Path 05 · Module 07 · Synthesis

System Trade Study &
Architecture Review

A precise score can still select the wrong radio. Establish what must pass, make preferences explicit, and ask which missing evidence could change the decision. Bring the gateway’s RF budgets into one defensible architecture review.

Compare
Zero IF / discrete IF100 / RF sampling
Decide
Feasibility → values → evidence
Carry forward
A conditional choice and one useful experiment
01 / 10

Failure: the highest score violates a must-pass requirement

The cheapest, fastest gateway has the highest score. Can it receive the required signal?

A review team compares three receive concepts. C, RF sampling, is cheap and quick in the illustrative allocation sheet. Its spreadsheet total looks compelling. But C’s conditional sensitivity is −110 dBm, and the selected receive mode requires −115 dBm or better. The concept misses a must-pass requirement by 5 dB. A schedule advantage cannot recover that lost sensitivity.

Frozen naive matrix · deliberate counterexample, never the production engine’s ranking
CandidateCost / scheduleNaive preference scoreHidden hard question
A · integrated zero IF18 units / 8 weeks0.546666667−117 dBm sensitivity
B · discrete IF10012 units / 12 weeks0.653333333−118.165007875 dBm sensitivity
C · RF sampling10 units / 6 weeks0.933333333 · invalid naive winner−110 dBm fails ≤−115 dBm
Think about itWould you carry C forward because its score is 0.933, or remove it before scoring?
Answer

Remove C from the eligible comparison. This frozen 0.933333333 is a counterexample calculated without feasibility. The real engine produces no score, no rank and no Pareto participation for C. A or B may advance as a conditional allocation; neither has hardware evidence closure.

Common misconceptionHard failures can be compensated by high preference scores.

A hard requirement defines the feasible set. Compensation is allowed only within the declared preference model after that gate. If the requirement is wrong, change it through an owned, explicit rebaseline and rerun every candidate.

Review artifact: a rejected naive matrix with its failure lineage: C-SENS → −110 dBm at R2 → −5 dB margin → excluded. Start the real comparison by freezing exactly what each column means.

02 / 10

Freeze scenario, assumptions, candidates, and common basis

Are the candidates solving the same problem at the same boundary?

Continue the fictional node and gateway from 05.1’s requirements baseline. This decision concerns gateway receive architecture. A node transmitter, a gateway receiver and shared-system infrastructure are separate power and cost boundaries. Sleeping for most of a telemetry period can reduce node average energy; it cannot reduce a gateway’s always-on receive power.

Common-basis sheet · frozen illustrative exercise, 2026-09-07
Boundary / quantityDeclared basisSource / open condition
Wanted waveform / R0 → R32.450 GHz selected QPSK channel; 20 kbit/s uncoded, no overhead; 10 ksymbol/sPORTFOLIO-v1 / Path 02; no named air interface selected
Signal support versus noise band13.5 kHz ideal total support; 20 kHz detector ENBW; 290 K reference05.3-ENBW; support is not an ENBW substitute
Sensitivity / R2 antenna feedR3 SNR criterion 8 dB + RX-IMPL 2 dB; ≤−115 dBm conditional requirement05.3 before optional ADC loss; design/production/fading reserves remain separate
Power / gateway electronicsSteady nominal RX electronics, ≤200 mWNode TX duty and shared-system power excluded, not assumed zero
Cost / assembled gatewayIllustrative cost units per assembled gateway at exercise quantity 1000; ≤25 unitsCOST-Q1000; no currency, quote, supplier or purchasing recommendation
Schedule / common completionWeeks from frozen project start until an architecture is ready for the same review; 4 best / 16 worst preference anchorsREVIEW-WEEKS; illustrative allocations, not actual dates or delivery promises
Market / antenna / channelPlaceholders with named future owners; no assumed authorization, realized gain or channel truthPaths 06/07/09; no legal or radio-technology conclusion
Candidate snapshot ledger · every summary cell has its own source and assumption
CandidateSensitivity dBm / sourceRX mW / sourceCost units / sourceSchedule weeks / sourceSelected-plan evidence
A · zero IF−117 / A-SENS100 / A-RX-POWER18 / A-COST-Q10008 / A-REVIEW-WEEKSA-SELECTED-PLAN-v1: supplied geometry clears; calibration open
B · IF100−118.165007875 / P05-M03-R2-NOMINAL160 / B-RX-POWER12 / B-COST-Q100012 / B-REVIEW-WEEKS05.2 selected IF plan with filtering assumptions; rejection evidence open
C · RF sampling−110 / C-SENS90 / C-RX-POWER10 / C-COST-Q10006 / C-REVIEW-WEEKSC-SELECTED-PLAN-v1: narrow preselection supplied; clock/amplitude evidence open

All A/C summary allocations belong to p05-m07-trade-fixture-v1. Their detailed noise chains are absent. They are not altered versions of B’s shared chain. B alone inherits the exact analog sensitivity from 05.3, with version p05-m03-receiver-ledger-v1. Do not attach its noise provenance to another topology.

Go deeperFreeze the completion definition, not just the unit

A prototype ready for an internal review and a product ready for certification are different schedule outcomes. Using “weeks” for both does not make the entries comparable. The same applies to bare IC price versus assembled gateway cost. Record scope, quantity, inclusions and omissions before normalizing values.

The common-basis sheet is now the review contract. The hard screen can compare quantities only when their conditions and definitions match it.

03 / 10

Screen hard constraints before scoring preferences

What is the difference between a modeled pass, a known failure and a missing answer?

m=limitsupplied valuefailtakesprecedenceunknowndefersallpasspermitspreferences\begin{aligned}m &= \mathrm{limit} - \text{supplied value} \\ &\mathrm{fail} \mathrm{takes} \mathrm{precedence} \to \mathrm{unknown} \mathrm{defers} \to \mathrm{all} \mathrm{pass} \mathrm{permits} \mathrm{preferences}\end{aligned}For these maximum limits, margin ≥0 passes, including equality. A dBm sensitivity subtraction gives dB. No arithmetic is performed on an unknown value.
Default hard-screen artifact · local supplied-model feasibility
CandidateSensitivity margin dBRX power margin mWCost margin unitsGeometry / interfacesEligibility
A+2.000+100+7Supplied pass / passEligible allocation
B+3.165007875+40+13Supplied pass / passEligible allocation
C−5.000+110+15Supplied pass / passExcluded: sensitivity fails

The word “pass” here means the declared numerical or typed interface screen clears. It does not mean measured product compliance. An absent sensitivity, unsupported unit or required source metadata is unknown. Invalid data such as an out-of-domain number suppresses evaluation altogether; it must not become zero or an apparently safe unknown.

Think about itThe review now requires verified performance beside the co-located blocker. Does changing “illustrative” to “measured” let B pass?
Answer

No. A source/configuration/method assertion can be recorded, but the studio has no verified hardware record. A and B become unknown for the required blocker evidence. C retains its known sensitivity failure. No candidate is eligible: defer all selection and request the discriminating blocker evidence.

Common misconceptionAn unknown is a conservative pass, or a waiver can be buried in a weight.

Unknown is an unresolved hard question. A proposed waiver needs the changed requirement, rationale, decision authority, affected modes and owner. Logging that proposal does not change feasibility. Only an explicit new baseline can do that.

