#Characterising Generative Novelty and Value

C.17 · Cluster C.IV - Composite & Macro-Scale · Stable · Part C - Kernel Extension Specifications

#About this pattern

This is a generated FPF pattern page projected from the published FPF source. It is canonical FPF content for this ID; it is not a FPF Reference product feature page.

#How to use this pattern

Read the ID, status, type, and normativity first. Use the content for exact wording, the relations for adjacent concepts, and citations to keep active work grounded without pasting the whole specification.

Status. Evaluation and measurement-use pattern; normative where stated.

Depends on. A.17, A.18, and A.19 for Characteristics, Scales, and CharacteristicSpaces; A.19.ECS for the evaluation-space specification; C.16 for measurement; C.2.1 for claim-bearing results and model epistemes; A.1.1 for model applicability, use, and expression coherence when those relations matter; A.10 and B.3 for evidence, reliance, and assurance; and the patterns that define the current objective, acceptance criterion, and must-constraints.

Coordinates with. E.10.LRN when learning progress or related wording still hides the bearer or result; C.18 for generation, Archive, Front, and possibility-space change; C.19 for pool policy and tie-break use; G.5 for selector-facing declarations; C.11.CRC for a missing finite configuration-relative comparison; C.11 for choice; F.9 for an actual cross-reference-scheme Bridge; F.18 for naming-candidate diversity; B.4 and G.11 for evolution and refresh; A.13 and A.15.1 for exact evaluator recovery and independently admitted dated overall-assessment Work; A.2.1 and F.6 only when exact assignment-bound attribution is expressly consumed; A.3.1 for the enacted Method; A.3.2 when a relied-on MethodDescription matters; and A.6.1 only when the assessment also uses one exact operation declared by a separately admitted U.Mechanism and a claim needs that operation's application or bindings.

Use C.17 when someone must say whether a design, code change, theory, policy proposal, dated Work occurrence, or finite candidate set—the bearer being discussed—is new relative to a named comparison basis and useful for a stated objective or must-criterion.

#Keywords

• qualitative-first evaluation
• named comparison basis
• Novelty
• Use-Value
• ConstraintFit
• bounded quantitative result
• evidence
• uncertainty
• incomparability.

#Relations

#Content

#Use this when

Use C.17 when someone must say whether a design, code change, theory, policy proposal, dated Work occurrence, or finite candidate set—the bearer being discussed—is new relative to a named comparison basis and useful for a stated objective or must-criterion.

If the claim arrives as learning progress, learned novelty, or information gain described as learning, and the bearer or result is still hidden, apply E.10.LRN and the direct result owner first. Enter C.17 only after the bearer and the novelty, use, surprise, creativity, or other characterization question are exact. Stop at the direct result when no such characterization is current.

Begin with the smallest useful answer:

1. identify the bearer being discussed;
2. say what it is new compared with;
3. say which objective, acceptance criterion, or must-constraint matters;
4. state the supported difference, its practical consequence, the evidence used, and the limit of that support.

Stop there when a qualitative answer is enough. A discussion does not need a score, profile, reusable record, or dated assessment merely because it uses the words novel, useful, or creative.

Open the stronger branch only when the receiving use needs a quantified coordinate, a comparison-ready result, later reliance, or an audit trail. Then every coordinate must follow its truthful measurement or ascription route before the coordinate claims can support a bounded aggregate result.

What changes in practice. A team can replace an unqualified creativity label with the comparison basis, practical criterion, supported difference, consequence, uncertainty, and stopping point. When numbers matter, it can also reproduce the complete result chain.

Not this pattern when. Use C.18 to generate candidates or maintain an Archive or Front, C.19 to change pool treatment, G.5 to declare selector-facing set results, C.11 to make a choice, and A.13 to characterize agency or autonomy. Use C.17 to report characteristics and bounded conclusions, not to generate, retain, rank, approve, fund, enact, make a service promise, classify a person as creative, measure a person's creative capacity, organize a team, or prescribe a workflow.

Do not infer a person's or System's creative capacity from a C.17 result. Strong agency can still yield a weak result, while useful scaffolding can help produce a strong result; C.17 characterizes the bearer and evidence named in the current claim.

