Test portfolio
Which small tests reduce decision uncertainty most usefully per unit of effort?
Fund qualified electrical assurance first, then buy the best combination of decision tests that fits the remaining cap. The £1,200 electrical review is mandatory: a promising score cannot replace it. Compare whole portfolios, because the highest individual ratio does not always produce the most useful affordable set.
Fictional what-if workspace. Changes stay in this tab and reset on reload; no record is saved and no external system is changed.
1. Set a test budget
The launch envelope is £240,000 over 12 weeks. This is a separate illustrative test allowance; the fixtures do not say whether test costs are already in that envelope. £3,000 is an authored starting scenario, not approved funding.
Whole pounds, £0–£240,000. Applying a budget chooses the best affordable portfolio and clears simulated review evidence. All ten tests cost £6,200.
Best feasible portfolio
Next useful budget step
2. Keep the electrical pilot gated
Supplied mandatory gate T02: Qualified electrical assurance must be funded, completed and passed before the electrical pilot can proceed.
- 1 · Funding
- 2 · Completion
- 3 · Passed
Selection only reserves hypothetical funds. No review has occurred. The following controls simulate evidence for the gate; they cannot approve, certify or release a real pilot.
No simulated result applied.
3. Compare ten assumption → test → decision chains
Check an optional test to build your own portfolio. T02 is reserved whenever affordable. An addition that exceeds the accepted budget is rejected without changing the selected set. Below £1,200, this recommendation holds all spend for the mandatory gate.
| Select | Assumption → test → decision | Cost / severity | Adjusted value proxy | Value per £ of cost |
|---|
“Value proxy” means the supplied base estimated decision value, adjusted by your stated scenario weight. It is not money earned, probability of success, measured uncertainty reduction or assured risk reduction.
4. Make one test decision-ready
Use this inspector to challenge an estimate and inspect the evidence a small test should produce. Test designs and decision rules below are authored proposals; no sample sizes, decision thresholds or results were supplied.
0–200%, whole numbers. 100% retains the fixture estimate. Lower a test whose result is unlikely to change a decision. This sensitivity assumption is not a calibrated probability.
Small test & evidence receipt
Decision rule to agree before testing
Limitation / prerequisite
How to use the recommendation
Run T02 and the selected prerequisite checks before the affected decisions. If the electrical review fails or remains inconclusive, address its findings and retest; keep that pilot blocked. Use demand, cover, supplier and pricing tests to change a specific operating choice. Run the first-job quality review only when those jobs can occur safely. Reassess remaining estimates after each result.
Adjusted valueᵢ = supplied base estimated decision valueᵢ × scenario weightᵢ / 100 Individual efficiencyᵢ = adjusted valueᵢ / supplied test costᵢ Portfolio proxy = sum of selected adjusted values Choose the subset with the greatest proxy, subject to: sum of selected costs ≤ accepted budget T02 included whenever budget ≥ £1,200 Ties: lower spend first, then ascending test-ID order. If budget < £1,200: select nothing; funding gap = £1,200 − budget. Electrical gate = T02 selected AND simulated completion AND simulated passed finding.
All 1,024 subsets of ten tests are evaluated, with T02 required for feasible portfolios. The displayed ratios aid judgement; the optimiser uses total proxy. Severity is visible as context and does not multiply value. A legal or safety necessity cannot be traded away for score.
What this calculation does and does not establish
- Money is the only supplied effort measure. Staff hours, calendar duration, access, sample sizes, reliability and test dependencies are unknown. A cost-efficient set may be infeasible in the 12-week schedule.
- Estimated values are not formal expected value of information. There are no prior probabilities, outcome likelihoods or explicit losses avoided. Values may overlap; adding them assumes independent, non-overlapping decision benefit. Use the weights to challenge this, then obtain better estimates.
- The mandatory gate is necessary, not sufficient. Only T02 is explicitly mandatory in these fixtures. Passing it does not prove readiness of the whole service or satisfy any unlisted approval, safety or quality requirements.
- Test design remains to be agreed. Scope, qualified reviewer, acceptance criteria, representative samples and named decision owner are missing. The inspector offers proposals, not acceptance authority.
- Spending the cap is not the objective. Zero-weight optional tests may be omitted even with spare funds. Some optional tests can run in parallel; no schedule feasibility is asserted.
Evidence, provenance & acceptance checks
Inputs: “Repair & Reuse: two-hub launch”, benchmarkVersion 1.0.0; candidateTests T01–T10 and mandatoryTestGates.T02. All records and economics are fictional. The ten supplied chains, costs, base estimated values and severity scores are preserved. Default £3,000 allowance, scenario weights and proposed small-test methods are authored assumptions.
- Ten assumption–test–decision chains are displayed in section 3.
- Every accepted selection is within the accepted budget; invalid entries preserve the last valid state.
- An unfunded T02 blocks all recommended test spend and quantifies its funding gap. Funding alone leaves the electrical pilot blocked until completion and a passed finding are simulated.
- Inputs can be corrected and results recalculated in this page. Reset restores £3,000, original estimates and no review evidence. Reload has the same effect; no durable receipt or history is created because this workspace is intentionally ephemeral.
Development: gpt-6-astra, ultra reasoning, 25 October 2026, delegated worker within a reviewed run. Plain HTML/CSS/JavaScript; local file authoring and bounded independent arithmetic/interaction checks. No external model calls, supplier research or real test execution. Browser QA belongs to the parent run; its results are recorded separately. No performance improvement is claimed.