Review artifact: the signed-margin screen plus each candidate’s excluded, deferred or eligible state. Keep different margin units separate; “40 mW versus 2 dB” is not a normalized risk comparison. Next, assign value only to outcomes you are actually permitted to trade.

04 / 10

Define independent criteria and value functions

What does one unit of preference mean, and does it change when a candidate disappears?

A value function maps an outcome in its own units onto a disclosed 0…1 preference scale. Here cost and schedule are distinct: assembled gateway unit cost excludes project delay valuation, while schedule counts elapsed weeks to the common review. Adding “cheapness,” “BOM value” and “cost saving” as three separate criteria would count the same benefit repeatedly.

vcost(c)=clip(25c15,0,1)vschedule(t)=clip(16t12,0,1)Score=wivi,only for eligible candidates\begin{aligned}v_{\mathrm{cost}}(c) &= \operatorname{clip}(\frac{25-c}{15}, 0, 1) \\ v_{\mathrm{schedule}}(t) &= \operatorname{clip}(\frac{16-t}{12}, 0, 1) \\ \text{Score} &= \sum w_{i}v_{i}, \text{only for eligible candidates}\end{aligned}clip(x,0,1) explicitly saturates preference beyond the anchors; it does not clamp raw inputs or relax hard limits. Higher value is always preferred.
Value-function artifact · fixed anchors and nonoverlapping ownership
CriterionBest → 1 / worst → 0A valueB valueOwner / excluded meaning
Cost10 / 25 exercise cost units7/15 = 0.46666666713/15 = 0.866666667Cost lead; excludes schedule delay, service effort and shared-system scope
Schedule4 / 16 weeks2/3 = 0.6666666671/3 = 0.333333333Program lead; common review milestone, not lifetime service time

The hard cost cap of 25 and the worst preference anchor of 25 happen to share a number. They play different roles. Changing one must not silently change the other. Do not normalize against the cheapest or fastest visible candidate: adding irrelevant D would then move everyone’s scale without changing any engineering outcome.

Go deeperLinear and tabulated preferences require a defensible model

An additive model assumes the chosen attributes and preference tradeoffs can be represented separately on these scales; statistical independence of cost and schedule is a different question. If the value of a faster schedule changes sharply with cost, a constant additive trade may be unsuitable. The studio can expose a monotonic piecewise-linear curve, with strictly ordered anchors, but it cannot prove preferential independence. Reframe overlapping criteria and document interactions before using their sum.

For example, schedule anchors (4 weeks,1), (10 weeks,0.8), (16 weeks,0) give a visible bend. This expresses a human preference for reaching a useful milestone by week 10; it is not empirical evidence about schedule uncertainty.

Common misconceptionThe highest weighted score is objectively best.

The score answers a conditional question under chosen values and weights. It does not establish technical superiority or business truth. An architecture outside the assumed operating mode needs a different comparison basis.

The value-function table makes the scales inspectable. The next artifact states whose preferences those scales represent.

05 / 10

Assign weights with rationale and ownership

Does 60% cost mean cost is “more important” regardless of its range?

Weights describe the relative value of moving across each declared best-to-worst range. In this exercise, the review chair gives a full 15-unit cost swing weight 0.60 and a full 12-week schedule swing weight 0.40. If you change the anchors, revisit the swing rationale. Equal weights are also a choice; they are not neutral.

Weight rationale and gaming review
EntryOwner / rationaleReview challenge
Cost: 0.60Cost lead; a complete 25→10 unit swing carries 60% of modeled preferenceAre assembly/test yield and quantity consistent? Has cost been counted elsewhere?
Schedule: 0.40Program lead; a complete 16→4 week swing carries 40%Are both candidates ready for the same review, with the same evidence maturity?
Adoption authorityArchitecture review chairPublish changes to anchors/weights before seeing the desired winner; retain dissent.
A=0.60(715)+0.40(23)=0.546666667B=0.60(1315)+0.40(13)=0.653333333\begin{aligned}A &= 0.60(\frac{7}{15}) + 0.40(\frac{2}{3}) = 0.546666667 \\ B &= 0.60(\frac{13}{15}) + 0.40(\frac{1}{3}) = 0.653333333\end{aligned}A and B already passed the supplied hard screens. These fixtures use unrounded fractions; nine digits are checking precision, not evidence quality.

Require every weight to be nonnegative and the sum to equal one within 10⁻⁹. The studio rejects invalid weights. Its explicit Normalize action shows before/after values; a zero vector cannot be normalized. The single-weight sweep distributes the remaining mass in the other weights’ prior proportions. If those weights previously sum to zero, you must supply a redistribution instead of letting the engine invent priorities.

Common misconceptionSensitivity analysis means changing one weight and quietly renormalizing everything.

That can change more than the intended preference. Use the declared proportional remainder rule and record it. A view-only cursor should never alter the committed comparison.

Review artifact: signed-off swing rationale and a change log. Before trusting either total, inspect what the supporting evidence actually establishes.

06 / 10

Separate evidence maturity from performance

Does a well-documented weak design become a strong design if you add an evidence-confidence score?

Evidence maturity describes what supports a claim. Performance describes the claim itself. Mixing a maturity ordinal into the performance sum hides both. A synthetic −118 dBm allocation is not measured receiver behavior, regardless of how many decimals it contains.

Independent evidence matrix · default hardware questions remain open
CandidateSupplied basisAbsent evidenceWhat would change eligibility?
ASynthetic sensitivity/power/cost/schedule; selected-channel geometry suppliedI/Q and DC calibration, wanted-plus-blocker behavior, temperature/mode cornersAn explicit hard-evidence requirement makes the absent row unknown.
B05.3 analog sensitivity; 05.2 IF100 plan with filtering assumptions; synthetic power/cost/scheduleActual filter rejection, mode-specific blocker/AGC response, converter and driver validationFailure or incompatible version in an imported hard row excludes/defers B.
CSynthetic RF sensitivity/power/cost/schedule; narrow preselected geometryAnalog-input clock/driver and full sensitivity modelA new traceable sensitivity allocation must clear the hard limit; evidence labels alone cannot repair −5 dB.

The source/configuration/method fields make a user assertion traceable. They do not verify it. A reviewer would still need the exact DUT state, stimulus, plane, bandwidth, instrument method, calibration and uncertainty. The studio deliberately keeps a requested verified-blocker row unknown because no such record is supplied.

Common misconceptionMore decimal places mean stronger evidence.

B’s extra decimals make the inherited mathematical fixture reproducible. They do not reduce production spread, model discrepancy or measurement uncertainty. Learner scores use three decimals; raw values and source IDs remain in the local record.

Go deeperAllocation review and hardware release have different gates

An allocation study may legitimately rank modeled candidates while carrying every unclosed risk. A hardware release can require measured evidence as a hard condition. Neither review is improved by pretending the two bases are the same. Name the gate in the decision record, and reopen the decision when its basis changes.

The evidence matrix now exposes missing truth without distorting value. Test whether the preference recommendation survives reasonable changes.

07 / 10

Test sensitivity to weights and uncertain inputs

How far must your preferences or assumptions move before the decision changes?

