Zero Overshoot claim ladder (2026-08)

1. What Zero Overshoot is

Zero Overshoot (GSRF Practical) is a deterministic soft thermostat for noisy signals. Most filters either chase the noise or smear the truth. ZO does neither: it pulls mid-band ring down and compresses peaks toward a declared normal, so detectors, controllers, and models can see a calmer input before the expensive stage.

Locked packs. Published failure modes. No slide-deck metrics.

2. Job it is built for

• Pre-stage signal hygiene on industrial / SCADA-style and command–measure loops where mid-band chatter and peak excursions hurt the next stage.
• Damp mid-band ring on locked spectral / oscillation pack surfaces (see Evidence · frequency).
• Compress peaks toward a declared normal on locked characterization packs (e.g. peak-deviation surface with pack downloads on home / evidence).
• Auditable path: frozen implementation, handable receipts, local gsrf-bench on your CSV.

3. Identity limit (not a bug)

ZO is not a general tracker. Tracking MAD loses by design on locked multi-family packages (0 / 52 MAD wins vs best EMA on the excellence-hunt surface). The spring is the product. If the KPI is hug-the-raw, use EMA / Kalman / a tracker — not balanced Practical as a pure tracker.

4. Where locked packs show strength (pack identity first)

Public figures already on the site. Do not invent new percentages. Every number below is bound to a named pack / surface — not free-floating marketing.

• Mid-band — locked Phase 6.2 pack: ~35% improvement on osc31; ~70% reduction on the defined sweet-band spectral surface versus best EMA. download pack · Evidence · frequency.
• Peaks — Excellence Hunt v2 pack: ~38% mean peak-deviation advantage; 49 / 52 characterization wins. download 49/52 CSV pack · Evidence.
• Adaptive normal — adaptive \(x^*\) probe (Evidence): regime-shift surface; frozen vs adaptive normal comparison published on Evidence · adaptive.
• Robot true command+measured continuous fleet (Phase 5 MAIN true-cmd surface; ABB / UR / Franka, n=298): wear holds 298/298, 0 continuous losses; osc/rate are surface-labeled only — do not merge one robot rate figure. Evidence · robot true-cmd.
• Pathology-scoped industrial synthetic map (independent adversarial / domain-map surface): domain dependence published (overall independent rates limited; stiction-class osc higher when labeled). Always keep domain + metric labels. Evidence · independent map · industrial batteries note.

Full downloads and tables: Evidence · packs directory. If a number cannot name its pack and metric surface, do not use it as a portable claim.

5. Where we lost (sell the opposite job and we lose)

Same tone as When GSRF lost:

1. Multi-minute delayed command path — MAD-to-measured worse vs EMA at 3 / 6 / 10 min on locked delay packages (~+59% / +97% / +106%). Do not deploy as a delay-tolerant command smoother.
2. Fixed false-alarm residual detection — residual recall ~0.44 (GSRF) vs ~0.83 (EMA) at matched FA on locked probe. Use a detector; use GSRF to calm.
3. Tracking MAD vs raw — 0 / 52 vs best EMA on excellence hunt multi-family. Identity, not noise.
4. Axiom industrial composite (PID-style pure evaluate, 2026-08) — on real PID-style loops with frozen Practical and Axiom industrial metrics, the sealed eligible set showed no clear composite wins vs strong causal smoothers; wear / command-churn-style terms often favour baselines. Boundary mapped; not a hero claim. Do not conflate Axiom industrial oscillation_power with Workbench mid-band / spectral pack claims — different protocols and series paths.

DAMADICS multi-day plant hardness remains published (composite mostly loss; wear better on 1 / 75 days — not a plant wear championship).

6. What we do not claim

• Not a universal industrial-PID composite winner under Axiom pure evaluate.
• Not “always beats EMA.”
• Not a delay-tolerant closed-loop command filter under multi-minute plant delay.
• Not a fixed-FA residual alarm engine product claim.
• Not a joint / PLC controller replacement.
• Not a certification authority; not AI-invented metrics.
• Not a guarantee of plant ROI or downtime dollars.

7. ZO vs Axiom (roles)

• Zero Overshoot — commercial candidate / pre-stage job for a soft thermostat.
• Axiom — controlled evaluation lab: sealed multi-method comparison under a frozen claim (not a public unlimited workbench). Request a sealed evaluation.

Bake-offs belong in Axiom (or gsrf-bench locally). Marketing adjectives do not replace a sealed pack.

8. How to evaluate

1. Run gsrf-bench or try on your CSV (evaluation license).
2. Open locked packs on Evidence.
3. For multi-method sealed comparison under one protocol → Axiom.
4. License when the job matches the green lights — not the red lights.

9. Relationship to older public PDFs

Older short PDFs under Papers (framework / product / limits stubs) are superseded for claim ladder wording by this note and the live Evidence / When-GSRF-lost surfaces. Live site numbers with pack links always win over older PDFs, offline copies of this note, or remembered slogans.

Optional PDF twin (same text discipline, no equations): zero-overshoot-claim-ladder-2026-08.pdf.

10. Contact

Product / pilots: gsrf@boonmind.io · Info: info@boonmind.io

Sources (public): Evidence packs · notes/when-gsrf-lost · FAQ · Workbench characterization surfaces already published on zeroovershoot.com · Axiom industrial pure-evaluate boundary (2026-08 sealed eligible set, PROMOTION NO).