Bounded Review Agents · Hardware

Thousands of artifacts. A ranked review packet. No signoff claim.

Ark Axiom compresses the review surface where generated hardware, board evidence, and the software stack meet human engineering judgment. The packet shows what should be reviewed first, what evidence exists, and which claims remain blocked.

Fixed scope. Three business day standard evaluation. No ranked review packet produced → no fee. Review-prioritization metrics are not correctness, coverage, readiness, or signoff claims. Every Ark Axiom review agent runs behind an Evaluation Gate: it can rank review targets and preserve evidence, but it cannot approve signoff, fabrication, tapeout, benchmark validity, or production readiness.

427x
RTL compression
Median object reduction across 15 historical replay cases.
1224x
PCB localization
Motor controller layer-transition sample compression.
16
Compiler targets
Fixture-backed early lane packet: 8 P1 and 8 P2 review targets. Compression metrics in progress.
3 days
Evaluation
Standard bounded packet turnaround after scoped intake.

Bounded Review Agents

Three review surfaces. One packet discipline.
Neither replaces your engineers.

Ark Axiom compresses the review surface where generated hardware, board evidence, and the software stack meet human engineering judgment.

01 / Current · RTL Review Agent

Turn massive RTL drift into manageable review targets.

Compresses RTL review surfaces before formal verification, DV, or manual review begins. It ranks signals, cones, FSM drift, and candidate assertion questions for engineer inspection.

Signal drift and cone changes
FSM state and transition changes
Semantic drift regions
Clock and reset-domain flags
Historical repair-region replay overlay
Candidate SVA review targets and assumption questions
02 / Current · PCB Review Gate

Review board evidence before fab, assembly, or flash.

Compresses board, BOM, fab-package, and firmware-to-board assumptions into ranked review blockers. Gate means claim/action boundary, not manufacturing approval.

Net changes and power/ground anomalies
Decoupling targets and connector mismatches
Floating nets and BOM substitution risks
Manufacturing-rule flags
Fab-package completeness review
Firmware-to-board assumption mismatches
03 / Early Lane · Compiler-Hardware

Expose software-to-silicon assumption gaps early.

Fixture-backed v0.1 lane for compiler output, ISA assumptions, simulator behavior, runtime assumptions, benchmark claims, and RTL evidence. Early lane means normalized evidence packet, not compiler verification.

ISA semantic mismatch review targets
Lowering-path and unsupported fallback flags
Simulator/RTL behavior questions
Memory-layout and runtime assumptions
Benchmark-claim boundaries
Generated artifact provenance gaps
Explicit Boundaries

We prepare the evidence. Your engineers make the call.

A review agent that overclaims is a liability. Ark Axiom blocks correctness, signoff, manufacturing, benchmark, and replacement claims. The packet is a bounded review queue, not permission to tape out, fabricate, ship, or trust a benchmark.

Every packet preserves evidence and hashes.
packet_authorizes_execution:false
human_operator_handoff:true
01
PCB Tamper Witness
Detects board and BOM revision drift: nets, components, traces, vias, text, BOM substitutions, component movement, and routing changes.
02
PCB ECO Gate
Turns revision drift into ordered review blockers: power/ground, debug/boot/control nets, interface routes, connector geometry, 0-ohm bridges, BOM evidence.
03
PCB Fab Readiness Gate
Checks whether the fab/assembly package is complete enough to review: source files, Gerbers, drills, BOM, CPL/PNP, stackup, DRC/ERC reports, and fab drawing.
04
Firmware-Hardware Assumption Gate
Checks whether firmware assumptions match board evidence: MCU identity, GPIO mappings, UART/I2C/SPI expectations, boot/reset/debug pin usage, build/flash/log evidence.

Axiom Evaluation Gate

The bounded agent boundary before every review packet.

Before Ark Axiom reviews RTL, PCB, BOM, firmware, fab-package, or compiler-hardware evidence, the evaluation scope is converted into a bounded authority envelope.

What governs the review agent before it touches customer artifacts?

Axiom Evaluation Gate controls what a review agent may read, derive, claim, export, and hand off during a bounded hardware evaluation. The gate defines what artifacts may be inspected, what evidence may be derived, what claims are blocked, what outputs may be exported, what must be handed to a human engineer, and what remains outside scope.