Think about itAt cost weight 0.30 and schedule weight 0.70, which eligible candidate leads?
Answer

A scores 0.606666667 and B scores 0.493333333. A’s faster review schedule now outweighs B’s cost benefit. Both remain feasible allocations. This is a disclosed preference reversal, not inconsistent engineering.

ScoreAScoreB=131115wTie at w=1/311/15=511=0.4545454545\begin{aligned}\mathrm{Score}_A-\mathrm{Score}_B&=\frac13-\frac{11}{15}w\\\text{Tie at }w&=\frac{1/3}{11/15}=\frac5{11}=0.4545454545\ldots\end{aligned}w is the cost weight and 1−w the schedule weight. The difference is linear under the fixed value functions.
Exact crossover and eligible Pareto artifact · comparison tolerance 10⁻⁹
CaseResultInterpretation
0 ≤ w_cost < 5/11A leads outside the score-tolerance tie bandA faster; B cheaper. Both are nondominated.
w_cost = 5/11A = B = 19/33 = 0.575757576Equality is a tie; no arbitrary topology preference.
5/11 < w_cost ≤ 1B leads outside the score-tolerance tie bandDefault w=.60 favors B.
Optional D: −116 dBm, 180 mW, cost 20, schedule 14Eligible; v=(1/3,1/6); default score 0.266666667A and B both dominate D in preferences. D is loaded only in an explicit comparison.
Add, remove or reorder DA/B scores and crossover unchangedFixed anchors and stable IDs prevent candidate-set artifacts.

Dominance means no worse in every consistently oriented value and strictly better in at least one, within 10⁻⁹. Equal vectors are ties. C cannot suppress A/B’s Pareto status because it is ineligible. D’s dominance says nothing about unspecified RF physics; it says its disclosed cost and schedule are inferior to both eligible alternatives.

Interval corner bounds · default weights, hard quantities initially fixed
CandidateCost bound / schedule boundAnalytical preference envelopeCorner derivation
A[16,20] units / [6,10] weeks[0.400000000, 0.693333333]Worst: .6(5/15)+.4(6/12); best: .6(9/15)+.4(10/12)
B[10,16] units / [9,15] weeks[0.393333333, 0.833333333]Worst: .6(9/15)+.4(1/12); best: .6(15/15)+.4(7/12)
B hard-bound variantSensitivity [−119,−114] dBm at R2Not robustly feasible against ≤−115 dBmNominal point is distinct from range. Each failed scenario excludes B before scoring.

Overlapping envelopes mean the default winner is not robust across all independent interval combinations. Samples help inspect the pattern but cannot replace endpoint bounds. In the hard-bound variant, averaging B’s passing and failing sensitivities would conceal a real requirement violation.

n=dkbkub(n)=dkbk1base2:12,14,34,18base3:13,23,19,49\begin{aligned}n &= \sum d_{k}b^{k} \to u_{b}(n) = \sum d_{k}b^{-k-1} \\ &\text{base} 2: \frac{1}{2}, \frac{1}{4}, \frac{3}{4}, \frac{1}{8} \\ &\text{base} 3: \frac{1}{3}, \frac{2}{3}, \frac{1}{9}, \frac{4}{9}\end{aligned}Halton radical inverse expands positive integer n in base b. Use n=1…256, not zero; map x=lo+(hi−lo)u. Bases A.cost=2, A.schedule=3, B.cost=5, B.schedule=7.

The studio reserves dimensions for all five candidate IDs and all six input quantities, so reorder and deletion do not change the source realization. Each sample reuses its cost realization in the cost hard limit and value function. It compares 256 and 1024 points as a coverage sensitivity check. A leader fraction means fraction of explored states only; there is no calibrated probability of success or statistical confidence interval.

Think about itThree optimistic assumptions move together. Do they provide three independent reasons to trust the choice?
Answer

No. A shared source can move cost and schedule in the same direction. In the shared-u preset, every uncertain cell uses the same u. In the opposed variant A/C/E use u and B/D use 1−u. These are explicit comoving/opposed scenario constructions, not inferred Pearson correlations. They change which joint cases are explored, without turning the bounds into probabilities.

Go deeperWhat the sample count does and does not establish

A deterministic low-discrepancy sequence spreads points through a box. It does not establish that the box is correct, all inputs are independent, or frequency equals probability. Report sampled extrema alongside analytical corner bounds, and test plausible shared-source structures. A larger count cannot repair a missing physical model or an inconsistent reference plane.

The sensitivity artifact identifies reversible preferences and nonrobust hard bounds. The next step is to check that the budgets being compared can coexist in one system.

08 / 10

Reconcile every RF interface and budget

Which apparently harmless copy across a spreadsheet boundary invalidates the review?

The workbook imports bounded typed snapshots, not live calculators or a backend. Every row names quantity, plane, unit, bandwidth/rate/power definition, mode, waveform, configuration, criterion, source, model version and evidence. Compare quantities only under their exact matching condition keys; convert supported units only after that check.

System crosswalk · retain the six preceding allocations and their open boundaries
Owner / imported artifactReview reconciliationStill outside this model
05.1 · requirementsR2 sensitivity criterion, gateway RX power/cost, mode, owner and evidence requestProduct claims are conditional, not verified
05.2 · frequency planB selected 2.450 GHz → IF100, explicit filtering and admitted image assumptionsReal rejection and spur amplitude
05.3 · noise/gain42.5 dB nominal R2→A1 gain; R2 −118.165007875 dBm; R3 ENBW 20 kHz; RX-IMPL onceOptional converter loss and other mode corners require recomputation
05.4 · blocker / AGCWanted, blocker offsets, per-tone/total levels, first overloaded stage and settling stateNo verified co-located threat result is imported
05.5 · node TXRequested linear +10 dBm R2 requires +12 dBm before 2 dB post-PA loss; actual compression qualifier remainsNode energy, antenna acceptance/realized gain and shared power kept separate
05.6 · A1 / A0 / D3Actual analog frequency for jitter; 80 MS/s physical A0 versus 80 kcomplex-sample/s D3; no SINAD double countRF driver validity, calibrated narrowband clock contribution and hardware current
Paths 06 / 07 / 08 / 09 / 10Owned requests for antenna/channel, named technology, measurement, market and implementation evidenceNo placeholder is silently turned into a pass
Eight frozen interface rules · triggering and repaired fixture pairs
RuleContradiction / consequenceExplicit repair
IF-01 sample-rate-boundaryD3 80 ksample/s copied as A0 80 MS/s; decimation omitted.80 MS/s at A0 → filter and decimate by 1000 → 80 ksample/s at D3. The factor and filtering are supplied, not a relabeling.
IF-02 power-plane+10 dBm PA output copied to R2 despite 2 dB loss.Separate repair example: +10 dBm actual PA output − 2 dB loss = +8 dBm actual R2. 05.5 instead requests +10 dBm R2, requiring +12 dBm before compression.
IF-03 bandwidth-type13.5 kHz signal support substituted for 20 kHz ENBW.Restore 20 kHz detector ENBW; retain 13.5 kHz wanted support as a different row.
IF-04 margin-duplicateRX-IMPL=2 dB charged in its original row and again as reserve.RX-IMPL=2 dB is already included. Do not add RX-IMPL again in reserve.
IF-05 clock-jitter2.45 GHz RF jitter computed at 30 MHz alias; total plus its own components.Use 2450 MHz analog input, not its 30 MHz alias. Replace components (.8 ps and .6 ps) with 1 ps total, or RSS components alone.
IF-06 evidence-overlapSupplied SINAD plus included quantization/noise.Use supplied SINAD under its sine-test conditions. Do not add its included ADC-Q or ADC-NOISE again; no narrowband truth is inferred.
IF-07 mode-boundarySleeping-node power copied to always-on gateway RX.Carry the always-on gateway RX boundary. Keep sleeping-node energy and shared-system power separate.
IF-08 source-stalenessDependent p05-m07-trade-fixture-v0 differs from reviewed parent.Recompute against the reviewed parent version and record that source. Age alone is not an RF failure.

