hw-testing

name	hw-testing
description	Hardware testing capability pack. Covers power-on testing (voltage rails, smoke test), functional peripheral verification, per-mode power measurement and battery life calculation, environmental test planning (IEC/MIL standards), EMC pre-compliance (emissions/ESD), production test fixture design, and human-AI pair testing with physical instruments (4D Protocol) — embedded MCU prototypes first. Use for any hardware bring-up, prototype validation, power profiling, compliance pre-check, or production test task.
version	0.1.0
type	reference-based
keywords	["hardware testing","硬件测试","power-on","上电测试","smoke test","冒烟测试","voltage rail","电压轨","functional test","功能测试","peripheral","外设","power measurement","功耗测量","deep sleep","深度睡眠","battery life","电池续航","environmental test","环境测试","IEC 60068","EMC","电磁兼容","ESD","静电","FCC","pre-compliance","预检","test fixture","测试治具","pogo pin","探针","pair testing","人机协作测试","multimeter","万用表","oscilloscope","示波器","embedded","嵌入式","MCU","prototype","样机","量产测试"]

Hardware Testing Capability Pack

From power-on to production validation — physical measurement + AI-guided analysis. Scope (v0.1, inherited from source v1.0): Embedded hardware prototypes (MCU-based, battery-powered devices). Hardware testing requires PHYSICAL instruments (multimeter, oscilloscope, power analyzer) — this pack's value is what to test + how to test + pass/fail criteria + documentation. UNIVERSAL RULE: 编造数据 = FAIL. Every voltage, current, RSSI, timing, or sensor value must come from an actual instrument reading (human-reported for pair testing). Fabricated measurements, standard citations, or harmonic numbers = FAIL. Measure it or report "not measured".

---

Step 1: Context Detection

User Signal / Task Type	Load Reference
new board bring-up, first power-on, voltage rails, smoke test, 上电, 冒烟测试, 电源轨	references/power-on-test.md
peripheral verification, display/sensor/button/WiFi/BLE/SD test, I2C scan, 外设, 功能测试	references/functional-test.md
power profiling, sleep current, battery life, µA, 功耗, 续航, 深度睡眠	references/power-measurement.md
temperature/humidity/drop/vibration test plan, IP rating, IEC 60068, MIL-STD, 环境测试	references/environmental-test.md
EMC, radiated/conducted emissions, ESD, FCC/CE/CCC, harmonics, 电磁兼容, 预检	references/emc-precheck.md
production test, bed-of-nails, pogo pin, jig, cycle time, 治具, 量产	references/test-fixture.md
human at the bench with instruments, guided measurement, 4D protocol, 人机协作	references/hw-pair-testing.md
before Gate review / reviewing hw test work / acceptance check	references/review-checklist.md

Multi-signal: load all matched references.

---

Step 2: Decision Entry Point

Q1 — Where is the board in its lifecycle?

• Fresh from fab, never powered → START with power-on-test.md (ALWAYS first — nothing else before rails verify)
• Powers on, peripherals unverified → functional-test.md (priority: power → comms → display → sensors → wireless → storage)
• Works, battery-powered product → power-measurement.md (per-mode profile + battery life calc)
• Heading to test lab / certification → environmental-test.md + emc-precheck.md (design review BEFORE paying for lab time)
• Heading to production → test-fixture.md

Q2 — Is a human at the bench with instruments? → hw-pair-testing.md (Mixed type: human operates, AI guides + analyzes; human reads every value)

Q3 — Capability type determines workflow shape:

• Code B (power-on, functional, power): select → execute → verify → optimize — generates scripts/checklists/procedures
• Doc A (environmental, EMC, fixture): search → analyze → derive → generate — standards MUST be looked up, never guessed
• Mixed (pair testing): prepare → discover → discuss → decide → deliver

Q4 — Reviewing or accepting work? → ALSO load review-checklist.md (personas + Gate 2/Gate 4 checklists).

