~70%
Kills mid-band ring
Spectral power reduction in the sweet band (period ≈ 50–120 samples; often minutes on 1‑min industrial traces) vs EMA.
phase6.2 · download pack · section
Gradient-Stabilized Recursive Filtering · GSRF
A Soft Thermostat for Noisy Signals
Calm sensor signals before they hit expensive models—without pretending to be a tracker. Mid-band ring damp + peak compression toward a declared normal. Evidence locked. Limits published.
Self-serve first — download receipts, no book-a-call · Evidence · Pricing
~38%
Compresses peaks
Mean peak-deviation advantage vs best EMA; 49 / 52 dataset wins across characterization families.
excellence hunt v2 · download 49/52 CSV pack
9.7 → 1.5
Adapts to regime shifts
Frozen \(x^*\) stuck after permanent jumps; adaptive \(x^*\) tracks 50 → 70 → 40 without abandoning the spring.
Workbench adaptive \(x^*\) probe · section
The new angle is trust
Not “we reduce oscillation.” Not “we beat EMA.” Those can be true in a band—and still be the wrong pitch.
What we sell
We know exactly what this filter does, exactly where it works, and exactly where it fails— and we keep the receipts: pre-declared rules, locked packs, and documented red lights.
What everyone else sells
A magic black box and a hope you don’t test the edge cases. We tested GSRF to destruction—and left the failures in the evidence page on purpose.
Self-verify on your own signals (~10 minutes)
Your CSV beats our 52 datasets for a purchase decision. gsrf-bench is the free credibility engine — run GSRF Practical vs EMA locally (osc31 · peak_dev · MAD · spectral).
Loudest path:
install eval/ → gsrf-bench --input your.csv --normal <x*> → read the metrics.
No sales call. Non-production LICENSE-EVAL.
Evaluation only. Production needs a license after you like your own numbers. Pricing · email only after self-serve: gsrf@boonmind.io
Audit Workbench — the system behind the claims
Local offline characterization engine: frozen evaluators, pre-declared rules, multi-method overlays, synthetic + adversarial + DAMADICS real-plant packs. Marketing only climbs the claim ladder.
Explore Zero Overshoot
Evidence →
Identity, frequency, adaptive x*, safety cage, industrial packs with charts.
Workbench →
Characterization system, gallery, claim ladder.
Research →
Long-form notes from locked packs.
Papers →
PDFs + technical library index.
Industrial actuators →
Commercial flagship. Kill mid-band ring without pretending to track. Valve/SCADA map + true cmd+meas robot wear holds. Not a controller.
Compare →
GSRF vs EMA vs Kalman decision table.
Pricing →
Eval / Pro / Enterprise / QEC path.
PdM pipelines →
Long-tail: pre-model sensor hygiene for industrial AI.
Applications →
Robotics, control, integration, production checklist.
Sources →
DAMADICS paper, eval package, PDFs.
Safety cage detail: Evidence §4 · Losses: When GSRF lost · Quantum: separate ladder
Questions people actually ask
Short-tail answers. Long-tail deep dives live in Research and the full FAQ.
What is Zero Overshoot / GSRF?
A deterministic log-space soft thermostat that kills mid-band oscillation and compresses peaks toward a declared normal. Not a general tracker. Identity →
Does GSRF always beat EMA?
No. Often wins mid-band osc and peaks on locked packs; loses tracking MAD, delayed command paths, fixed-FA residual. Compare →
Does GSRF win industrial actuators?
Not universally. Stiction-class osc and provisional slow-track wear on independent synthetics; plant DAMADICS hardness published. True cmd+meas robot continuous wear holds (ABB/UR/Franka) with surface-labeled osc/rate. Actuators flagship guide →
How do I try it without sales?
Python snippet or local gsrf-bench on a CSV. Evaluation only.
Where does it fail?
MAD tracking, multi-minute delay as command smoother, fixed-FA residual alarms. When GSRF lost →
Is the snippet a production license?
No. Commercial license required for production. Pricing →