These rules block only their declared dependent candidates. The supplied B review crosswalk includes clearly named node-TX and RF-jitter side examples; it does not silently replace B’s IF100 physical model. The studio shows exactly which candidate each diagnostic affects. Missing metadata or an unsupported unit is unknown; known incompatible plane/mode/definition is a contradiction. A stale parent version requests recomputation, not a physical failure based merely on elapsed age.

Common misconceptionIf the units match, the budgets reconcile.

dBm at a PA package and dBm at an antenna feed need a loss transformation. Two values in kHz can represent signal support and noise integration. Preserve meaning before arithmetic; a keyword engine cannot certify a free-form RF budget.

Review artifact: the crosswalk and repaired source lineage, copied and printed with stable IDs. With contradictions exposed, choose the smallest experiment that can change the decision.

09 / 10

Retire risk with a discriminating experiment

Which open question is worth answering before the next architecture commitment?

Prioritize experiments lexicographically: first a missing hard condition for a currently eligible candidate, next a known hard failure that could remove or restore an alternative, then a preference refinement. Within a class, use lower declared effort, then stable experiment ID. The effort values below are illustrative engineer-days, and the time boxes are exercise allocations. They are not financial value-of-information estimates.

Experiment priority artifact · inspectable human policy, stable IDs
Order / experimentHard relevance / effortWhat can change?
1 · EXP-01 Co-located-blocker comparisonUnclosed hard evidence; 2 engineer-days / 1 weeksA fails / B clears; B fails / A clears; both fail; or neither result discriminates within uncertainty.
2 · EXP-02 Clock and driver at actual RF inputKnown hard failure investigation; 3 engineer-days / 2 weeksReject C if actual 2.45 GHz clock/driver performance cannot support the required budget; otherwise rebuild its sensitivity allocation before re-entry.
3 · EXP-03 Refine assembled-gateway BOMPreference refinement; 1 engineer-days / 0.5 weeksNarrow A [16,20] and B [10,16] under quantity 1000; rerun if relative value reverses or cost crosses the cap.
RISK-BLOCK · EXP-01 · owner Receiver lead

Co-located-blocker comparison

Candidate / requirement: A / B · Wanted criterion under a declared co-located threat. Consequence: Gateway loses telemetry beside a nearby transmitter.

Uncertain assumption: Selected plan and nominal gain preserve the wanted criterion under the threat. Evidence gap: No verified wanted-plus-blocker evidence.

Discriminating outcomes: A fails / B clears; B fails / A clears; both fail; or neither result discriminates within uncertainty.

Acceptance / stop: Agree blocker offsets, level, waveform, coupling and wanted criterion before testing. Stop selection if every candidate fails; defer if test validity is unresolved.

Commitment: Prototype/gain-state choice is reversible; tooling and committed hardware are not. Bounded effort: 2 engineer-days within 1 weeks from the evidence request, illustrative only.

RISK-RF-CLOCK · EXP-02 · owner Converter lead

Clock and driver at actual RF input

Candidate / requirement: C · R2 sensitivity and A1 timing/driver allocations. Consequence: Aliased geometry looks clean while RF information is corrupted.

Uncertain assumption: A revised RF-sampling budget could recover at least 5 dB. Evidence gap: No compatible RF driver or clock measurement; C already fails nominal sensitivity.

Discriminating outcomes: Reject C if actual 2.45 GHz clock/driver performance cannot support the required budget; otherwise rebuild its sensitivity allocation before re-entry.

Acceptance / stop: Use actual RF input, voltage, band and clock inclusion metadata. Stop RF sampling if the revised hard screen fails.

Commitment: Bench characterization is reversible; selecting an RF ADC and PCB stack is costly to reverse. Bounded effort: 3 engineer-days within 2 weeks from the evidence request, illustrative only.

RISK-COST · EXP-03 · owner Cost lead

Refine assembled-gateway BOM

Candidate / requirement: A / B · Preference interval and cost cap. Consequence: Preference ranking reverses after scope correction.

Uncertain assumption: Both assembly estimates use the same boundary. Evidence gap: Illustrative allocations only, no costed BOM.

Discriminating outcomes: Narrow A [16,20] and B [10,16] under quantity 1000; rerun if relative value reverses or cost crosses the cap.

Acceptance / stop: Same assembled scope, test yield and quantity; no quotes fabricated. Stop comparison if scope differs.

Commitment: Estimate refinement is reversible; purchase commitments are not. Bounded effort: 1 engineer-days within 0.5 weeks from the evidence request, illustrative only.

For EXP-01, agree the threat frequency, level, waveform, duty and coupling before running the comparison. Use the same wanted criterion and gain-control condition for A and B. If B fails while A clears, the experiment can reverse the conditional allocation choice. If both fail, stop selection and revisit the architecture or requirement basis. If both clear, retain the preference study and close only the tested corner.

Common misconceptionA risk is retired by adding margin everywhere.

Margin can reserve capacity for an allocated effect. It cannot prove that a receiver survives an unspecified blocker or that three dependent assumptions are correct. A risk-retirement action needs an observation that distinguishes the alternatives and an acceptance/stop condition.

The risk register and experiment card state what could reverse the decision, who will obtain that evidence, and what commitment waits for it. Now conduct the review.

10 / 10

Conduct and record the architecture review

Can another engineer reproduce your decision—and explain what would change it?

Start with A/B/C. Verify C is excluded, then change weights to 0.30/0.70 and evaluate. Use the exact crossover key point, compare the explicit D preset, and require verified blocker evidence to defer all. Explore bounded inputs, then B’s hard-bound variant. Inject each interface contradiction and repair it. Finally, select one evidence request and copy the committed record.

Interactive allocation study · Class 1

Constraint-First Trade Studio

Choose a comparison basis, screen the hard requirements, then explore preference reversals. The default A/B/C study is already evaluated. Edits stay in a draft until you select Evaluate.

Named presets atomically restore candidates, limits, ranges, source rows, weights and decision metadata.
Hard limits · gateway steady RX electronics
-150…-80; step 0.1. Reset restores the fixture.
1…1000; step 1. Reset restores the fixture.
0…100; step 0.1. Reset restores the fixture.
No verified hardware record is supplied. Requiring this evidence makes the row unknown, even with a measurement assertion.

Candidates · 3 of 5

Numbers are illustrative allocation snapshots. Cost means units per assembled gateway at exercise quantity 1000; schedule uses the same frozen project start and ready-for-review definition. Open a candidate to edit. Missing means explicitly unknown; clearing a numeric field without declaring unknown is invalid.

