Every source has a load rating.

Claim

Adoption rate is governed mainly by five perceived attributes of the innovation — relative advantage, compatibility, complexity, trialability, observability — plus the structure of the communication network. Adoption, not invention, is the rate-limiting step.

Evidence

A synthesis of thousands of diffusion studies across agriculture, medicine, public health and communications, accumulated over five editions. The S-curve and the attribute set replicate across domains and decades.

Doesn't cover

It will not tell you what to build. The adopter categories are statistical partitions of a curve, not personality types, and Moore's "chasm" is a later extension with much weaker support — do not cite it as Rogers.

In our process

Score every concept on all five attributes from the customer's perception before a pilot is designed. Low trialability plus low observability predicts slow uptake even where relative advantage is high — which is precisely the standing problem with home storage and dynamic tariffs.

Load-bearing stories: CASE-0019 · missing: Iowa hybrid seed corn diffusion

• Publisher page, Simon & Schuster purchase
• Borrowable copy, Internet Archive free

Claim

Innovation is a discipline with seven identifiable sources of opportunity — the unexpected, incongruity, process need, industry and market structure change, demographics, changed perception, new knowledge — ordered by decreasing reliability. New knowledge is the most glamorous source and the worst bet: longest lead time, highest failure rate.

Evidence

Practitioner synthesis from decades of consulting; no formal testing. The ordering is assertion, though later work on lead times is broadly consistent with it.

Doesn't cover

No method, no metrics, and a management world that no longer exists in several of its examples.

In our process

Use the seven sources as a scan taxonomy alongside STEEP-V. Regulatory and market-structure change is Drucker's highest-yield category for a utility — while the deep-technology bets the lab finds most exciting sit in his worst-odds category. Say that out loud in portfolio reviews.

• Publisher page, HarperCollins purchase
• Borrowable copy, Internet Archive free

Claim

A large share of commercially important innovation originates with users rather than manufacturers. Lead users experience needs ahead of the market and routinely build their own solutions, unpaid, before any supplier prices the problem.

Evidence

Functional-source studies across many industries, plus representative national surveys of consumer innovation in several countries finding single-digit percentages of adults developing or modifying products for their own use — a small share of the population but a very large absolute volume. Replicated across countries; the measurement of "importance" remains debated.

Doesn't cover

Almost nothing on how to industrialise a user innovation, which is exactly the hard part for us.

In our process

Balkonkraftwerk and DIY plug-in storage are a textbook lead-user field. Treat the German and Austrian forums, the ESPHome and OpenEMS repositories, and Home Assistant integration activity as a primary signal source — earlier, cheaper and more honest than any trend report. Both later books are free from MIT Press.

• All three books, free PDF (CC licence), author's MIT page free
• Democratizing Innovation, MIT Press free
• Democratizing Innovation, OAPEN mirror free

Claim

Well-managed incumbents fail because their resource-allocation processes rationally starve low-margin, small-market technologies that later improve enough to take the mainstream.

Evidence

Originally the disk-drive industry, later extended by case selection. Replication is poor: King and Baatartogtokh examined 77 cases claimed for the theory and judged only a small minority to satisfy all four of its elements. Lepore attacks the case selection and the retrospective fit directly.

Doesn't cover

Prediction. It cannot tell you which technology will disrupt you, or when. Every forecast built on it is a story with a bibliography.

In our process

Read the dilemma as an organisational diagnosis — "our margin structure will kill this bet" — and respond with mandate and P&L separation, which is a governance move. Never use it as forecasting evidence in a gate paper.

Load-bearing stories: missing: The disk drive industry 1976-1992

• Lepore, 'The Disruption Machine', New Yorker 2014 paywalled
• King & Baatartogtokh, MIT Sloan Management Review 2015 free
• Borrowable copy, Internet Archive free

Claim

Technologies are combinations of existing technologies, recursively assembled around exploited natural phenomena. Novelty is mostly recombination, and every new technology becomes a building block that makes the next one possible.

Evidence

Conceptual and historical rather than tested, though consistent with the recombination patterns visible in patent data and with the history of engineering practice.

Doesn't cover

Markets, organisations, money. It is a theory of technology, not of business.

