Beacon's Requirements: From Wish to Specification
"Long battery life" is a wish. "≥ 6 months at one report per minute at 22 °C" is a requirement — and it constrains every decision after it.
🎯 By the end you'll understand
What separates a requirement from a wish: a measurable quantity, units, the conditions it holds under, and a stated method of verification. How to write Beacon's nine requirements so that each one could actually reject a design you liked. How to derive a top-down power budget from R1 — usable capacity divided by hours gives an allowed average current — and how to compare it against the bottom-up design figure to get a margin you can defend. How R1 propagates into sleep architecture, R2 into antenna layout, R5 into every part choice, and R8 into placement, so that later arguments become arithmetic. Which requirement combinations are mutually exclusive and how to find that out in an afternoon rather than after a board spin. And why requirements written after the design are a description rather than a constraint.
KEY INSIGHTA requirement is a statement that could have rejected a design you wanted to build
"It should have good battery life and be small and cheap." Everyone nodded. Six weeks later two engineers were arguing about a capacitor, and the specification could not settle it, because it agreed with both of them.
The purpose of a requirements table is not documentation. It is to convert future arguments into arithmetic. When someone proposes a bigger inductor, a longer advertising burst, or a second sensor, a real requirements table answers immediately: that costs 4 µA of average current, the allowance is 183 µA, we are at 38, therefore yes — or that adds 8 mm to the board and R8 caps it at 35, therefore no unless R8 changes and here is who has to agree. Without that table, the same question is a matter of taste, seniority, and whoever is most tired. The test of whether you have written requirements or wishes is brutally simple: name a design you would like to build that this document forbids. If you cannot, you have written a description of what you were going to do anyway.
Every requirement needs five things: an identifier so it can be referred to, a measurable quantity with units, the conditions under which it must hold, a verification method saying how you will know, and a priority saying whether it is negotiable. Beacon's R1 has all five: ≥ 6 months (quantity and units), at one report per minute at 22 °C on a 1000 mAh cell (conditions), verified by a measured average current plus a 72-hour bench run and a 30-day pilot (method), mandatory (priority). "Long battery life" has none of them, and that is why it can never be met, never be violated, and never settle anything.
CORE CONCEPTThe anatomy of a requirement
The obligation words, and why they are worth being pedantic about
The convention, formalised in the ISO/IEC directives for standards and in RFC 2119 for internet specifications, is narrow and useful: shall (or must) means mandatory, should means recommended and negotiable with justification, and may means optional. Using them consistently means a reader can tell at a glance what is up for discussion. Mixing them — or writing "will", "needs to", "is expected to" — hides the negotiability of every line and pushes the argument to a meeting. Beacon's canon marks R9's IP54 enclosure as optional and everything else as mandatory, and that single distinction saves an argument in Stage 6.
The four verification methods
Systems engineering recognises four, and choosing one at the time you write the requirement is what stops it from being unfalsifiable:
- Inspection — look at it. R8's board dimensions are verified by inspecting the fabrication drawing. Cheapest, and available immediately.
- Analysis — compute it from other verified numbers. R1's battery life is verified in the first instance by analysis: measured average current divided into usable capacity. Analysis can be done before hardware exists, which is why it is the method that shapes the design.
- Demonstration — operate the thing and observe that it does what it should, without instrumenting the quantity. R7's OTA-with-rollback is demonstrated: push a bad image, watch it roll back.
- Test — instrument and measure. R3's ±0.3 °C is verified by test against a reference thermometer; R1's six months is confirmed by test in the 30-day pilot and a 72-hour bench run, which cannot prove six months but can falsify it.
Notice R1 uses three methods at three different times. That is normal and correct: analysis early to shape the design, bench test at Stage 4 to catch gross errors, and pilot data at Stage 6 for the real distribution. A requirement whose only verification is "we will see in six months" is a requirement you cannot manage.
