Find your next learning step
A bootstrap consensus envelope identifies uncertain mechanism boundaries, but the next experiment should be chosen for the boundary it can resolve. A discriminating design predicts how candidate measurements would alter the posterior or bootstrap support for competing maps. Expected map resolution values an experiment through a scientific or release decision, while protocol selection accounts for cost, delay, specimen use and model uncertainty. The aim is targeted evidence, not the most data from the cheapest instrument.
Open the chapter contents · Polymer Science root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the decision, or read from Chapter 1. The chapters remain open; these grouped routes are a navigable index.
Route 1 · Define the decision
Route 2 · Work the ledgers
Route 3 · Transfer the method
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE DECISION
1. Name the competing map decisions
Begin with a boundary that matters: whether two ageing pathways form one mechanism class, whether a recycled blend belongs in an approved family or whether a rare failure mode requires a separate group.
List the plausible maps and the downstream action each would support. An experiment is discriminating only if its outcomes differ meaningfully across those maps.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Predict outcomes under every candidate map
For each protocol, specify the outcome distribution under each competing mechanism map, including calibration error, missingness and batch effects. These predictive distributions drive expected resolution.
Avoid using the same data to invent a candidate map and claim decisive separation without adjustment. Hold out validation evidence or propagate selection uncertainty.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. A protocol-selection ledger
A boundary pair can be studied with spectroscopy, thermal cycling, microscopy or mechanical recovery. Scores are evaluated before the new experiment is run.
| Protocol | Expected decision resolution | Cost | Delay | Specimen impact |
|---|---|---|---|---|
| Spectroscopy scan | 0.18 | 1.0 | 1 day | Low |
| Thermal cycle | 0.42 | 2.4 | 4 days | Moderate |
| Interface microscopy | 0.51 | 3.1 | 2 days | Destructive |
| Recovery sequence | 0.37 | 1.8 | 6 days | Low |
Resolution is not a universal accuracy score. It is the expected reduction in uncertainty for the declared map decision under the current model.
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Worked expected map resolution
Suppose thermal cycling yields one of three outcome bands. For each band, update the probability that specimens A and C share a mechanism, then measure the reduction in decision entropy.
| Outcome band | Predictive probability | Updated shared-map probability | Decision-entropy reduction |
|---|---|---|---|
| Low divergence | 0.30 | 0.84 | 0.28 |
| Intermediate | 0.45 | 0.58 | 0.05 |
| High divergence | 0.25 | 0.17 | 0.43 |
| Expected | 1.00 | 0.21 |
An intermediate result teaches little because it leaves the boundary unresolved. Experimental value comes from the full outcome distribution, not the most favourable result.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Target the decision, not every parameter
A protocol may estimate kinetic parameters precisely yet add little evidence about the contested grouping. Decision-focused active learning values directions that can change the mechanism action.
Keep a secondary scientific-value record for discoveries outside the immediate decision. A narrow decision objective should guide the commissioned experiment without pretending other knowledge has no value.
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Protect against model-sensitive rankings
Experiment rankings can change with priors, observation models and proposed mechanism classes. Recalculate under credible alternatives and report protocols that remain near-optimal across the sensitivity set.
Robust expected information gain offers one way to account for prior ambiguity. When rankings reverse, a small pilot or diverse two-protocol design may be safer than committing the entire budget.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Changed-context transfer: recycled blend families
Transfer the design to recycled blends where microscopy, spectroscopy and mechanical cycling disagree about interfacial failure. Choose experiments using production-lot variation and actual release consequences.
The transfer check prevents a high-throughput spectroscopy campaign from winning solely on volume when a smaller interface study is more likely to resolve the release boundary.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Connect uncertainty envelopes to active choice
Use the consensus-envelope guide to identify boundary pairs. The protocol-weighting guide defines how new evidence enters the map.
Return to the protocol-influence guide before scaling a design whose leverage could dominate the consensus.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Limits and common misconceptions
Expected information gain is model-based. A confidently misspecified observation model can rank the wrong protocol first. Sensitivity and posterior predictive checks remain essential.
The experiment with the greatest expected resolution is not automatically the best choice. Safety, destructiveness, throughput, latency and the value of preserving scarce specimens can override a small information advantage.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Practical evidence checklist
Archive competing maps, decision loss, candidate protocols, predictive outcome models, priors, batch structure, acquisition calculation, simulation error, cost, delay and specimen impact.
Pre-register the selection rule and update path. After the experiment, compare observed and predicted outcomes, rebuild the consensus envelope and document whether the downstream decision actually changed.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Should we always test the most uncertain pair?
No. Prioritise pairs whose resolution can change a scientific or release decision and for which a feasible protocol is genuinely discriminating.
Can two modest experiments beat one strong one?
Yes, especially when they have complementary failure modes, but value their joint outcomes and dependence rather than adding scores blindly.
What if every protocol has low expected resolution?
The current experiment set cannot resolve the boundary reliably. Redesign the protocol, accept multiple maps or revise the decision to remain safe under ambiguity.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Sources and next reading
Javdani and colleagues study Bayesian active learning for decision regions. Filstroff and colleagues develop targeted active learning for Bayesian decision-making. Go and Isaac propose robust expected information gain under prior ambiguity. The polymer protocol ledgers are original teaching examples.
Continue through the Polymer Science root, the Gold Standard of Education and How X Works.
Top · Contents · Previous · How X Works