop area and keeps CM current off sensitive reference paths; keep routing symmetric and the midpoint connection short.

Pass criteria EMI gate: ≥ X dB improvement at the target band. Integrity gate: no added noise coupling to internal references beyond Y (unit placeholder). DM gate: errors ≤ Z per corner test.

Radiated emission passes, but RF injection immunity is still weak — check CM loop first or DM sampling margin first?

Likely cause Injection failures can be CM-driven (return path and conversion points) or DM-margin-driven (edge/sampling window). Passing radiated emission does not prove CM robustness under injection.

Quick check Under injection, log: (1) error counters vs injected band, (2) DM waveform distortion at the receiver, (3) whether failures correlate with a specific harness routing/return discontinuity.

Fix If failures track return-path/harness changes, prioritize CM loop control (symmetry + return continuity). If failures track edge/sampling shrinkage, reduce DM-impacting tuning (C/RC/CMC mismatch).

Pass criteria Immunity gate: error-free (≤ X errors per Y minutes) across injected bands and power levels per spec. DM gate: margin ≥ Z% of baseline. Repeatability gate: holds for N harness variants.