Requirements say what, design decisions say how
A requirement should constrain the outcome, not prescribe the mechanism, because prescribing the mechanism forecloses solutions you have not thought of yet. "The tag shall achieve ≥ 10 m indoor range through one plaster wall" is a requirement. "The tag shall use an ESP32-C3-MINI-1" is a design decision.
Beacon's canon fixes the hardware — the module, the sensors, the regulator, the board size — and it is worth being honest about why that is not a contradiction. Those are design decisions made early and deliberately, each traceable to a requirement, recorded so the course has a concrete thread to follow. The right way to keep the distinction alive is to write the trace: ESP32-C3-MINI-1 chosen to satisfy R2 (pre-certified module with a characterised PCB antenna), R5 (lowest-cost module meeting the radio requirement), and R8 (module size), with the ESP32-S3 alternative considered and rejected in Lesson 1.6. Now if a requirement changes, you know exactly which decisions to revisit — and if you find a decision with no trace to any requirement, you have found either a missing requirement or a decision nobody needed to make.
No units. "Fast", "small", "low power", "reliable". No conditions. "6 months of battery life" without saying at what report rate and what temperature is unmeasurable, because the answer depends entirely on those. No verification. If nobody can say how it will be checked, it will not be. Comparative without a baseline. "Better range than the competitor" — which competitor, measured how? Two requirements in one sentence. "Shall be small and cheap" cannot be partially met, partially traded, or independently verified; split it. Each smell has the same cure: add the missing part, and if the missing part turns out to be un-addable, the requirement was never real.
DEFINITIONBeacon's requirements: R1 through R9
These are the canon's nine requirements, reproduced exactly, with the verification method that this lesson adds to each. Everything in the rest of the course traces back to this table.
| # | Requirement | Number | Why it constrains later work | Verified by |
|---|---|---|---|---|
| R1 | Battery life at 1 report/min, 22 °C | ≥ 6 months on 1000 mAh LiPo | Drives sleep architecture (3.7, 4.3), regulator choice (4.2), BLE payload (4.6) | Analysis from measured average current; 72 h bench test (4.3); 30-day pilot (6.9) |
| R2 | Indoor BLE range to gateway | ≥ 10 m through one plaster wall | Drives antenna/module choice (4.5), RF layout (5.7) | Link-budget analysis (4.5); test with a gateway at 10 m through a wall, on 5 units |
| R3 | Sensors | Temp ±0.3 °C, RH ±3 %, motion wake ≤ 100 ms | SHT40 + LIS2DH12 on I2C (3.5, 3.6) | Test against a reference instrument across 0–45 °C; scope the LIS2DH12 INT1 line on GPIO3 through to wake |
| R4 | Report latency after motion | ≤ 2 s | Interrupt path (3.4), advertising interval (4.6) | Test: marker GPIO from INT1 (GPIO3) to first advertisement, captured on a scope (1.7) |
| R5 | Unit BOM cost at 25 units | ≤ IDR 150,000 (~USD 9.5) excluding enclosure | Part selection everywhere; cost curve (6.8) | Inspection of the costed BOM at 25-unit pricing |
| R6 | Charging | USB-C, 500 mA, with protection | 4.4 | Test: charge current measured; demonstration of over-voltage and reverse protection |
| R7 | Updates | OTA with rollback | 4.9, partition table (3.2) | Demonstration: push a deliberately failing image and observe rollback |
| R8 | Size | PCB ≤ 35 × 25 mm, 4-layer | Placement + stackup (5.4, 5.5) | Inspection of the fabrication drawing and the stackup |
| R9 | Environment | Indoor, 0–45 °C, IP54 enclosure optional | Enclosure (6.5), reliability (6.6) | Test at both temperature extremes; IP54 by inspection of the enclosure specification if fitted |
"≥ 6 months on 1000 mAh LiPo" is the quantity. "At 1 report/min" is a condition, and it is the one that does most of the work: the report interval is the single biggest lever on average current, so without it the number means nothing. "At 22 °C" is a second condition, and it is deliberately not the top of R9's 0–45 °C range — which means R1 as written does not guarantee six months in a warm ceiling void, and everyone should know that. "On 1000 mAh LiPo" fixes the energy source, which in turn interacts with R8 and R5 through the cell's physical size and price. Four interacting statements in fourteen words, and every one of them is load-bearing.