A · Integrated zero IF · candidate allocation and evidence
Required; 1–80 characters. Stored only in this unsaved local record.
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
R2 sensitivity · source A-SENS / p05-m07-trade-fixture-v1
-150…-80 dBm; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Gateway steady RX electronics power · source A-RX-POWER / p05-m07-trade-fixture-v1
1…1000 mW; step 1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Assembled gateway cost · source A-COST-Q1000 / p05-m07-trade-fixture-v1
0…100 exercise cost units; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
0…100; step 0.1. Reset restores the fixture.
0…100; step 0.1. Reset restores the fixture.
Ready-for-review schedule · source A-REVIEW-WEEKS / p05-m07-trade-fixture-v1
1…52 weeks; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
1…52; step 0.1. Reset restores the fixture.
1…52; step 0.1. Reset restores the fixture.
A evidence · independent of value scores
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
B · Discrete IF100 · candidate allocation and evidence
Required; 1–80 characters. Stored only in this unsaved local record.
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
R2 sensitivity · source P05-M03-R2-NOMINAL / p05-m03-receiver-ledger-v1
-150…-80 dBm; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Gateway steady RX electronics power · source B-RX-POWER / p05-m07-trade-fixture-v1
1…1000 mW; step 1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Assembled gateway cost · source B-COST-Q1000 / p05-m07-trade-fixture-v1
0…100 exercise cost units; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
0…100; step 0.1. Reset restores the fixture.
0…100; step 0.1. Reset restores the fixture.
Ready-for-review schedule · source B-REVIEW-WEEKS / p05-m07-trade-fixture-v1
1…52 weeks; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
1…52; step 0.1. Reset restores the fixture.
1…52; step 0.1. Reset restores the fixture.
B evidence · independent of value scores
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
C · RF sampling · candidate allocation and evidence
Required; 1–80 characters. Stored only in this unsaved local record.
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
R2 sensitivity · source C-SENS / p05-m07-trade-fixture-v1
-150…-80 dBm; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Gateway steady RX electronics power · source C-RX-POWER / p05-m07-trade-fixture-v1
1…1000 mW; step 1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Assembled gateway cost · source C-COST-Q1000 / p05-m07-trade-fixture-v1
0…100 exercise cost units; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
Ready-for-review schedule · source C-REVIEW-WEEKS / p05-m07-trade-fixture-v1
1…52 weeks; step 0.1. Nominal point allocation; B’s inherited exact sensitivity may be off step.
Intervals are bounds, not confidence limits. They are used only when scenario exploration is enabled.
C evidence · independent of value scores
Closed choices. A state or evidence label alone is not hardware verification.
Closed choices. A state or evidence label alone is not hardware verification.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
Required; 1–160 characters. Stored only in this unsaved local record.
Preferences · fixed scales, explicit owners

2–4 nonoverlapping criteria. Footprint and service time have no supplied candidate allocations; adding them requires those values before eligible scoring. No free-form formulas are executed.

Assembled gateway cost · value function and weight
0…1; step 0.01. Reset restores the fixture.
Changing direction requires matching anchors. Preferences are human-defined; explain any unusual direction.
0…100; step 0.1. Reset restores the fixture.
0…100; step 0.1. Reset restores the fixture.
Clipped at declared endpoint values. Tabulated curves linearly interpolate each segment; anchors must be ordered and monotonic.
Required; 1–300 characters. Stored only in this unsaved local record.
Required; 1–300 characters. Stored only in this unsaved local record.
Ready-for-review schedule · value function and weight
0…1; step 0.01. Reset restores the fixture.
Changing direction requires matching anchors. Preferences are human-defined; explain any unusual direction.
1…52; step 0.1. Reset restores the fixture.
1…52; step 0.1. Reset restores the fixture.
Clipped at declared endpoint values. Tabulated curves linearly interpolate each segment; anchors must be ordered and monotonic.
Required; 1–300 characters. Stored only in this unsaved local record.
Required; 1–300 characters. Stored only in this unsaved local record.

Draft weight sum: 1.000000000. Must equal 1 within 10⁻⁹.

Sensitivity exploration
The named B sensitivity preset supplies [−119,−114] dBm. This mode selector alone never rewrites intervals.
Deterministic source grouping, not a statistical correlation coefficient.
Indices start at 1. 256 and 1024 summaries are also compared; no probability inferred.
Other weights retain their prior proportions. A zero prior remainder requires explicit redistribution.
Inject and repair typed interface contradictions

The supplied crosswalk rows are B’s imported review dependencies, including explicitly named side examples. Every repair restores a complete typed fixture and transform. Free-form text is never certified.

IF-01 · sample-rate-boundary

05.6-A002-D3; 80 MS/s at A0 → filter and decimate by 1000 → 80 ksample/s at D3. The factor and filtering are supplied, not a relabeling.

IF-02 · power-plane

TX-PA-EXAMPLETX-R2-EXAMPLE; Separate repair example: +10 dBm actual PA output − 2 dB loss = +8 dBm actual R2. 05.5 instead requests +10 dBm R2, requiring +12 dBm before compression.

IF-03 · bandwidth-type

05.3-ENBWREVIEW-ENBW; Restore 20 kHz detector ENBW; retain 13.5 kHz wanted support as a different row.

IF-04 · margin-duplicate

05.3-IMPLREVIEW-IMPL; RX-IMPL=2 dB is already included. Do not add RX-IMPL again in reserve.

IF-05 · clock-jitter

RF-ANALOG-INPUTJITTER-EVALUATION; Use 2450 MHz analog input, not its 30 MHz alias. Replace components (.8 ps and .6 ps) with 1 ps total, or RSS components alone.

IF-06 · evidence-overlap

ADC-SINADREVIEW-ADC-EXTRA; Use supplied SINAD under its sine-test conditions. Do not add its included ADC-Q or ADC-NOISE again; no narrowband truth is inferred.

IF-07 · mode-boundary

GW-RX-BOUNDARYREVIEW-GW-POWER; Carry the always-on gateway RX boundary. Keep sleeping-node energy and shared-system power separate.

IF-08 · source-staleness

REVIEW-PARENTDEPENDENT-RECOMPUTATION; Recompute against the reviewed parent version and record that source. Age alone is not an RF failure.

Decision owner, rationale, dissent and evidence request
Required; 1–1000 characters. Stored only in this unsaved local record.
Required; 1–1000 characters. Stored only in this unsaved local record.
Required; 1–1000 characters. Stored only in this unsaved local record.
Required; 1–1000 characters. Stored only in this unsaved local record.
Selection is an unsaved local record, not a calendar event or message.

Conditional allocation leader: B. Hardware evidence remains open; this is not product compliance.

Evaluate the hard rows first. A passing synthetic allocation is eligible only for this study. Calibration, rejection, blocker, clock and driver evidence remain visible in the risk register.