In our process

The most effective antidote to genius-flash mythology in a technical team, and the right lens on our own architecture work: a sovereign firmware stack is a recombination of existing blocks — control cores, protocol adapters, secure-boot chains — not an invention. Decompose before estimating.

• Publisher page, Simon & Schuster purchase
• Borrowable copy, Internet Archive free

Claim

The closed-laboratory model of corporate R&D stopped paying around the end of the twentieth century; firms profit more by letting ideas, licences and people flow across their boundary in both directions.

Evidence

Argued from a handful of deliberately chosen cases — Xerox PARC's spin-offs, IBM, Intel, Lucent — plus the observable erosion of the conditions that made closed labs work (worker mobility, venture capital, distributed university research). Coherent and influential; not a tested proposition, and the subsequent academic literature is a sprawl of definitions more than of replications.

Doesn't cover

When openness loses: appropriability regimes where the licence-out leaks the business. The book names the risk and moves on; von Hippel (SRC-0003) covers the user side of the boundary with better evidence.

In our process

Carries the framing for any make-buy-license discussion at the portfolio gate. As language it is load-bearing; as a promise that openness raises returns it may not be cited.

• No link recorded yet — see CONTRIBUTING.

Claim

New ventures fail from lack of customers, not lack of product. Customer discovery and validation must run as a formal process in parallel with product development, and no scaling happens until both are complete.

Evidence

Practitioner method with no controlled outcome data. The strongest external signal is institutional adoption rather than trial results — it became the basis of the US NSF I-Corps curriculum.

Doesn't cover

Regulated, certification-bound, capital-heavy products. You cannot iterate a grid-connected device weekly, and the method quietly assumes you can.

In our process

Blank is the original; Ries's Lean Startup is the popularisation and adds less than its reputation suggests. Use the discovery/validation split as the skeleton of the concept phase, and state the certification cycle as an explicit cap on iteration rate rather than pretending it away.

• Author's book page, Steve Blank free

Claim

Desirability, viability and feasibility assumptions can each be tested with a catalogued experiment, chosen by evidence strength, cost and setup time.

Evidence

None beyond craft. The catalogue is the contribution; there is no theory being defended and no outcome data offered.

Doesn't cover

Technical feasibility and grid-code compliance. The catalogue was built for digital and service propositions and has nothing for "will this pass EN 50549 recertification".

In our process

The best experiment catalogue in print. Pair every riskiest assumption from the Heilmeier charter with one named experiment and a kill threshold declared before the experiment runs. An experiment without a pre-declared threshold is a demonstration.

• No link recorded yet — see CONTRIBUTING.

Claim

A firm should run an explore portfolio and an exploit portfolio side by side, with different metrics, funding rhythms and governance for each.

Evidence

A practitioner framework resting on ambidexterity research that it does not itself test. Card 09 is where that evidence actually lives.

Doesn't cover

The internal politics that decide whether an explore portfolio is permitted to exist at all — which is the binding constraint, not the framework.

In our process

Use the portfolio map as the artefact for the annual conversation with the business units. Its real function is making the explore-to-exploit ratio visible enough that somebody has to defend it.

• No link recorded yet — see CONTRIBUTING.

Claim

Firms sustain innovation by being ambidextrous: structurally separate exploratory units, a common identity, targeted access to the core's assets, and a senior sponsor with real authority. Separation without sponsorship gets orphaned; integration without separation gets starved.

Evidence

An actual academic literature, with meta-analytic support for a positive association between ambidexterity and firm performance. Causality and construct measurement remain genuinely contested.

Doesn't cover

Small organisations, and — critically for us — what happens when the sponsor changes role.

In our process

The closest thing to real evidence on why labs die. It imposes a testable requirement on us: name the senior sponsor with budget authority, and name our asset-access rights to the core — grid teams, metering data, customer base. If either cannot be named, that is the lab's actual top risk, not any technology on the register.

• No link recorded yet — see CONTRIBUTING.

Claim

Initiatives die in execution, not ideation. Each needs a dedicated team, an explicit partnership contract with the core organisation, and a plan judged on learning rather than variance-to-budget.