HOW IT WORKSDeriving the power budget from R1, top-down and bottom-up
A requirement becomes useful the moment you can turn it into an allowance that every later decision spends against. For R1 that conversion is three lines of arithmetic.
Top-down: what R1 allows
Convert the duration to hours
Six months, taken as six thirty-and-a-bit-day months, is 6 × 730 = 4,380 hours. Use a consistent convention and write it down; the difference between 4,380 and 4,392 hours is irrelevant to the answer and endlessly relitigable if you leave it implicit.
Convert nameplate capacity to usable capacity
A cell's 1000 mAh is measured at a stated discharge rate down to a stated cutoff voltage under laboratory conditions. Your product cannot use all of it: some sits below the voltage at which your regulator stops working, some is lost to self-discharge over the six months (a few percent per month for lithium polymer), and some is lost to ageing and temperature. An 80 % usable figure — 800 mAh — is a reasonable and openly illustrative derating for a six-month window; Lesson 4.1 does this properly and argues about every term in it. What matters here is that the derating is written down, so that when someone later quotes a different battery life you can find the disagreement in seconds.
Divide
Allowed average current = 800 mAh / 4,380 h = 183 µA. That is R1, restated as a number every subsequent design decision can be measured against. Every microamp anyone proposes to spend comes out of 183, and the question "can we afford this?" now has an answer instead of an opinion.
Bottom-up: what the design costs
Lesson 1.6's budget assembled the design's own figure from datasheet rows and measurements: a deep-sleep floor of about 8 µA for the whole board, and an active window of about 45 ms at 40 mA once per minute, which averages to 0.045 × 40 mA / 60 = 30 µA. Total: 38 µA.
Set the two side by side. The allowance is 183 µA and the design spends 38 µA, a margin of about 4.8×. Stated the other way — the canon's way — the design's theoretical life on nameplate capacity is 1000 mAh / 38 µA ≈ 26,000 hours ≈ 3 years, and the six-month requirement therefore survives a 6× real-world derating for self-discharge, temperature, retries, and everything else the bench does not show you. Both framings are the same fact, and having both is useful because the first answers "may we spend this?" and the second answers "how wrong can we be?"
Of the 38 µA, sleep is 8 (21 %) and the active window is 30 (79 %). So halving the active window buys 15 µA, while halving the sleep current buys only 4 µA. That single comparison redirects the whole of Stage 4: shorter bursts, a smaller payload (R4.6's 14 bytes), fewer packets per burst, and a faster sensor read are worth several times more than any heroic effort on leakage. It also explains why Rev A's 1.85 µA divider leak — a 23 % overrun on sleep — costs only about 5 % of the total and does not threaten R1. Engineers reflexively optimise sleep current because it is the number that sounds like low power. The budget says otherwise, and the budget is the reason to trust the answer over the reflex.