Decision boundary
Gateway · steady RX · nominal mode
Wanted channel / detector
2.450 GHz QPSK · 13.5 kHz support / 20 kHz ENBW
Feasible allocation set
A, B
Preference Pareto set
A, B
Hard screen → eligible preference comparison · full precision before display
Candidate / stateHard reasons / signed marginsValues in criterion orderScore / rankPreference dominance
A · Integrated zero IF · passAll supplied screens clear. sensitivity: 2.000 dB; power: 100.000 mW; cost: 7.000 exercise cost units. Different units remain separate.0.467 / 0.6670.547 / 2Nondominated
B · Discrete IF100 · passAll supplied screens clear. sensitivity: 3.165 dB; power: 40.000 mW; cost: 13.000 exercise cost units. Different units remain separate.0.867 / 0.3330.653 / 1Nondominated
C · RF sampling · failsensitivity: -110 > -115 dBm; violates the closed limit.Not evaluated— / excluded or deferredOutside Pareto comparison
Inspect every hard row, signed margin and source
Signed margin = limit − supplied value; nonnegative passes closed numerical limits
Candidate / requirementState / marginCondition / reasonSource / fixture
A / sensitivitypass / 2.000000 dB-117 ≤ -115 dBm; clears the closed limit.A-SENS / p05-m07-trade-fixture-v1
A / powerpass / 100.000000 mW100 ≤ 200 mW; clears the closed limit.A-RX-POWER / p05-m07-trade-fixture-v1
A / costpass / 7.000000 exercise cost units18 ≤ 25 exercise cost units; clears the closed limit.A-COST-Q1000 / p05-m07-trade-fixture-v1
A / geometrypass / — stateSupplied geometry screen only; hardware truth remains open.A-SELECTED-PLAN-v1
A / interfacepass / — stateSupplied interface screen only; hardware truth remains open.A-SELECTED-PLAN-v1
B / sensitivitypass / 3.165008 dB-118.165007875 ≤ -115 dBm; clears the closed limit.P05-M03-R2-NOMINAL / p05-m03-receiver-ledger-v1
B / powerpass / 40.000000 mW160 ≤ 200 mW; clears the closed limit.B-RX-POWER / p05-m07-trade-fixture-v1
B / costpass / 13.000000 exercise cost units12 ≤ 25 exercise cost units; clears the closed limit.B-COST-Q1000 / p05-m07-trade-fixture-v1
B / geometrypass / — stateSupplied geometry screen only; hardware truth remains open.B-SELECTED-PLAN-v1
B / interfacepass / — stateSupplied interface screen only; hardware truth remains open.B-SELECTED-PLAN-v1
B / IF-01pass / — typed comparison80 MS/s at A0 → filter and decimate by 1000 → 80 ksample/s at D3. The factor and filtering are supplied, not a relabeling.05.6-A0 → 02-D3 → IF-01-TRANSFORM-v1
B / IF-02pass / — typed comparisonSeparate repair example: +10 dBm actual PA output − 2 dB loss = +8 dBm actual R2. 05.5 instead requests +10 dBm R2, requiring +12 dBm before compression.TX-PA-EXAMPLE → TX-R2-EXAMPLE → IF-02-TRANSFORM-v1
B / IF-03pass / — typed comparisonRestore 20 kHz detector ENBW; retain 13.5 kHz wanted support as a different row.05.3-ENBW → REVIEW-ENBW → IF-03-TRANSFORM-v1
B / IF-04pass / — typed comparisonRX-IMPL=2 dB is already included. Do not add RX-IMPL again in reserve.05.3-IMPL → REVIEW-IMPL → IF-04-TRANSFORM-v1
B / IF-05pass / — typed comparisonUse 2450 MHz analog input, not its 30 MHz alias. Replace components (.8 ps and .6 ps) with 1 ps total, or RSS components alone.RF-ANALOG-INPUT → JITTER-EVALUATION → IF-05-TRANSFORM-v1
B / IF-06pass / — typed comparisonUse supplied SINAD under its sine-test conditions. Do not add its included ADC-Q or ADC-NOISE again; no narrowband truth is inferred.ADC-SINAD → REVIEW-ADC-EXTRA → IF-06-TRANSFORM-v1
B / IF-07pass / — typed comparisonCarry the always-on gateway RX boundary. Keep sleeping-node energy and shared-system power separate.GW-RX-BOUNDARY → REVIEW-GW-POWER → IF-07-TRANSFORM-v1
B / IF-08pass / — typed comparisonRecompute against the reviewed parent version and record that source. Age alone is not an RF failure.REVIEW-PARENT → DEPENDENT-RECOMPUTATION → IF-08-TRANSFORM-v1
C / sensitivityfail / -5.000000 dB-110 > -115 dBm; violates the closed limit.C-SENS / p05-m07-trade-fixture-v1
C / powerpass / 110.000000 mW90 ≤ 200 mW; clears the closed limit.C-RX-POWER / p05-m07-trade-fixture-v1
C / costpass / 15.000000 exercise cost units10 ≤ 25 exercise cost units; clears the closed limit.C-COST-Q1000 / p05-m07-trade-fixture-v1
C / geometrypass / — stateSupplied geometry screen only; hardware truth remains open.C-SELECTED-PLAN-v1
C / interfacepass / — stateSupplied interface screen only; hardware truth remains open.C-SELECTED-PLAN-v1
Committed point allocations, intervals and cell provenance
Committed inputs · nominal point is separate from its bound; gateway RX / quantity 1000 / common review milestone
Candidate / quantity / planeNominal point / unitSupplied intervalSource / model version
A / R2 sensitivity / R2 → R3 criterion-117 dBmpoint onlyA-SENS / p05-m07-trade-fixture-v1
A / Gateway steady RX electronics power / gateway steady RX100 mWpoint onlyA-RX-POWER / p05-m07-trade-fixture-v1
A / Assembled gateway cost / assembled gateway18 exercise cost units[16, 20] exercise cost unitsA-COST-Q1000 / p05-m07-trade-fixture-v1
A / Ready-for-review schedule / project start → ready for review8 weeks[6, 10] weeksA-REVIEW-WEEKS / p05-m07-trade-fixture-v1
B / R2 sensitivity / R2 → R3 criterion-118.165007875 dBmpoint onlyP05-M03-R2-NOMINAL / p05-m03-receiver-ledger-v1
B / Gateway steady RX electronics power / gateway steady RX160 mWpoint onlyB-RX-POWER / p05-m07-trade-fixture-v1
B / Assembled gateway cost / assembled gateway12 exercise cost units[10, 16] exercise cost unitsB-COST-Q1000 / p05-m07-trade-fixture-v1
B / Ready-for-review schedule / project start → ready for review12 weeks[9, 15] weeksB-REVIEW-WEEKS / p05-m07-trade-fixture-v1
C / R2 sensitivity / R2 → R3 criterion-110 dBmpoint onlyC-SENS / p05-m07-trade-fixture-v1
C / Gateway steady RX electronics power / gateway steady RX90 mWpoint onlyC-RX-POWER / p05-m07-trade-fixture-v1
C / Assembled gateway cost / assembled gateway10 exercise cost unitspoint onlyC-COST-Q1000 / p05-m07-trade-fixture-v1
C / Ready-for-review schedule / project start → ready for review6 weekspoint onlyC-REVIEW-WEEKS / p05-m07-trade-fixture-v1
Frozen preferences · no normalization against visible candidates
Criterion / unitOwner / rationaleBest / worst / curveWeight
Assembled gateway cost / exercise cost unitsCost lead: A full 15-unit cost improvement is worth 0.60 of the total preference swing.10 → 1; 25 → 0; linear0.600000
Ready-for-review schedule / weeksProgram lead: A full 12-week schedule improvement is worth 0.40; labor cost is excluded.4 → 1; 16 → 0; linear0.400000

Evidence remains a separate axis

Evidence matrix · user assertions never manufacture verified blocker evidence
CandidateLabel / maturitySource / configuration / methodOpen hardware question
Aillustrative / allocationA-SELECTED-PLAN-v1; GW-RX-NOM-QPSK-2450; Supplied allocation snapshot; hardware tests openZero-IF calibration and wanted-plus-blocker validity.
Billustrative / allocationB-SELECTED-PLAN-v1; GW-RX-NOM-QPSK-2450; Supplied allocation snapshot; hardware tests openReal filter rejection, gain-state performance and selected IF budget.
Cillustrative / allocationC-SELECTED-PLAN-v1; GW-RX-NOM-QPSK-2450; Supplied allocation snapshot; hardware tests openRF input clock/driver performance and missing sensitivity margin.

