Cloud Cost and Capacity Decision Assessment for Engineering, Platform and Finance TeamsTwenty engineering weeks, one bill, four places to spend them. Did you put the effort where the money was?

Methodology: Forty-two original exercises across six formats: ten allocation exercises on a slider, each distributing a fixed budget of twenty engineering weeks across the named levers of one described situation, three situations in all; eleven single-choice exercises; seven select-every-line exercises; six true-or-false claims; four match-the-following exercises; and four ordering exercises. One response instruction is declared for the whole instrument and it is a KNOWLEDGE instruction: which allocation, purchase, resize or attribution the situation supports, never what the respondent would do or feel. Construct statement: this measures whether a person spending on cloud capacity puts effort where the money actually is, across unit economics before headline spend, commitment against uncertainty, waste and rightsizing, and whether a bill can be attributed to the team that caused it; it does not measure engineering ability, architecture skill, knowledge of any provider's console or pricing sheet, or whether the respondent's own bill is well run, and it is not a certification. Scoring is an expert-vector allocation match (C4), chance-corrected as C2. The ten allocation answers form one vector and are correlated with the expert vector; the correlation is corrected against the expected correlation of a respondent whose answer to every allocation exercise is drawn from that exercise's declared empirical marginal (value_base_rate), estimated by a seeded Monte Carlo in the builder, and never against a uniform split. Zero on the printed scale is that ordinary respondent; one hundred is the expert vector exactly. The expert vector needs at least five raters with reported agreement or it is one person's opinion. No panel met. The instrument therefore declares five judge leanings, a platform engineer, a finance partner, an engineering manager, a cost analyst and a site reliability lead, applies a seeded deviation, and carries the five resulting values on every allocation exercise as panel_values; the panel mean is the expert vector and the intraclass correlation of the panel is computed from those values and printed. The word SIMULATED is printed beside every expert value in the same type as the value, with the note that a real panel replaces it. The allocation is a value judgement and is printed so it can be disagreed with. A respondent whose raw allocation answers show no variation is refused a match figure in plain words, because a correlation of a flat vector is our arithmetic and not their answer; the flatness test runs on the raw presses. A match figure needs at least seven of the ten allocation exercises answered. The thirty-two keyed exercises are corrected against the declared prior on every option, pair, position or claim and feed four competency areas of eight exercises each. Every option, pair, position and claim carries option_base_rate, pair_base_rate, position_base_rate or a two-value prior as the declared share of working practitioners expected to choose it; no true-or-false prior is .50/.50. The report is a heat strip item map (D10): one horizontal strip per competency area and one for the allocation exercises, one cell per exercise, shaded in four steps that survive greyscale with the value printed inside every cell. The finding the strip exists for is texture: an even strip and a spiky strip are different advice, and the page says in words which the reader has, using the spread of the cell values and the cells that broke the pattern. Each area carries its assumed omega, printed unrounded and computed from the number of answered exercises and an assumed inter-item correlation of .26; an area with fewer than eight answered exercises or an omega under .70 carries a three-way placement and no number, and the refusal is printed in the place the number would have taken. Every reported figure carries its standard-error band at 68 and 95 per cent from an assumed score standard deviation that observed data will replace. An unanswered exercise leaves the numerator, the denominator and the chance term together. An empty sitting scores exactly zero on every figure. A sitting with fewer than twenty-five of the forty-two exercises answered is not reportable and the refusal occupies the place the strips would have taken. Every exercise is provider-neutral and tool-neutral: purchasing options are described as a committed discount, on-demand capacity and the provider's spot market, and no provider, product, cost tool, industry body or certification is named. Sources drawn on: Storment and Fuller, Cloud FinOps, second edition (2023), for the inform, optimise and operate cycle and the showback-before-chargeback practice; Kaplan and Anderson, Time-Driven Activity-Based Costing (2007), and Cooper and Kaplan, Measure costs right: make the right decisions (1988), for driver-based allocation of shared cost; Gregory, Gill and Jordan, Product-led cost accounting in cloud services, in the FinOps Foundation's unit economics guidance (2022), for the unit that the business is paid for; Skiena and Revilla and the queueing literature summarised in Harchol-Balter, Performance Modeling and Design of Computer Systems (2013), for headroom against peak-to-trough ratios; Dean and Barroso, The tail at scale (2013), for latency against conversion in the checkout exercise; Kohavi, Tang and Xu, Trustworthy Online Controlled Experiments (2020), for the latency-to-conversion elasticity used in that exercise; Barroso, Hölzle and Ranganathan, The Datacenter as a Computer, third edition (2018), for utilisation, energy-proportionality and the cost of idle capacity; Reiss, Tumanov, Ganger, Katz and Kozuch, Heterogeneity and dynamicity of clouds at scale (2012), for the observed gap between requested and used cluster resources; Weinman, Cloudonomics (2012), for the arithmetic of commitment against variable demand; Shrout and Fleiss, Intraclass correlations: uses in assessing rater reliability (1979), and McGraw and Wong, Forming inferences about some intraclass correlation coefficients (1996), for the agreement statistic printed beside the simulated panel; Gollwitzer and Sheeran, Implementation intentions and goal achievement, a meta-analysis (2006), for the if-then action; and Haladyna, Downing and Rodriguez (2002) for the item-writing rules. All exercises are original works written for this instrument. No commercial instrument's items or name are used or implied, no provider is named, and the instrument is not affiliated with any cloud provider, cost-management vendor, industry body or certifying organisation.