# Requirements are more useful as data than as prose: the gateway's subscriber # (canon: 40-line Python, MQTT -> CSV) can then check every report against them. REQS = { "R1_avg_current_uA": {"max": 183, "cond": "1 report/min, 22 C, 1000 mAh"}, "R3_temp_accuracy_C": {"max": 0.3, "cond": "0-45 C vs reference"}, "R3_rh_accuracy_pct": {"max": 3.0, "cond": "0-45 C vs reference"}, "R4_motion_latency_s": {"max": 2.0, "cond": "INT1 to first advertisement"}, } def check(name, value): """Return (ok, message). A requirement you cannot evaluate is not a requirement.""" r = REQS[name] ok = value <= r["max"] return ok, f"{name}: {value} vs max {r['max']} ({r['cond']}) -> {'PASS' if ok else 'FAIL'}" if __name__ == "__main__": # the design's own numbers, Lesson 1.6 assert check("R1_avg_current_uA", 38.0)[0] # 38 uA against a 183 uA allowance assert not check("R1_avg_current_uA", 210)[0] # and it really can fail print("requirements table is evaluable")
SEE ITRequirement propagation map — change a number, watch the design change
Edit any requirement and the downstream constraints recompute: the allowed average current, the transmit power the range demands and what that costs in current, the physical envelope the cell forces, the BOM total, and the latency budget. Combinations that cannot coexist flag themselves. The cost and link-budget models here are illustrative and stated as such — the point is the coupling, which is real and is what the table exists to expose.
Things to try. Start at the canon: 6 months, 60 s, 1000 mAh, 10 m, IDR 150,000, 35 × 25 mm, 2 s. Everything is consistent with real margin, and the chart shows you could report roughly every ten seconds and still just meet R1 — useful ammunition next time someone asks for faster reporting. Now push the report interval to 10 s and watch the margin evaporate, because the active term scales directly with reporting rate and it is 79 % of the budget. Put the interval back and raise R2 to 50 m instead: the link budget demands more transmit power, the active current rises, and a range requirement has quietly become a battery requirement — the coupling that a requirements table exists to make visible. Then ask for 24 months while shrinking the cell to 200 mAh, and read the flag. Finally take the cell to 2000 mAh with the 25 × 20 mm board: the requirement set is satisfiable electrically and physically absurd, because the cell is now four times the area of the PCB and R8 has stopped being the constraint that sets the product's size.
IN PRACTICEWhere each requirement lands
The value of the table is that it reaches forward. Below is the propagation, requirement by requirement, into the decisions the rest of this course will make. When you read a later lesson and wonder why a choice was made the way it was, this is the answer.
Two of the couplings deserve naming, because they are where real projects break.
Range costs battery. R2 is met by a link budget, and the levers in that budget are transmit power, antenna gain, receiver sensitivity, and the number of retries. Transmit power costs active current directly, and retries cost active current by extending the window — so a requirement expressed in metres lands on R1's microamps. Concretely, in the model used above, moving R2 from 10 m to 50 m raises the required transmit setting from 0 dBm to the top of the module's range, which lifts the active current from about 40 mA to about 72 mA and the average from 38 µA to 62 µA. This is exactly why R2 says "10 m through one plaster wall" and not "as far as possible": the second version has no upper bound, so it silently consumes the entire power budget.
Size costs energy and area. R8's 35 × 25 mm PCB is a placement constraint, but the cell that R1 assumes is a 1000 mAh pouch measuring roughly 34 × 50 × 5 mm — larger than the board. The product's physical size is therefore set by the battery, not by R8, and shrinking the board past a point buys nothing until you also shrink the cell, which spends R1. Discovering that relationship in Lesson 1.9 costs an afternoon. Discovering it in Stage 6, when the enclosure is quoted, costs a redesign.
BEACONClosing Stage 1
The stage's milestone, from the canon: requirements table (R1–R9) written; LED + button + LDO on a breadboard; first current measurement. All three are now done, and it is worth stating precisely what each is worth.
The requirements table exists and is falsifiable
Nine requirements, each with a quantity, units, conditions, and a named verification method. R1 has been converted into a 183 µA allowance and compared against a 38 µA design, giving a 4.8× margin that is written down rather than assumed. The couplings between R1, R2, R5, and R8 are known. If someone proposes a change tomorrow, the answer is arithmetic.
The breadboard proves what it can and no more
An XC6220 LDO — deliberately not the product's regulator — with an LED and a button, running firmware. Lesson 1.8 established exactly which questions this apparatus can answer: DC behaviour, logic, firmware, and current at the tens-of-milliamps scale. The active-window current has been confirmed to the right order of magnitude, which validates the dominant term in the budget. The sleep current has not been measured, and the log says so in words rather than in a misleading number.