Evidence

Case-based, and unashamed about it.

Doesn't cover

The front end entirely — by design. It starts where the ideation books stop.

In our process

The boring, correct book on this shelf. Its single operational demand — never evaluate an experimental initiative on budget variance — is the one change that would most improve how our projects are reviewed, and the one most likely to be refused.

• No link recorded yet — see CONTRIBUTING.

Claim

Customers buy to get a job done. Needs are stable desired outcomes, and opportunity is scored as importance + max(importance − satisfaction, 0), which double-weights importance and refuses to penalise already-satisfied needs.

Evidence

Practitioner method carrying vendor-reported success rates that have never been independently replicated. Treat the headline hit-rate statistic as marketing. The formula itself needs no such support — it is defensible on its own arithmetic, which is why we adopted it.

Doesn't cover

Needs the customer cannot articulate, and needs created by regulation rather than by the customer — a large fraction of ours.

In our process

This is the direct lineage of our opportunity-scoring model. Keep the refined formula. Stop quoting the success statistics, including internally.

• Author's book page, Strategyn free

Claim

Under extreme uncertainty, the fastest route to a viable product is a Build–Measure–Learn loop: state assumptions as hypotheses, test them with the smallest artefact that produces evidence, and pivot or persevere on the result.

Evidence

Founder narrative and consulting practice — the book offers no outcome data. The underlying principle, hypothesis-driven experimentation, has since acquired replicated RCT support (SRC-0031), but that evidence tests the scientific core, not the book's full apparatus (innovation accounting, engines of growth), which remains untested.

Doesn't cover

Contexts where the minimum viable artefact is expensive or regulated — hardware, infrastructure, clinical, grid-connected anything. "Ship fast and measure" assumes shipping is cheap and failure is private.

In our process

Supplies the loop vocabulary the team already speaks. Cite SRC-0031, not this book, when the question is whether hypothesis testing works; cite this book only for how to phrase the loop.

Load-bearing stories: missing: Dropbox demo-video MVP · missing: IMVU and the shipped-too-early lesson · missing: Zappos founding experiment

• No link recorded yet — see CONTRIBUTING.

Claim

A business model can be described completely enough for design work on one page of nine blocks — customer segments, value propositions, channels, relationships, revenue, key resources, activities, partners, costs.

Evidence

None offered for outcomes. The contribution is a shared language, and the evidence for the language is its uptake: Strategyzer reports "over 6 million" practitioners (vendor-reported, undated methodology, unaudited). One exploratory study of 271 accelerator teams associated heavy Canvas use with better pitch-competition results — competition performance, not survival or returns.

Doesn't cover

Whether any canvas is true. The tool records a theory of the business; it has no mechanism for testing one, which is why it pairs with SRC-0007 rather than replacing it.

In our process

The one-page representation every concept must have before it may request experiment budget — because a model that cannot be drawn cannot be tested. The canvas is admissible as a map, never as evidence.

• No link recorded yet — see CONTRIBUTING.

Claim

Customers hire products to make progress in a circumstance — a "job to be done" — and innovation succeeds by targeting the job, not the demographic.

Evidence

Anecdote and consulting cases (the milkshake story chief among them), presented as theory. No systematic outcome data; the authors call it a theory of causality but do not test it. The clearest book-length statement of the jobs lens, which is a real contribution of language, not of evidence.

Doesn't cover

Jobs the customer cannot articulate or does not know they have — and any market where the buyer, user and payer are different people, which is most of B2B and all of regulated infrastructure.

In our process

The readable introduction handed to anyone new to the jobs lens; the working method behind it is SRC-0011. Neither may appear in a gate paper as a finding.

Load-bearing stories: missing: The milkshake job interviews

• No link recorded yet — see CONTRIBUTING.

Claim

Creative capability is a learnable behaviour, not a trait: fear of judgment is the main suppressor, and guided small successes with prototyping restore it in most adults.

Evidence

IDEO client stories and the Kelleys' teaching experience at the Stanford d.school, with a supporting nod to Bandura's self-efficacy research — the one genuinely evidenced thread, though the book borrows rather than tests it. Warm, persuasive, undemonstrated.