---

Step 3: Core Judgment Rules

Power-On (references/power-on-test.md)

• Spec-first measurement: ALL pass/fail specs MUST be declared BEFORE executing tests (thresholds from datasheet/schematic design intent, never from the measurement itself — source: OpenHTF)
• Before power: continuity check VCC↔GND expect >1kΩ (short = DO NOT POWER ON); bench supply current limit 100mA initially
• At power-on: idle draw expect <50mA typical MCU board; >200mA immediately → POWER OFF; smoke test (5s heat/smell watch) is a mandatory explicit step
• Rails: every rail has expected value, ±5% tolerance, measurement point; any rail >±10% off nominal = FAIL; quiescent current within datasheet typical ±20%; no component >50°C at idle
• Record actual measured values, never just PASS/FAIL

Functional (references/functional-test.md)

• Every peripheral on the BOM has a test; mark CRITICAL vs OPTIONAL — PASS = all CRITICAL pass
• Every test has quantitative pass/fail criteria (never "it works"); sensor readings compared to reference instrument or datasheet typicals
• Wireless MUST include RSSI (expect >-70 dBm at 1m from AP); storage MUST use write-read-verify byte match
• Completeness = % of BOM peripherals covered, not test count; define a hardware-mocking path (pyvisa-sim or equivalent) for CI regression

Power Measurement (references/power-measurement.md)

• Deep sleep on µA range, never auto-range (injects noise); switch range BEFORE entering sleep; wait 10s for stable reading
• Sleep current >100µA → suspect a peripheral not powered down; flag if > datasheet typical ×2
• Sanity: deep sleep < idle < active, or the measurements are inconsistent
• Battery life: formula shown with actual measured values (I_avg over a realistic usage profile), then apply 80% derating; compare against the product requirement

Environmental (references/environmental-test.md)

• Every test condition traceable to a specific IEC/MIL/ISO standard clause — look up, never guess
• Severity matches the product's real environment; sample size ≥3 units per test; non-destructive tests BEFORE destructive
• Battery behavior at temperature extremes explicitly addressed (LiPo degrades below 0°C)

EMC Pre-Compliance (references/emc-precheck.md)

• Design review BEFORE measurement (cheaper to fix on paper); inventory ALL clock frequencies and calculate harmonics from actual values
• ESD protection verified on EVERY external port: USB ≥±8kV contact (IEC 61000-4-2 Level 4), other ports ≥±4kV contact
• Thresholds = FCC Part 15 Class B radiated (3m): 40.0 dBµV/m @30-88MHz, 43.5 @88-216MHz, 46.0 @216-960MHz, 54.0 @>960MHz; conducted (QP): 66→56 dBµV @150-500kHz, 56 @0.5-5MHz, 60 @5-30MHz
• Standards cited with edition/year and matched to target market (FCC/CE/CCC)

Test Fixture (references/test-fixture.md)

• Test point map from actual PCB design files, never assumed; every PCB test point mapped to a fixture contact
• Pogo pins: ≥2A rating for power, ≥0.5A signal, contact resistance <50mΩ, >100K cycle life for production; alignment repeatability <±0.2mm
• Test sequence <30s per board; pass/fail automated via UART self-test parsing; operator guide requires no engineering knowledge
• Fixture BOM complete with part numbers; maintenance plan included (pogo pins wear out)

HW Pair Testing (references/hw-pair-testing.md)

• ONLY the human reads instruments — AI never produces a measurement value; AI specifies instrument + range for every measurement request
• Anomalies get differential diagnosis (ranked causes), not "it failed"; AI analysis cites datasheet values and tolerances
• Decide in-session while the probe is on the board; human makes severity/priority decisions and approves any hardware modification
• Record instrument model/calibration status for traceability; log complete enough to reproduce the session

---

Anti-Patterns

Power-On

• ❌ Powering on without checking for shorts first (can destroy board)
• ❌ Setting bench supply to unlimited current (use 100mA limit initially)
• ❌ Skipping visual inspection (solder bridges are the #1 prototype killer)
• ❌ Recording only PASS/FAIL without actual measured values
• ❌ Testing with battery before bench supply verification (battery has no current limit)
• ❌ Measuring first, then deciding pass/fail (spec-first measurement: declare thresholds from datasheet/design BEFORE testing — source: OpenHTF)