Weight reversals and ties

Eligible score versus cost weightIllustrative allocation scores from zero to one. Other weights retain their prior proportions. Ineligible candidates have no trace. Exact crossings and scores are in the following table.000.250.250.50.50.750.7511Value-weighted score [0…1] · modeled allocationcost weight [0…1]; dimensionlessA: solidB: dashed
Fixed value anchors; unchanged hard screens. Preference dominance concerns these values, not superior RF physics. The 0…1 grid uses 0.01 steps, exact analytical crossings and ±10⁻⁷ adjacent probes. This is a decision-support view.
Key weights for cost; all pairwise eligible crossings, not automatic recommendations
Selected weightEligible score valuesLeading allocation / tie
0.0000000000A: 0.666666667; B: 0.333333333; C: —A
0.3000000000A: 0.606666667; B: 0.493333333; C: —A
0.4545453545A: 0.575757596; B: 0.575757522; C: —A
0.4545454545A: 0.575757576; B: 0.575757576; C: —Tie: A / B
0.4545455545A: 0.575757556; B: 0.575757629; C: —B
0.6000000000A: 0.546666667; B: 0.653333333; C: —B
1.0000000000A: 0.466666667; B: 0.866666667; C: —B

Exact eligible pair crossings in this basis: 0.4545454545. Between crossings each pair’s linear score order is unchanged. Equal vectors remain tied throughout. Read the leading set, since a pair crossing may be below another candidate.

Uncertainty is off for the committed allocation. Enable bounded scenarios to explore intervals; the static worked envelopes in Section 7 remain available.

Published stable variable-to-base map
p05-m07-halton-map-v1 · reservations survive reordering, deletion and re-addition
IDCost / scheduleSensitivity / RX powerFootprint / service
A2 / 331 / 3741 / 43
B5 / 747 / 5359 / 61
C11 / 1367 / 7173 / 79
D17 / 1983 / 8997 / 101
E23 / 29103 / 107109 / 113

Interface diagnostics and failure lineage

p05-m07-interface-rules-v1 · typed snapshot crosswalk; source IDs retained
Rule / dependent candidateState / reconciliationSource lineage
IF-01 / Bpass: 80 MS/s at A0 → filter and decimate by 1000 → 80 ksample/s at D3. The factor and filtering are supplied, not a relabeling.05.6-A0 → 02-D3 → IF-01-TRANSFORM-v1
IF-02 / Bpass: Separate repair example: +10 dBm actual PA output − 2 dB loss = +8 dBm actual R2. 05.5 instead requests +10 dBm R2, requiring +12 dBm before compression.TX-PA-EXAMPLE → TX-R2-EXAMPLE → IF-02-TRANSFORM-v1
IF-03 / Bpass: Restore 20 kHz detector ENBW; retain 13.5 kHz wanted support as a different row.05.3-ENBW → REVIEW-ENBW → IF-03-TRANSFORM-v1
IF-04 / Bpass: RX-IMPL=2 dB is already included. Do not add RX-IMPL again in reserve.05.3-IMPL → REVIEW-IMPL → IF-04-TRANSFORM-v1
IF-05 / Bpass: Use 2450 MHz analog input, not its 30 MHz alias. Replace components (.8 ps and .6 ps) with 1 ps total, or RSS components alone.RF-ANALOG-INPUT → JITTER-EVALUATION → IF-05-TRANSFORM-v1
IF-06 / Bpass: Use supplied SINAD under its sine-test conditions. Do not add its included ADC-Q or ADC-NOISE again; no narrowband truth is inferred.ADC-SINAD → REVIEW-ADC-EXTRA → IF-06-TRANSFORM-v1
IF-07 / Bpass: Carry the always-on gateway RX boundary. Keep sleeping-node energy and shared-system power separate.GW-RX-BOUNDARY → REVIEW-GW-POWER → IF-07-TRANSFORM-v1
IF-08 / Bpass: Recompute against the reviewed parent version and record that source. Age alone is not an RF failure.REVIEW-PARENT → DEPENDENT-RECOMPUTATION → IF-08-TRANSFORM-v1

Local architecture decision record

Architecture review chair

Engine assessment: Conditional allocation leader: B. Hardware evidence remains open; this is not product compliance.

Chair rationale (user assertion): Carry B into the allocation review at 0.60 cost / 0.40 schedule, conditional on unclosed RF evidence.

Dissent / evidence that changes the decision: Program lead prefers A below cost weight 5/11. Blocker failures or a reversed preference basis would reopen the choice.

Change / waiver log: No waiver. A proposed requirement change needs rationale, owner and a new comparison baseline.

Selected evidence request: EXP-01 · Co-located-blocker comparison · Receiver lead.

The record is local and unsaved. Review the assertion against the current hard rows, crossings, evidence and risk register before adopting it. No automatic business or legal decision.

p05-m07-constraint-trade-v1 · p05-m07-trade-fixture-v1 · p05-m07-interface-rules-v1 · p05-m07-decision-json-v1. Scores display three decimals; independent fixtures use absolute 10⁻⁹. Plane/mode and source metadata are retained in copy and print.

Keyboard weight key points

View only · point 1/104: cost weight 0.0000000000; leader A.

Ungraded capstone · System architecture review

Chair a review that leaves a usable engineering record.

Assume Modules 05.1–05.6 and Paths 02–04. Submit the ten linked artifacts below. You may choose different traceable values and weights, or defer the choice. Explain why the selected architecture clears the declared hard screen, and what evidence would reverse your decision. There is no score, saved progress, certificate or grading engine.