#The practitioner route

#Qualitative first move

Name the bearer, comparison set, practical objective or must-criterion, observed or argued difference, supported consequence, evidence, and limitation. A useful statement can be as simple as:

Compared with the admitted five-year pump-design set, P-22's inspected split-clamp arrangement is a supported difference. The cited assembly evidence supports shorter assembly, and the design must not require new tooling. The inspection does not support claims about other pump families.

This is already a valid C.17 result for an ordinary design discussion. It does not imply a numerical Novelty value or an overall assessment occurrence.

#Quantified or reusable branch

When a quantified, comparison-ready, or reusable result is needed:

1. select one A.19 CharacteristicSpace and its A.19.ECS specification;
2. fix the finite comparison corpus, inclusion rule, source editions, scope, comparison window, and evidence;
3. identify the objective, acceptance criterion, and must-constraints actually used;
4. identify the similarity or measurement Method used and any model, encoder, distance definition, invariances, calibration, and uncertainty basis it needs;
5. for each coordinate, cite an existing complete C.16 measurement result, perform and constitute the missing C.16 measurement, or state a C.2.1 non-measurement ascription under its declared rule;
6. form only the aggregate conclusion needed by the receiving comparison;
7. add an optional profile payload, representation, or record only when a named receiver needs it.

Return only the missing premise that blocks the current coordinate or conclusion. A missing measurement does not invalidate independent qualitative claims or other coordinates.

#Dated overall-assessment branch

Open this branch only when the claim says that an overall assessment actually occurred and later reliance needs that fact. Recover the evaluator System through A.13 and the admitted Method through A.3.1, then use A.15.1 to independently admit the dated Work that enacts that Method. Add the exact A.2.1 assignment occurrence and F.6 relation only when later reliance expressly consumes precise assignment-bound attribution through the same obtaining A.13 assignment; missing or failed F.6 leaves the assessment Work intact. Only the evaluator System performs the Work. A separate MethodDescription may explain the reusable Method; it is not enacted. Coordinate-measurement Work remains its own C.16 Work, and the aggregate result states claims.

Do not infer an A.6.1 operation application from the Method, Work, configuration, or result. If a receiving claim separately asserts an exact operation application or binding, satisfy the current A.6.1 application account and cite that application. Otherwise retain only the C.17-local evaluator System, assignment, Method enactment, dated assessment Work, coordinate results, and aggregate result that the claim actually needs.

#Keep the evaluation objects distinct

Changing a space slot, corpus membership or inclusion rule, model claims or training basis, objective, criterion, constraint, Scale meaning, scope, or window reopens only the coordinate claims that depend on that change and any aggregate conclusion that uses them.

#Retired predecessor heads

The following names no longer introduce root kinds:

Earlier results remain historical epistemes under their original editions. Relate an earlier and later result as editions only when both results and the continuity claim are recoverable; do not retype an old result merely because the current ontology is clearer.

#Evaluation configuration

The configuration must make these questions answerable:

• Which bearer is evaluated, for which use and ClaimScope, over which selected slices and window?
• Which CharacteristicSpace and A.19.ECS specification supply the Characteristics, Scales, value meanings, missingness rules, and admissible comparisons?
• Which finite corpus or reference set supplies the comparison basis, and how were its members admitted?
• Which Method, model, encoder, distance definition, calibration, and uncertainty basis produced or supports each value?
• If the Method compares a representation or observation instead of the bearer directly, which bearer and representation or observation are related, what describing, projection, measurement, or other stated relation supports the inference, and do the corpus members use a compatible comparison basis? If not, state the mapping and relevant loss.
• Which objective, acceptance criterion, and must-constraints are current, and which source epistemes state them? Their inclusion in the configuration does not make their claims true.
• If the result claims improvement or gain, which baseline and comparison or counterfactual Method make the difference testable?
• Which evidence supports the difference, consequence, coordinate, or comparison conclusion, and what use may rely on it?

For each objective, criterion, or constraint on which the result depends, keep its EntityOfConcern, effective ReferenceScheme, edition or currentness basis, and subject-defined predicate recoverable. Do not use a generic container label to answer several of these questions at once. Add only the source, scheme, scope, model-use, comparison, or evidence relation the current claim needs.