The open questions are named, with a platform and a stage attached
Four things Stage 1 could not settle, each now scheduled rather than forgotten. The buck converter's stability needs a real board (4.2). The 8 µA sleep floor needs a low-burden instrument on Rev A (4.3). The antenna's match and R2's range need the module at a board edge with a defined keep-out (5.7, and measured in 4.5). The ADC's absolute accuracy needs per-unit calibration in production test (6.3). Naming an unknown and giving it a date is the difference between a plan and an optimism.
Handoff to Stage 2
Stage 2's milestone is a schematic on paper: battery sense, an I2C bus with pull-ups, button debounce, LED driver, and protection. Every one of those is now constrained by something written down. The battery-sense divider's 2 × 1 MΩ and 100 nF came from Lesson 1.3's loading analysis and Lesson 1.4's dielectric choice. The pull-up values will come from R3's sensors, the I2C specification's rise-time limit, and Lesson 1.8's warning not to copy what happened to work on a breadboard. The LED driver's high-side PMOS came from Lesson 1.5, along with the gate-drive rule that would have caught Rev A's divider leak. The protection comes from R6 and R9. Stage 2 begins with logic levels and pull-ups, which is the first place where the numbers in this table meet a real bus.
They will change, and that is fine — what is not fine is changing them silently. Beacon's rule, and the canon's, is that a requirement change is a deliberate act: the number changes in the table first, the design decisions traced to it are re-examined, and the change is dated with a reason. That is why the canon says "if a lesson needs to change a number here, the canon changes first". The alternative is the state most projects are actually in, where the specification and the product have quietly diverged and nobody can tell which one is wrong. Rev A ships with three known defects (canon, 6.9) precisely because they were compared against requirements and found to be tolerable — a judgement that was only possible because the requirements existed first.
WEAKNESSWhere this bites
This is the failure the whole lesson exists to prevent, and it is almost never deliberate. What happens is that the design starts first — reasonably, because someone had a good idea — and the requirements get written later, from the design, to satisfy a process or a customer. Every number in such a table is achievable by construction, so the document cannot reject anything and never will. It looks identical to a real requirements table and has none of its function. The test: for each requirement, name a design you would plausibly have wanted to build that this number forbids. If R1's answer is "a version that reports every ten seconds", the requirement is real. If the answer is "nothing", you have written a description. The cure is not to throw the table away but to tighten the numbers until they bind — and to notice, honestly, that tightening them after the fact is much harder than choosing them before.
Each stakeholder contributes a number that is defensible on its own: marketing wants 50 m of range, the customer wants two years of battery, procurement wants IDR 100,000, and industrial design wants a 25 × 20 mm board. Every one is arguable; the set is not satisfiable, and nothing in the individual reviews reveals that. The only defence is to compute the couplings early — which is what the widget above does in seconds — and to bring the conflict back as arithmetic rather than as reluctance. "50 m needs about 14 dB more transmit power, which raises the active current by roughly 80 %, which with a 200 mAh cell gives eleven weeks, not two years" is a conversation. "That sounds hard" is not. Note also that the impossible combination is usually discovered by whoever is closest to the physics, which means the obligation to raise it early falls on you.
"Six months of battery life" with no report interval is not a weak requirement, it is several incompatible requirements sharing a sentence — at one report per minute it is comfortable, at one every ten seconds it is impossible, and everyone reading it assumes the interval they had in mind. The same failure hides in "±0.3 °C" without a temperature range, "10 m range" without a wall, and "IDR 150,000" without a quantity. Each of those omissions has ended a project's schedule at least once. The habit that fixes it is the one from Lesson 1.6, applied to a document you are writing rather than reading: every number gets its conditions, every time, even when they feel obvious. They feel obvious because you know them; the person reading it in Stage 5 does not.