The review agent may rank, localize, summarize, and preserve evidence. It may not approve signoff, fabrication, tapeout, benchmark validity, compiler correctness, RTL correctness, BOM correctness, DRC/ERC, SI/PI, safety, or production readiness.

scope_envelope
bounded artifact and source-access scope
artifact_manifest
what was provided, hashed, and reviewed
allowed_review_modes
RTL, PCB, fab, firmware, compiler-hardware
allowed_agent_actions
rank, localize, summarize, hash, packetize
blocked_agent_actions
signoff, fab, tapeout, benchmark, source modification
human_operator_handoff:true
engineers decide consequence
packet_authorizes_execution:false
packet does not approve action
zero_signoff_authority:true
no correctness or readiness approval
{
  "axiom_evaluation_gate": {
    "gate_id": "AXIOM_EVAL_GATE_v0_1",
    "review_mode": "rtl_review | pcb_eco | fab_readiness | firmware_hardware | compiler_hardware",
    "scope_envelope": {
      "artifact_types_allowed": ["rtl", "pcb", "bom", "firmware_config", "compiler_evidence"],
      "source_access": "sanitized | nda | public_fixture",
      "production_access_allowed": false,
      "credential_access_allowed": false,
      "external_side_effect_allowed": false
    },
    "allowed_agent_actions": [
      "rank_review_targets",
      "localize_evidence",
      "summarize_drift",
      "emit_reason_codes",
      "hash_artifacts",
      "produce_review_packet"
    ],
    "blocked_agent_actions": [
      "approve_signoff",
      "approve_fabrication",
      "approve_tapeout",
      "approve_benchmark",
      "modify_source_artifacts",
      "submit_to_fab",
      "replace_engineer_review"
    ],
    "human_operator_handoff": true,
    "packet_authorizes_execution": false,
    "zero_signoff_authority": true
  }
}

Explicit Boundaries

What Ark Axiom does not do.

These boundaries are not disclaimers. They are the architecture. A review agent that overclaims is a liability, not a tool.

— Blocked Claims · Every Packet
RTL correctness or formal signoff
Tapeout readiness or production safety
PCB fabrication or assembly approval
DRC, ERC, DFM, SI/PI validation
Firmware correctness or release approval
BOM correctness or AVL approval
Compiler correctness or simulator/RTL equivalence
Benchmark validation or hardware readiness
Bug detection or coverage closure
Replacement of human electrical, RTL, or compiler engineering review

Sample Packets

See the deliverable before sending files.

Controlled ECOs and historical replay cases on public boards and public RTL. Customer artifacts evaluated only under NDA. View the packet mode map → · Open evidence graph →

PCB — Public Board ECOs
antmicro/artix-dc-scm
FPGA / Data-Center Control Board ECO
4409x
2 affected nets · 3 review targets. Local component movement: route displacement and net-level compression.
Open packet →
mjbots/moteus
Motor Controller Power-Stage ECO
306x
3 affected nets · 4 review targets. Coordinate drift, tstamp-matched segments, dual compression ratios.
Open packet →
mjbots/moteus · layer transition
Motor Controller Layer-Transition ECO
1224x
Hard case: F.Cu→B.Cu transition, via detection, nearby motor-phase net triage at 0.117mm.
Open packet →
RTL — Public Generated HDL
WarpCore · Amaranth HDL GPU Core
Public HDL GPU Review Packet
12 targets
Bounded review-compression pass: scheduler timing, SIMT lane execution, ISA parity, memory assumptions, generated-Verilog provenance.
Open packet →
Compiler-Hardware — Early Lane
Fixture-backed normalized evidence
Compiler-Hardware Assumption Gate Demo
16 targets
8 P1 · 8 P2. ISA, simulator, RTL, runtime, memory, benchmark, fallback, and provenance review targets. Early lane; compression metrics in progress, not compiler verification.
Open packet →
RTL — Historical Replay
Ibex · historical replay
Instruction-Fetch Control Replay
2278x
Top-1 exact-signal overlap · ibex_id_stage. Public RTL historical repair region.
Open packet →
CVA6 · historical replay
MMU / TLB Context Replay
762x
Top-1 exact-signal overlap · cva6_mmu. Public RTL historical repair region.
Open packet →
OpenTitan · historical replay
SPI Host FSM Replay
349x
Top-1 exact-signal overlap · spi_host_fsm. Public RTL historical repair region.
Open packet →

Proof Matrix

What the current evidence supports.

Scoped proof artifacts and technical demos. Not production signoff claims.