Keep prospective and observed readings distinct. A prospective reading may use a surrogate model and stated assumptions to compare designs before use; an observed reading uses later Work, service, or other outcome evidence. Do not overwrite the prediction as though it had been an observation. Preserve both results when the comparison matters, and reopen only the coordinates and conclusions that relied on the superseded prediction.

#Core characteristics

Each selected characteristic has a declared Scale, polarity, admissible operations, missingness rule, and evidence route. An ordinal value is not averaged unless a declared model justifies an interval interpretation.

#Novelty: unlike which admitted set?

Novelty describes supported difference from a finite comparison corpus under one declared similarity Method. When that Method returns a calibrated similarity result on [0,1] for each like-for-like comparison, that source result keeps its declared Similarity Scale. The following transformation defines a bounded value on a corresponding declared Novelty Scale, whose meaning is difference from the corpus and whose positive polarity increases as maximum similarity falls:

Novelty = 1 - max similarity(bearer, corpus member)

A declared normalization to [0,1] may be part of the Method; state its source Scale and transformation. If the Method uses another similarity or distance Scale, state the lawful coordinate construction and resulting Scale instead of reusing this formula. Subtracting an unrestricted result from one does not make it bounded.

When the Method compares representations or observations rather than the evaluated bearers themselves, name each bearer and the value actually compared, together with the describing, projection, measurement, or other stated relation that lets the comparison support the bearer-level claim. Use a compatible basis for corpus members, or state the mapping and relevant loss. A direct like-for-like comparison of epistemes needs no extra representation relation.

A robust top-k variant is allowed when declared. The result identifies the Novelty Characteristic and Scale editions, corpus and inclusion rule, source editions, comparison window, Method, model or encoder edition, distance definition, invariances, calibration, uncertainty, ClaimScope, evidence, and intended use. Changing any load-bearing element creates a different comparison basis or result edition. Those identifiers make the result reproducible; they do not by themselves show that the value is robust. When the value materially affects a comparison or pool treatment, use diagnostics suited to the claim. For example, inspect the nearest corpus members and their distances, repeat the reading with a plausible alternative corpus or similarity Method and report the sensitivity, and remove a claimed invariance to see whether it materially changes the result. These are bounded diagnostic examples, not one mandatory algorithm. If no robustness check was performed, report the supported value and uncertainty without calling it robust.

Novelty is neither timeless originality nor a property detached from its comparison basis. A label such as Novelty@context is not an executable input and must not substitute for the result chain.

#Use-Value: useful for which objective?

Use-Value, historically also called ValueGain, reports the bearer's supported usefulness or contribution to one declared objective or acceptance criterion. An ordinary usefulness statement may use an ordinal Scale such as Fail | Partial | Pass; it does not need a counterfactual merely because the bearer is useful for the stated purpose.

When the result claims an improvement or gain, identify both the baseline and the comparison or counterfactual Method that makes the difference meaningful. One bounded measured construction is ValueGain = metric_after - metric_before under a fixed metric and window. An A/B comparison, back-test, or causal-inference Method may provide the comparison when it fits the claim. Cite the before and after measurements and the Work or other evidence on which they rely. A predicted gain identifies its model, error, baseline, and intended later update. Without a baseline and an appropriate comparison Method, say what usefulness is supported; do not report an observed gain.

Use-Value may be one member of a declared Q-set, but it is not the whole Q-set by default. If it stays outside Q, name its actual use as a side condition or tie-breaker. Do not silently promote Novelty, Surprise, DeltaDiversity_P, or Illumination into dominance.

#Surprise: unexpected under which model?

Surprise reports how improbable one declared sample of the bearer is under one generative model. For a discrete probability, a common raw result is -log p(sample) in bits or nats. State the modeled sample unit and encoding and how bearer size is handled. Compare bearers only under a justified common extent, a declared per-unit or code-length normalization, or another calibrated rule suited to the model. For a continuous model, identify the measure as well as the representation; a density value alone is not representation-independent. Otherwise keep the raw model result within its exact basis and do not treat it as a comparable Surprise coordinate. Also identify the model episteme and edition, training basis, preprocessing, fit and out-of-distribution checks, calibration, refresh condition, and limits.

Novelty and Surprise answer different questions. A bearer may be unlike the selected corpus yet unsurprising under a broad model, or close to known examples yet surprising under a narrow model. Keep both results visible when both matter.

#ConstraintFit: which must-criteria hold?