KEY INSIGHTHow this connects
- What an Embedded System Is (1.1) — the loop-in-a-box with constraints. This lesson writes those constraints down with units so the rest of the course can spend against them.
- Reading a Datasheet Without Getting Fooled (1.6) — a requirement is a datasheet row you write for yourself, and it is judged by the same standard: a value, a statistic, and the conditions it holds under.
- Measurement (1.7) — every verification method in the table above is a measurement, and this stage's instrumentation lesson is what makes those verifications trustworthy.
- Breadboards, Prototyping, and the Bench (1.8) — which requirements can be verified on which platform, and which have to wait for real hardware.
- Logic Levels, Pull-ups, and Open-Drain (2.1) — the next lesson, and the first place R3's sensors meet a real bus with real timing constraints.
- Sleep Modes and the Power Budget (4.3) — where R1's 183 µA allowance is spent explicitly and defended on a bench for 72 hours.
- Capstone: Beacon Pilot Deployment and Post-Mortem (6.9) — where every requirement is checked against 25 units for 30 days, and where the three known Rev A defects are judged against this table rather than against taste.
Stage 1 has installed the physics: voltage, current, power, the two conservation laws, real components with their fine print, switches that heat when half-driven, datasheets read with their conditions, instruments that change what they measure, and prototypes that only prove some things. This lesson turns all of it into a document that decides. Stage 2 starts drawing the circuit that has to obey it.
GLOSSARYKey terms introduced
- Requirement
- A statement with an identifier, a measurable quantity and units, the conditions under which it must hold, a verification method, and an obligation level. Distinguished from a wish by being falsifiable, and from a description by being able to reject a design.
- Obligation words
- Shall (mandatory), should (recommended and negotiable with justification), may (optional). Used consistently so a reader can tell instantly what is up for discussion.
- Verification method
- One of inspection (look at it), analysis (compute it from verified numbers), demonstration (operate it and observe), or test (instrument and measure). Chosen when the requirement is written, not when it comes due.
- Traceability
- The recorded link between a design decision and the requirement it satisfies. Lets you find every decision that must be revisited when a number changes, and exposes decisions that satisfy no requirement at all.
- Allowance (budget)
- A requirement restated as a quantity that later decisions spend against — R1's ≥ 6 months becomes 183 µA of allowed average current. Converts "can we afford this?" from an opinion into arithmetic.
- Top-down versus bottom-up budget
- Top-down divides the requirement into an allowance; bottom-up sums what the design actually costs. The ratio between them is the design's margin, and having both means you can answer both "may we spend this?" and "how wrong can we be?".
- Usable capacity
- Nameplate battery capacity minus what sits below the system's cutoff voltage, self-discharge over the required life, ageing, and temperature effects. Roughly 80 % here, and stated as a derating so that disagreements about battery life are locatable.
- Requirement coupling
- A relationship in which meeting one requirement consumes another's budget — range costing transmit power costing average current, or board size limiting the cell that battery life depends on. The reason individually reasonable requirement sets can be jointly impossible.
ACTIVE RECALLCheck your understanding
Answer in your own words first, then reveal.
1Rewrite "the tag should have good battery life and work reliably indoors" as proper requirements, and say what you had to find out in order to write them.›
It is two requirements sharing a sentence, and neither is measurable, so split first. R1: "The tag shall operate for ≥ 6 months on a 1000 mAh LiPo cell at one report per minute and 22 °C ambient, verified by analysis from the measured average current, a 72-hour bench run, and the 30-day pilot." R2 and R9 together: "The tag shall deliver ≥ 90 % of scheduled reports to a gateway at ≥ 10 m through one plaster wall, at 0–45 °C indoors, verified by test with five units over 24 hours." Notice how much I had to decide in order to write those: the report interval, the reference temperature, the cell size, what "reliably" means as a delivery rate, how far, through what, how many units, and for how long. That is the real work — the writing is trivial once the decisions are made, and every one of those decisions was previously hiding inside the word "good" or "reliably", being assumed differently by each person who read it. It is also worth noticing that I added a delivery-rate number that the canon's R2 does not have; a real project would need it, and finding that gap is exactly what this exercise is for.