Functional

• ❌ Testing only happy path ('it displayed something' without checking correctness)
• ❌ Skipping I2C address scan (address conflicts are common on prototype boards)
• ❌ Not recording actual sensor values (only PASS/FAIL loses diagnostic data)
• ❌ Testing wireless without RSSI measurement (range problems hide in 'connected = pass')
• ❌ Assuming SD card works after 1 write (test with larger payload for reliability)
• ❌ No hardware abstraction layer in test scripts (tests coupled to specific instruments = can't run in CI — source: awesome-hardware-test)
• ❌ Paper/Excel-based test management (no traceability, no version control — source: OpenHTF test database pattern)

Power Measurement

• ❌ Using auto-range multimeter for µA measurements (injects measurement noise)
• ❌ Calculating battery life from datasheet typical values instead of actual measurements
• ❌ Not derating battery life estimate (real-world is 70-80% of calculated)
• ❌ Measuring deep sleep current immediately after entering sleep (wait for stabilization)
• ❌ Ignoring peripheral quiescent current (LDO Iq can dominate in deep sleep)

Environmental

• ❌ Applying MIL-STD-810H to a consumer desk gadget (overkill, expensive)
• ❌ Not testing at temperature extremes with battery powered (LiPo at -20°C is a real failure mode)
• ❌ Testing only 1 sample (insufficient for reliability conclusions)
• ❌ Doing drop test before temperature test (drop test may damage device, invalidating later tests)
• ❌ Specifying IP67 without understanding the test cost and design implications

EMC

• ❌ Skipping design review and going straight to measurement (cheaper to fix on paper)
• ❌ Not testing ESD on USB port (most common compliance failure)
• ❌ Assuming shielding solves everything (shielding without proper grounding makes it worse)
• ❌ Ignoring cable radiation (cables are often the dominant emission source)
• ❌ Using unshielded DCDC inductor and hoping for the best
• ❌ Testing without chamber calibration (must verify 16 points in 1.5m×1.5m area first — source: GTEM cell testing guide)
• ❌ Frequency step size >1% of current frequency or dwell time <0.5s/frequency (insufficient resolution — source: IEC 61000-4-3)
• ❌ EUT cables touching reference ground plane (must be elevated ≥30mm — source: EMC pre-compliance testing guide)

Test Fixture

• ❌ Designing fixture without PCB test point layout (fixture must match PCB exactly)
• ❌ Using undersized pogo pins for power rails (need ≥2A rating for VCC/GND)
• ❌ No alignment mechanism (operator will misalign, pogo pins will miss pads)
• ❌ Manual pass/fail judgment (automate with UART self-test output parsing)
• ❌ Forgetting maintenance plan (pogo pins wear out, contact resistance increases)

HW Pair Testing

• ❌ AI fabricating measurement values (only human reads the instrument)
• ❌ Deferring all decisions to 'review meeting later' (decide while probe is on board)
• ❌ Measuring without specifying instrument range (auto-range on µA measurements = noise)
• ❌ AI deciding hardware modifications without human approval
• ❌ Not recording instrument model/calibration status (traceability)

---

Anti-Skip Table

Shortcut Attempt	Required Action
"Board looks fine, just power it on"	MUST run pre-power checks in power-on-test.md: visual inspection + VCC↔GND continuity (>1kΩ) + 100mA supply limit first
"I'll decide pass/fail once I see the numbers"	MUST declare measurement specs from datasheet/schematic BEFORE testing (spec-first, OpenHTF pattern)
"It connected to WiFi, wireless passes"	MUST record RSSI (>-70 dBm at 1m) per functional-test.md — connected-only hides range problems
"Datasheet says 5µA sleep, use that for battery life"	MUST measure actual sleep current on µA range and apply 80% derating per power-measurement.md
"Standard conditions are roughly -20 to +60°C"	MUST cite the specific IEC/MIL clause per environmental-test.md — never guess test conditions
"We'll find EMC issues at the lab"	MUST complete the design review + harmonic inventory in emc-precheck.md first — lab time is $5K-$15K
"Fixture layout from the board photos"	MUST derive test points from actual PCB design files per test-fixture.md
"I'll estimate the reading, human is busy"	Only the human reads instruments (hw-pair-testing.md). Fabricated values = FAIL across every capability