ConstraintFit reports satisfaction of the declared must-constraints under their predicates and source epistemes. Use E.5, D.1-D.5, or a service-acceptance pattern only when it supplies the actual predicate or source for the current must-criterion. A ratio such as passed declared must-constraints / all declared must-constraints is allowed when the set and any criticality weights are explicit.

A failing must-constraint makes the bearer ineligible for the affected use unless the receiving constraint or decision pattern recognizes an independently obtaining exception or waiver effect. A waiver speech act alone does not change eligibility. If no pattern defines the needed effect, return the missing relation rather than treating communication as authorization.

#AttributionIntegrity

AttributionIntegrity reports how completely the applicable provenance, authorship, source, and licence duties are met. First identify the duty set and the source that makes each duty applicable. One bounded local construction is satisfied required duties / all applicable required duties on a declared ratio Scale; mark unresolved and inapplicable duties separately rather than treating them as satisfied. Provenance links, licence scans, and acknowledgements are example evidence routes.

When the applicable-duty set is empty, do not compute the ratio. Omit AttributionIntegrity when the receiving use does not need it; when that use needs an explicit disposition, return not applicable under the declared Scale and missingness rule. That disposition is not a pass and cannot by itself pass a filter, break a tie, establish legal adequacy, or satisfy a must-constraint. Keep it distinct from an unresolved duty that does apply.

This reading does not by itself establish legal adequacy. It is measurable but not in the default dominance set; an applicable policy may use it as a filter or tie-breaker. When a duty is a must-constraint, its pass or failure belongs in ConstraintFit and affects eligibility there.

#EffortCost

EffortCost reports actual resource outlay through A.15.1 dated Work, B.1.6 resource aggregation, C.16 measurement, and A.10 evidence. Planned effort remains A.15.2 WorkPlan content. Use cost-normalized readings for planning or comparison only under a declared rule; cost is not itself creativity, and a profile does not carry operational actuals.

#Retained-set and optional applied readings

#Diversity and illumination

Diversity_P describes coverage or dispersion of one declared retained set under a named measurement policy. A local policy may, for example:

• take the average pairwise distance among the admitted members under one declared descriptor map, distance Method, and Scale; or
• report how much of a declared feature partition is covered, using a stated covering radius or k-cover rule.

Neither construction is universal. The result identifies the retained set and membership rule, measurement-policy and Scale editions, descriptor or feature source editions, distance or covering definition, comparison window, and evidence. A distance matrix or coverage map can show the calculation. When the reading affects a decision, vary a plausible kernel, distance definition, covering threshold, or admitted-member set and report whether the conclusion changes.

For a candidate h and retained set S, the same local policy may use the marginal reading DeltaDiversity_P, also written ΔDiversity_P:

DeltaDiversity_P(h | S) = Diversity_P(S plus h) - Diversity_P(S)

Illumination is a report over Diversity_P, such as a coverage map or QD-score summary. It is telemetry, not a primitive characteristic and not part of the default dominance set. Use C.18 to maintain an Archive or Front and C.19 to state any pool policy that uses these readings.

Optional retained-set readings include:

• FamilyCoverage: coverage of locally defined families under a named policy and Scale;
• MinInterFamilyDistance: the smallest distance among declared families, with descriptor map, distance definition, Scale, and family-representation rule;
• AliasRisk: a near-duplicate or alias diagnostic with collision policy, descriptor source edition, and Scale;
• DescriptorVector: an optional descriptor payload whose dimensions and interpretation the same local policy declares.

These readings characterize the named set. They do not admit sources, select members, establish universality, or widen applicability. For naming candidate sets, apply F.18's head-term-family anti-inflation rule rather than restating that lexical rule here.

#Optional applied characteristics

The following are executable local examples, not required universal templates. Use one only when the receiving question needs it and identify its bearer, rule, Scale, and evidence.