Evidence Track Best Current Proof What It Supports What It Does Not Claim
RTL Review Compression 15 / 15 public replay cases Ranked review-target and signal localization for bounded public RTL repair contexts Bug detection, formal proof, RTL correctness, DV signoff
RTL Assertion Gate 5 cases · 2 strong · 3 partial Candidate SVA, assumptions, vacuity warnings, human approval questions Final specs, proof success, non-vacuous proof, DV engineering replacement
PCB ECO Gate System76 Launch public KiCad history Board-revision review blockers for routing, power, ground, debug, boot/control DRC/ERC, SI/PI, thermal validation, manufacturing readiness
BOM / AVL-Style Evidence nSealr TROPIC01 public BOM path BOM/AVL-style review evidence in controlled and real before/after BOM path BOM correctness, AVL approval, substitute-part suitability, DFM
PCB Fab Readiness Gate Complete + missing fixture packages Checks for expected source, Gerber, drill, BOM, CPL/PNP, stackup, DRC/ERC presence File correctness, manufacturability, electrical behavior, certification
Firmware-Hardware Gate 7 deterministic fixtures Review blockers for MCU identity, pin assumptions, UART/I2C/SPI drift, boot conflicts Firmware correctness, HIL testing, board bring-up, safety certification
Compiler-Hardware Assumption Gate v0.1 Fixture-backed packet · 16 targets · SHA-256 verified Assumption review across ISA, simulator, runtime, benchmark, memory, and RTL evidence Compiler correctness, simulator/RTL equivalence, benchmark validation, performance portability, hardware readiness

How It Works

Submit artifacts. Receive ranked targets.
In 3 business days.

A fixed-scope engagement, not a software deployment. Runs alongside your existing review workflows without replacing EDA, DRC/ERC, or signoff.

01
Send a bounded artifact pair
A reference/candidate RTL pair, PCB/BOM revision set, fab-package folder, firmware pin map, or normalized compiler-hardware evidence set — scoped under NDA before intake begins. Sanitized inputs accepted.
RTL pair · KiCad / BOM
Fab package folder
Firmware project + board config
Compiler/ISA/runtime evidence
Sanitized inputs OK · NDA first
02
Ark compresses the review surface
Review ranking, cone isolation, visual diffing, and provenance capture reduce thousands of objects to the targets that deserve human attention first.
RTL Review Witness
PCB Revision Witness
Fab Readiness Gate
Firmware-Hardware Gate
Compiler-Hardware Assumption Gate
03
Receive a bounded evidence packet
Ranked review targets, cone or board-region notes, MD/JSON/HTML outputs, SHA-256 manifests, and explicit claim boundaries — ready to hand off to your engineering team.
Review packet · SHA manifest
Claim boundaries on every artifact
Review memo: what was checked,
what was not, what still requires human review

Review compression before signoff  ·  not signoff replacement  ·  Gate means claim/action boundary  ·  not manufacturing approval  ·  Humans decide consequence

Built For

Who sends files.

Hardware teams that need review compression before signoff — not autonomous AI making engineering decisions.

Semiconductor
IP teams reviewing RTL drift, FSM changes, and clock-domain flags before handing off to DV.
AI Hardware Startups
Teams using AI-generated RTL or board revisions who need human review compression before signoff.
Robotics / IoT
Teams preparing PCB revisions or firmware-to-board assumption reviews before fabrication or flash.
Design Services
Teams compressing review before handoff to clients — delivering ranked targets and claim-bounded packets alongside design files.
Board Revision Teams
Engineering teams managing PCB ECOs, BOM substitutions, or fab-package completeness checks across revisions.
Not for
Teams seeking autonomous chip design, formal proof replacement, DRC/ERC automation, or manufacturing signoff. Those are not this product.
Bounded Evaluation

Send a sanitized artifact.
Get a review packet.

One bounded RTL, PCB/BOM, fab-readiness, firmware-hardware, or early compiler-hardware assumption review packet. Scope and NDA agreed before intake.

No ranked review packet produced → no fee.
From $10K · 3 business day turnaround.
Request Evaluation
What to send. RTL reference/candidate pair, PCB/BOM revision set, fab-package folder, firmware project with board config, or normalized compiler/ISA/runtime evidence. Sanitized inputs accepted.
NDA before intake. Scope, artifact type, and claim boundaries agreed before any files are reviewed. No credentials, source secrets, or production access required.
What you receive. Ranked review targets, cone, board-region, or assumption-boundary notes, SHA-256 manifests, and a review memo with explicit claim boundaries on every artifact.
Built for. Semiconductor IP teams, AI hardware startups, robotics and IoT teams, board-revision reviewers, and teams evaluating AI-generated hardware packages.