Review package · minimum reproducible evidence contract
ArtifactRequired content / quality test
1 · Requirements and assumptions baselineScenario, hard limits, owners, configuration, date/version and named unknowns; node TX / gateway RX / shared-system boundaries separate.
2 · Frequency planSelected channel, RF/IF/LO/sample rates, images/aliases and real-filter evidence gaps.
3 · Receive budgetsGain/noise, ENBW, blocker/linearity/AGC states and converter headroom; preserve plane and mode on each quantity.
4 · TX and digital crosswalkRequested versus actual TX power, loss, duty, A0 versus D3 rates, clock/driver inclusion and shared interfaces.
5 · Hard screenEvery pass/fail/unknown with signed margin or reason and source; no score for excluded/deferred candidates.
6 · Preferences and sensitivityNonoverlapping criteria, fixed value anchors, owned weights, Pareto set, exact reversals, bounds and sequence assumptions.
7 · Reconciled interfacesAll eight rule results with triggering and repaired source IDs; recompute dependent versions.
8 · Decision recordSelect conditionally / defer / reject, architecture, basis, binding rows, reasonable reversals, conditions and authority.
9 · Dissent and assumption logWho disagrees, why, and the specific evidence or weight change that would change their position; proposed waivers do not silently pass.
10 · Risk register and one evidence requestRisk ID, consequence, uncertain assumption, missing evidence, reversible/irreversible choice, owner, experiment, outcomes, stop condition, effort and schedule.
Approved rubric · explanatory guidance only, no grading or completion tracking
DimensionEmphasisWhat good reasoning looks like
Traceability20%Every decision points to requirement, source, condition and owner.
Reconciled budgets30%Units, planes, bandwidths, modes and source versions agree; impairments counted once.
Conditional / worst-case reasoning20%Unknowns remain explicit; interval failure is not averaged away.
Trade and risk quality20%Disclosed values, reasonable reversals, dominance and one discriminating experiment.
Review communication10%A concise conditional decision, dissent and evidence request that another engineer can act on.
Go deeperModel review record: conditionally carry B forward

Decision: carry discrete IF100 B into the allocation review at weights 0.60/0.40. R2 sensitivity margin is +3.165007875 dB at 20 kHz ENBW and the 05.3 criterion; gateway RX margin is +40 mW and cost margin +13 exercise units. All supplied interface snapshots clear. This does not validate hardware.

Alternatives: C is rejected from this study because −110 dBm fails −115 dBm. A remains nondominated and leads when cost weight is below 5/11; equality is a tie. Optional D is eligible but both A and B dominate its cost/schedule values. The overlapping bounded envelopes permit additional reversals. B’s [−119,−114] dBm variant is not robustly feasible.

Defer condition: if verified blocker performance becomes a hard requirement, every candidate lacks an eligible path to selection. A/B are unknown and C still has a known fail. Record defer all rather than selecting a “least unknown” winner.

Dissent: the program lead prefers A when schedule receives over 6/11 of the weight. They would accept B after a shared review of swing values and a valid blocker comparison; a B-only blocker failure would move the chair toward A.

Evidence request: EXP-01, Receiver lead, two illustrative engineer-days within one week. Agree wanted criterion and threat conditions before testing. Stop selection if both fail. Defer irreversible hardware/tooling commitment until the relevant hard evidence closes. EXP-02 tests whether C can re-enter after a new allocation; EXP-03 then narrows preference uncertainty.

Common misconceptionAn architecture review is a presentation, and selection ends the need for dissent.

The useful output is a reproducible decision with conditions, source lineage and an evidence plan. A polished deck without those artifacts cannot guide the next engineering change. Keep dissent and reopening conditions attached after the meeting.

This lesson is Path 05’s synthesis. Availability of seven lesson routes does not certify completion of the entire path. Antenna/channel truth belongs to Path 06, named technology confirmation to Path 07, verification procedures to Path 08, market requirements to Path 09 and implementation lifecycle to Path 10.

Ungraded review

Check your understanding

Answer each question in your own words, then reveal the model answer.

  1. 01Why is C absent from the real rank and Pareto set?
    Model answer

    Its R2 sensitivity is −110 dBm against a closed maximum of −115 dBm: a −5 dB margin. Fail precedes preference scoring. Its 0.933333333 naive score exists only in the frozen counterexample.

  2. 02When do A and B reverse, and what happens at equality?
    Model answer

    At cost weight 5/11 the scores tie at 19/33. Below this boundary A leads, above it B leads, outside the declared 1e−9 score tie tolerance. At 0.30/0.70 the scores are A 0.606666667 and B 0.493333333.

  3. 03Why is D dominated while A and B can reverse?
    Model answer

    D clears the supplied hard screens but has worse cost and schedule values than both A and B. A trades better schedule for worse cost relative to B, so neither dominates the other. Ineligible C cannot dominate eligible candidates.

  4. 04When should the review defer all selection?
    Model answer

    When no candidate clears every declared hard requirement. If verified blocker evidence is required, A/B are unknown, C retains a sensitivity failure, and no evidence-label dropdown can manufacture the missing record.

  5. 05Does a 700/1024 leader fraction mean a 68.4% chance of success?
    Model answer

    No. It means leadership in that fraction of the explored deterministic states under the declared bounds and dependence construction. The sequence does not supply a calibrated probability model. Analytical corner bounds are still required.

  6. 06Which missing evidence comes first and what could it change?
    Model answer

    EXP-01 first addresses unclosed blocker performance for the eligible A/B pair. A B-only failure could favor A; both failing means stop or rebaseline. Clock/driver characterization addresses C’s feasibility, and BOM refinement narrows a preference interval after the hard questions.

Sources and further study

Methods, provenance and limits

Access checked 7 September 2026. Candidate values, costs, time boxes and evidence gaps are synthetic, supplied teaching allocations. The equations, rule fixtures and plots here are original derivations of the disclosed model; none is measured product data.

  1. NASA Systems Engineering Handbook, crosscutting technical management, SP-2016-6105 Rev. 2, §6.8.1 and its decision-analysis activities/report contents, as presented on NASA’s public handbook page. Used for the sequence from decision context to alternatives, uncertainty, recommendation and decision record. This fictional product is not bound by NASA process requirements; linked expanded guidance was not consulted.
  2. INCOSE Systems Engineering Handbook, 5th edition, 2023. Official edition listing and INCOSE’s April 2026 fifth-edition feedback notice consulted to verify status. The full member-only handbook was not accessed; no paywalled clause/page claim is made. Recommended for lifecycle and review context.
  3. Ralph L. Keeney and Howard Raiffa, Decisions with Multiple Objectives: Preferences and Value Trade-Offs, Cambridge University Press, 1993 reissue of the 1976 work; Chapter 3, Tradeoffs Under Certainty, and Chapters 5–6 as further study. Publisher contents/edition and public preview were consulted, not the full text. Additive preference separability is an explicit modeling assumption here; probabilistic independence is not inferred from it.
  4. J. H. Halton and G. B. Smith, Algorithm 247: Radical-inverse quasi-random point sequence, Communications of the ACM 7(12), 701–702, 1964. Original algorithm provenance; ACM full text was unavailable. Halton’s A Retrospective and Prospective Survey of the Monte Carlo Method, SIAM Review 12(1), 1970, public author-paper scan, and John Burkardt’s author-maintained Halton description, revised 12 August 2016, provide accessible sequence context. The local digit-reversal implementation starts at index 1 and is independently checked; no library code is copied.
  5. William F. Egan, Practical RF System Design, first edition, Wiley–IEEE, 2003, ISBN 9780471200239. Publisher edition/contents and public Chapter 6 abstract consulted for RF architecture/allocation context; full text was unavailable. The actual receive anchor comes from the independently checked 05.3 model, not an invented textbook table.

Model p05-m07-constraint-trade-v1; fixture p05-m07-trade-fixture-v1; interface rules p05-m07-interface-rules-v1; sequence/map p05-m07-halton-map-v1; serialization p05-m07-decision-json-v1; display p05-m07-display-v1. Score/value comparisons use absolute 10⁻⁹ on unrounded values. Sequence coverage, human preferences and synthetic allocations never establish hardware, probabilistic, business or legal truth.