• ReframeDelta. The bearer is an ordered pair of problem-frame epistemes. One local rule compares the earlier and later frame on an ordinal Scale such as None | Local | BoundaryShift | Systemic; a boundary or scope diff and a changed causal map support the reading. The frame change does not by itself prove improvement, so state Use-Value separately.
• Compositionality. The bearer is the design or episteme being assessed. One local rule requires reuse of at least a declared number of components and evidence of at least one new relation among them; it may return a boolean plus a separately defined structure reading. Cite the component graph and component provenance.
• Transferability. The bearer is the design, result, or episteme whose use is tested in one named receiving setting. One local ordinal Scale is not supported | supported with stated loss | supported for the stated use. Cite receiving-use pilot evidence and the preserved and lost meaning; use an F.9 Bridge only when a relation between different reference-scheme senses actually obtains.
• DiversityOfSearch. The bearer is a finite set of dated Work attempts. Count distinct approach classes under a declared local typology, optionally as a rate over a stated time window, and cite the tagged Work and typology. Cosmetic variants do not create new classes.
• Time-to-First-Viable. The bearer is one Work episode. Measure elapsed time from a declared start to the first dated result that passes the stated viability criterion; cite the timestamps and passing evidence. If no result passed, report not yet obtained or a right-censored duration rather than the time to the first runnable output.
• Risk-BudgetedExperimentation. Compare the applicable WorkPlan with the resulting dated Work set. One local rule reports planned exploratory resource use divided by the allowed risk budget and the realized ratio separately, with any overrun visible. Cite the WorkPlan, actual Work, and resource evidence; the reading does not grant the budget or authorize the Work.

These examples do not create another universal characteristic family. Readings about actual attempts, elapsed time, or realized experimentation depend on dated Work; planned experimentation depends on a WorkPlan.

#Other domain characteristics

The six applied readings above are examples, not the extension boundary. A configuration may select other Characteristics already established for the current use—for example, time or cost to probe, evidence sufficiency, safety or ethical risk, option value, or regret risk. Keep each selected Characteristic's bearer, Scale, polarity, defining source, Method, and evidence. A safety or ethical must remains an eligibility condition through ConstraintFit; evidence sufficiency does not become creativity; and scope does not become another coordinate merely because every claim needs one.

When one comparison covers several components, attempts, or Work occurrences, identify the lawful aggregation separately for each Characteristic. For example, compatible costs may sum, all declared must-constraints may have to pass, a domain risk rule may use its own conservative combination, and evidence may be combined for a named assurance claim under B.3. These are examples, not C.17 defaults. If no declared aggregation supports the combined reading, keep the component results separate.

Any prior or default used for Novelty, evidence, risk, or another selected reading remains a separately supported model or policy claim with its source and edition. C.17 does not publish domain priors merely because a reusable configuration is convenient.

#Results, profiles, and assessment Work

#Coordinate claims

For each selected characteristic, choose one truthful route:

• cite an already constituted C.16 measurement-result episteme and its complete measurand, Characteristic, Scale, Method, dated measurement Work, measurand relation or A.6.1 binding, any further bindings required by the claim, time, uncertainty, and evidence chain;
• if the current action measures the coordinate, constitute that complete C.16 chain before using the value;
• if the claim applies a declared criterion without measuring, state a C.2.1 ascription and its rule.

A numeric model output, criterion label, dashboard cell, or formula alone is not a measurement result.

#Aggregate result

CreativityEvaluationResult is one bounded C.2.1 episteme about the bearer. Its ClaimGraph cites the selected coordinate claims, CharacteristicSpace and specification, scope, use, window, evidence, current eligibility consequence, and any frontier or incomparability conclusion. It is not an arithmetic total and does not create a universal creativity kind.

The result may say that a bearer is eligible for one comparison, incomparable on a missing coordinate, or non-dominated within one declared set. It does not choose, approve, publish, retain, or enact the bearer.

#Optional payload, record, and rendering

Use CreativityProfile only as the local name for the optional non-arithmetic payload of coordinate-claim references, their arrangement, and current frontier or incomparability annotation. A table, chart, or dashboard is a separate representation. A CreativityEvaluationRecord is a separate optional episteme that packages references for a named receiver.

Do not alternate among result, profile, record, and dashboard as if they were synonyms.