2Derive R1's allowed average current, then explain why the design's 38 µA and the canon's "3 years theoretical" are both correct.›
Top-down: six months at 730 hours per month is 4,380 h. The 1000 mAh nameplate is derated to about 800 mAh usable, for the capacity below the cutoff voltage, self-discharge over six months, and ageing — an illustrative 80 %. Allowed average current = 800 mAh / 4,380 h = 183 µA. Bottom-up: 8 µA of board sleep plus an active window of 40 mA for 45 ms once per 60 s, which is 40 × 0.045 / 60 = 30 µA, giving 38 µA. Margin 183 / 38 ≈ 4.8×. The canon's framing: 1000 mAh divided by 38 µA is about 26,000 hours, or three years — but that is on nameplate capacity with no derating at all, which is why the canon labels it theoretical. Six months is one sixth of three years, so the requirement survives a 6× derating for everything the bench does not show. The two framings differ only in where the derating is applied: the top-down number derates the battery and compares currents; the canon derates the result and compares durations. Both are correct, and quoting either without saying which one you used is how two engineers get four-times-different answers from the same hardware.
3A stakeholder asks for reports every 10 seconds instead of every 60, "since the battery lasts three years anyway". Answer them with the budget.›
The active term scales directly with reporting rate and it is 79 % of the budget, so this is close to a six-times increase on the dominant term. Sleep stays at 8 µA; the active contribution goes from 30 µA to 40 mA × 0.045 s / 10 s = 180 µA. New average: 188 µA, against R1's 183 µA allowance — so it fails R1, narrowly, and the three-year theoretical figure becomes about seven months on nameplate capacity, or just under six months usable. The answer is therefore "no at 10 seconds, and here is the boundary": setting the average to the full 183 µA allowance leaves 175 µA for the active term, which at 40 mA for 45 ms works out to one report every 10.3 seconds. So the fastest reporting this design can support is just over ten seconds, and the chart in this lesson's widget shows exactly where the curve crosses. Then offer the levers rather than just the refusal, because the stakeholder's need is probably real. Shorten the active window — fewer packets per burst, a smaller payload, a faster sensor read — since halving 45 ms halves the dominant term. Report on change rather than on schedule, using the motion interrupt that R4 already requires, which for an asset tag is usually what was actually wanted. Increase the cell, which then collides with R8 through the cell's physical size. Each of those is a trade with a number attached, which is a much better conversation than "that sounds hard".
4Beacon Rev A ships with three known defects. Explain why that is a defensible engineering decision and what would have made it indefensible.›
It is defensible because each defect was measured against a requirement and found not to violate it. The battery divider's extra 1.85 µA is a 23 % overrun on the 8 µA sleep term, but sleep is only 8 of 38 µA, so the average rises to about 40 µA and R1 — which had 4.8× of margin — is untouched. The button on GPIO9 confuses boot mode when held at power-up, which is a usability defect with a documented workaround and no requirement covering it, so it is a candidate for Rev B rather than a blocker. The rotated LED footprint is an assembly problem that costs rework time on 25 units, which is a schedule cost, not a requirement failure. In each case the sentence "this does not violate Rn, and here is the arithmetic" is available, and the decision to ship is a judgement made against a written standard. What would make it indefensible: not knowing. If the divider leak had been discovered in the field rather than measured at bring-up, or if there were no budget against which to judge whether 1.85 µA mattered, then shipping is not a decision but a gamble that happened to pay off. The requirements table is what converts "we shipped with defects" from an admission into an engineering position — and note that it only works because the table existed before the defects did.