#Comparison, gates, and selection boundary

1. Never use Novelty alone to approve or prefer a bearer. Pair it with Use-Value or the relevant ConstraintFit gate.
2. State the selected characteristic subset, polarities, eligibility conditions, and comparability basis of every dominance claim.
3. Preserve partial orders and incomparability. A Pareto or constraint-bounded Front follows from the declared rule; a visually pleasing hull is not a frontier.
4. Do not force one scalar creativity score. If a receiving policy uses an index, publish its weights or curves, admissible transformations, uncertainty treatment, sensitivity, and drift rule while keeping the primitive coordinates queryable.
5. Do not average ordinal Scales without an accepted model that supports the conversion.
6. A result may state a frontier relation over a declared set. Use C.18 to maintain the current Front and Archive, C.19 to state pool treatment and tie-break policy, G.5 to declare selector-facing results, and C.11 to make the choice.

#Evidence, uncertainty, and resistance to gaming

Every quantitative result names its evidence, calibration, uncertainty, and validity window. Keep aleatory and epistemic uncertainty separate and state any rule that combines them. A claim imported from another source or scale states the preserved meaning, lost meaning, direction, receiving use, and evidence limit; use F.9 only when an actual Bridge between reference-scheme senses is needed.

Apply these anti-Goodhart guards:

• pair Novelty with Use-Value or ConstraintFit;
• freeze the comparison corpus, encoder or model, and Scale edition for the result; a load-bearing change creates a new result basis;
• keep Illumination as telemetry unless an explicit policy promotes it for a named use;
• check delayed practical consequences, such as retention, maintenance burden, or cost-to-serve, before celebrating a proxy win;
• connect DiversityOfSearch to the declared experimentation allowance and report overspend;
• retain primitive coordinates and sensitivity when a composite index is used.

When a corpus, inclusion rule, model, encoder, Scale, criterion, window, or cross-source premise changes, leave a compact change account: the previous and new editions, what changed in the basis, which coordinates and aggregate conclusions reopen, whether eligibility or frontier membership changed, and which next observation can settle the difference. Latest is not a reproducible selector. Use B.4 and G.11 for the refresh; the change account explains the comparison and does not replace the new result.

Goodhart's law is a useful historical warning, not evidence for any result. The safeguards above do the operational work.

Evidence can support a result without establishing assurance. Ordinary use stops with proportionate evidence. Enter B.3 only when an actual named assurance claim is current.

When that receiving use requires independent assessment or segregation of duties, name the Systems that performed any bearer-producing, coordinate-measurement, or overall-assessment Work and the assignments relevant to the independence claim. State the relevant conflict or independence evidence. Do not impose this assurance arrangement on an ordinary qualitative discussion that makes no independence claim.

#Cross-scale and cross-source limits

When a reading moves across scales, state which Characteristics are preserved, aggregated, projected, or lost and cite the actual aggregation, projection, transition, temporal cross-scale, or Bridge relation on which the reading depends. Keep a plain distortion note next to the projection. If the receiving use needs the named A.0 qualifier structure, use A.0:QF.2a and its optional OutcomeMapRef, TransitionRelationRef, or BridgeDistortionNote; otherwise the plain loss statement is enough.

Different projections of the same retained set or frontier may preserve different information. A useful projection is not automatically information-preserving, and one atlas-like view does not cancel the original Front, Archive, or result.

When claims come from different source schemes or corpora, do not compare by matching labels. State the actual mapping, Bridge when required, direction, loss, calibration, and receiving use. A target-use pilot may improve evidence for that use; it does not make the source and target bases identical.

#Worked cases

#Pump design: stop early or open the measurement branch

Identify P-22 as the exact design episteme. Compared with the admitted five-year pump-design set, its inspected split-clamp arrangement is a supported difference. Assembly evidence supports shorter assembly, and the candidate must not require new tooling. State what was inspected and the limit of that support. Stop here when the discussion needs neither a quantified coordinate nor a reusable result.

To use Novelty value 0.42, cite P22-NoveltyResult-4, an already constituted C.16 measurement-result episteme. Its chain identifies P-22 as measurand, the Novelty Characteristic and Scale, exact similarity Method, the calibrated [0,1] CAD-graph similarity result used by the declared Novelty construction, encoder and model edition, uncertainty, dated measurement Work, actual bindings, time, and evidence. The Method compares P22-CADGraph-7, produced from and representing P-22 under CADGraphProjection-2, with graphs produced by the same projection for every corpus design; the result states that this projection omits surface finish and manufacturing tolerances. If the current action measures novelty, constitute that chain first. State the no-tooling-change coordinate as a C.2.1 ascription under its declared criterion unless it was independently measured.

One CreativityEvaluationResult may cite those coordinates and state only that P-22 is eligible for the current comparison and lies on the declared non-dominated set. Building that result does not assert separate overall-assessment Work.

If an audit later asserts that an overall assessment occurred, identify the evaluator System, its exact assignment, PumpCreativityAssessment-17 Work, and PumpCreativityAssessmentMethod-2; state that the Work enacts the Method. The MethodDescription explains the Method, the result states claims, and coordinate-measurement Work stays separate. Do not add an A.6.1 operation application merely because those values are recorded. If the audit separately asserts an exact operation application or binding, satisfy the current A.6.1 application account and cite that application.

#Software and algorithmic design

Software design. In this worked example, ETL-Parallel-12 is compared with 40 admitted internal pipeline designs through ASTGraphSimilarity-3, whose declared Novelty construction uses calibrated [0,1] similarities. The Method compares AST-graph representations produced by the same declared parser and projection for the evaluated design and every corpus design; the result states that runtime configuration and deployment topology are not preserved by that projection. Its Novelty result is 0.36. A fixed-workload benchmark reports an 18% lower p95 latency than the serial baseline, but the segregation-of-duties test fails. The benchmark run, corpus edition, nearest-neighbour report, and policy test are the evidence. The result therefore supports a latency gain but leaves the design ineligible for the stated use; redesign the isolation boundary and repeat the affected tests before any pool or choice decision.

Algorithmic search Work. SpikeSet-7 contains nine dated attempts in three declared approach classes over six hours. Tagged Work records support DiversityOfSearch = 3 classes; the first runnable output appeared after 2 h 10 m. The held-out viability test was never run, so Time-to-First-Viable is not established. The practical action is to run that test, not to relabel time-to-first-runnable as viability.

#Health analytics

For Cardio-Readmit-H4, the held-out AUROC is 0.79 against a 0.75 baseline under the frozen test set, clearing the declared uplift threshold of 0.03. The model card, held-out plot, and evaluation Work support that local Use-Value result. No receiving-use pilot has yet been performed at Hospital B, so Transferability there remains unsupported. Use the local result for the applicable pool or choice question, but leave the target-hospital claim open until pilot evidence exists. Use an F.9 Bridge only if the two hospitals' reference schemes require one.

#Product reframing

OnboardingFrame-v1 treats onboarding as one completion task; OnboardingFrame-v2 separates job setup from obtaining the first result. The declared ReframeDelta rule returns BoundaryShift, supported by the frame diff and a simpler causal map. In a four-week A/B comparison, v2 reduces median time-to-value by 22% against the control baseline, clearing the 20% objective. Exploratory Work used 9 of the 12 allowed staff-days, so the realized risk-budget ratio is 0.75 with no overrun. The frame diff, A/B report, WorkPlan, and Work records support the result. This evidence can inform the later choice; it does not make the choice.

#Scientific and policy proposals

Scientific proposal. ScalingRelation-S4 has Novelty 0.61 from the declared calibrated [0,1] similarity construction relative to the admitted literature corpus. The Novelty Method compares one text embedding that describes the proposal with embeddings produced by the same encoder for every corpus paper; the result names that projection and states that notation and experimental detail may be lost. Its Surprise is 4.1 bits per token under PriorModel-2, using the same versioned tokenizer and abstract-text sample unit for the proposal and model basis; the per-token normalization handles text length but does not support a claim about equations or full papers. The corpus, neighbour report, model calibration, and derivation evidence support those coordinates, but independent replication is missing. Report the proposal as a preliminary bounded result and seek replication before a reliance claim.

Policy proposal. In a municipal permit-triage pilot, Policy-P8 reduces median processing time by 12% against the prior-procedure baseline and passes the declared legal-form test. Subgroup error evidence required by the equity must-criterion is missing, so ConstraintFit and eligibility are not established. Keep the proposal out of an approval-facing comparison until the subgroup test is complete; the time result remains usable for its narrower operational question.

#Manager quick start

1. Name the bearer, comparison corpus, objective, and must-criteria.
2. State the smallest supported difference and consequence; stop if that answers the question.
3. When numbers matter, select the space and specification and build each coordinate through C.16 or C.2.1.
4. Compare with declared gates and a partial order; keep incomparability visible. When a retained set matters, report its declared Diversity_P or other needed set reading without turning that reading into selection policy.
5. Pass generation or retention questions to C.18, pool-policy questions to C.19, selector-facing declarations to G.5, and choice questions to C.11, with the result references each needs.

#Optional one-page comparison brief

When several people must reuse the same comparison, publish one short view containing only the fields they need:

• the working question, bearer, intended use, and stop;
• the finite corpus or reference set, inclusion rule, source editions, and window;
• the selected Characteristics, Scales, polarities, admissible operations, and missingness rules;
• the objective, must-criteria, eligibility consequence, and their sources;
• the Methods, models or priors actually used, with calibration, uncertainty, and evidence;
• the coordinate-result references and any declared set, frontier, or incomparability statement;
• the applicable C.19 policy, G.5 declaration, or C.11 choice reference when later work relies on one; and
• the basis-change and reopen condition when editions or evidence change.

This brief is a representation of the selected configuration and results. It does not create a result, policy, choice, WorkPlan, prior, or domain default. Put a reusable transform or objective form in the pattern that defines it and cite that definition here; do not make the brief a second source of the rule.

#Conformance checklist

#Common failures and repairs

#Consequences

Benefits. Teams can discuss novelty and value before building a metric stack; stronger results remain reproducible and comparable; trade-offs and incomparability stay visible; and generation, policy, choice, Work, evidence, and publication keep their own boundaries. No particular tool or rendering is required: an implementation is suitable when it preserves the declared configuration, result chain, evidence, and limits.

Costs. Quantified claims require a fixed corpus, explicit Scales and Methods, evidence, uncertainty, and edition discipline. Cross-source and cross-scale comparisons sometimes remain incomparable.

Limits. A C.17 result does not prove that a bearer is good, authorize Work, define universal novelty, supply a selection policy, or establish assurance. It makes the bounded claims and their dependencies inspectable.

#SoTA-Echoing and source use

Source-use boundary. The sources below change what a C.17 user inspects or reports. They do not install one creativity theory, automated judge, metric, or search algorithm as the FPF default. Reopen this source-use judgement when a cited source is corrected, retracted, or materially superseded; when new cross-domain evidence overturns one of the stated consequences; or when a proposal would make one automated metric, corpus, encoder, QD descriptor, or proxy score normative. Use G.11 for that refresh.

These decisions reinforce the existing route rather than add another assurance layer. In the pump case, inspect the admitted design set and tooling constraint; in the hospital case, keep the held-out result separate from unsupported transfer; in the policy case, keep the missing subgroup evidence as an eligibility gap. The source-use decisions therefore change the comparison and robustness work already required by the cases and checklist, not the practitioner-first entry.

#Open questions

The following are research questions, not current requirements:

• Under what evidence can a domain justify a stable distance across several creativity Characteristics without hiding their different Scales?
• How can several agents' partial frontiers be related without importing team-governance assumptions or forcing one scalar objective?
• What evidence is sufficient to treat an ordinal reading as interval-like for one declared frontier-estimation use?
• Which delayed observations best reveal when Novelty, Use-Value, or Illumination has become a gamed proxy?

#Relations

• Builds on: A.17, A.18, A.19, A.19.ECS, C.16, C.2.1, A.1.1, A.10, and B.3.
• Coordinates with: E.10.LRN for ambiguous learning-family wording, A.13 for exact evaluator recovery and any separate agency or autonomy claim, F.9 for an actual Bridge, F.18 for lexical candidate-family diversity, A.0:QF.2a for an optional structured cross-scale qualifier, B.4 and G.11 for evolution and refresh, A.15.1, A.15.2, B.1.6, A.3.1, and A.3.2 for Work, plans, resources, and Method descriptions, A.2.1 and F.6 only for an expressly consumed precise assignment-bound attribution, and A.6.1 only for a separately claimed application of one exact declared Mechanism operation.
• Supplies results to: C.18, C.19, and G.5 for their exact set-side questions, C.11.CRC only when one finite configuration-relative comparison is missing, and C.11 for choice, without taking over generation, set stewardship, pool policy, comparison, declaration, or choice.

#C.17:End

---

---

Last Updated: 2026-09-10 — upstream FPF commit a87d0ef4 (github.com/ailev/FPF)