Doesn't cover

Whether confident people produce better innovation outcomes. The book measures nothing downstream of the workshop.

In our process

Read for facilitation craft when a group is afraid of looking wrong — a real and recurring failure mode. It may set the tone of a workshop; it may not justify a concept.

• No link recorded yet — see CONTRIBUTING.

Claim

Design work moves through five teachable modes — empathize, define, ideate, prototype, test — and the card deck of methods under each mode is enough to run the process.

Evidence

Teaching practice at the d.school. The kit is free, widely used and offers no outcome data; the five-mode structure is pedagogy, not a tested model of how design actually proceeds.

Doesn't cover

Selection and economics. Nothing in the deck says which ideas deserve the process, what evidence should kill one, or what any of it costs — the S4/S5 half of the journey is someone else's problem.

In our process

The facilitation deck for discovery workshops (S2), used for method cards, not for the five-stage theology. Downstream of any workshop, evidence rules are SRC-0007's and SRC-0031's, not the deck's.

• No link recorded yet — see CONTRIBUTING.

Claim

Human-centered design proceeds through inspiration, ideation and implementation, and the guide's 57 methods let a team without design training run the process in the field.

Evidence

IDEO.org's development-sector project practice. Written for social-sector work, method by method, with worked examples and no outcome data. The three-phase frame is the same craft tradition as SRC-0032 with a different cut.

Doesn't cover

Business viability. The guide's origin in donor-funded projects shows: it tests desirability with users and treats "who pays, sustainably" as out of frame — precisely the question that kills most concepts at our gate.

In our process

Field interview and synthesis methods for S2 work outside our home domains, where cheap structured empathy beats no structure. Same standing rule as the Bootleg: cards yes, cosmology no.

• No link recorded yet — see CONTRIBUTING.

Claim

"Innovation" was a term of accusation for most of its recorded life — heresy, sedition, dangerous novelty. Its transformation into an unqualified good is a recent, political and largely commercial achievement, not a discovery.

Evidence

Documentary conceptual history across several centuries of sources. Strong within its own type; there is nothing here to replicate in the experimental sense, and nothing that needs it.

Doesn't cover

How to do any of it. That is the point of the book.

In our process

The historical spine of any honest account of what this lab is for. Reach for it the moment somebody in the room uses "innovative" as a synonym for "good", which is the moment a decision stops being examinable.

• Publisher page, Routledge purchase
• Catalogue record, Stanford SearchWorks free

Claim

"Innovation-speak" is a sales pitch, distinct from and often inversely related to actual innovation. The cultural elevation of novelty has systematically defunded maintenance — the work that determines whether infrastructure keeps working at all.

Evidence

Two historians of technology arguing from infrastructure and maintenance-spending evidence. Deliberately polemical; the argument is stronger than the data offered for it.

Doesn't cover

It undersells the cases where new capability genuinely was the answer, and it has no method to replace what it attacks.

In our process

The most uncomfortable and most necessary entry on this shelf for a distribution business, where novelty-versus-maintenance is not a metaphor but a capex line. Read it before writing anything about what the lab is for.

• The Maintainers - Vinsel and Russell's research network, not the book free

Claim

Novelty-stage invention is only one of four stages of production. Most regions and most firms should compete on design and prototyping, second-generation improvement, or production and assembly — and copying the Silicon Valley model outside its conditions wastes public money reliably.

Evidence

Comparative case studies across regions and decades. Qualitative but disciplined, and independently recognised in policy.

Doesn't cover

Firm-level method. It is political economy, and it will not tell you how to run Tuesday.

In our process

The strategic frame for a Czech lab inside a German utility. Our comparative advantage is almost certainly second-generation improvement and systems integration rather than frontier novelty — and Breznitz gives that a respectable name instead of treating it as a consolation prize.

• Publisher page, Oxford University Press purchase

Claim

R&D productivity is measurable as an output elasticity — her Research Quotient — and once measured, several standard prescriptions stop surviving contact with the data, including "buy startups rather than build" and "spend more on R&D".

Evidence

Production-function estimation over a large panel of US public firms, publicly funded. Unusually testable for this field; the measure is the author's own and independent validation remains thin.

Doesn't cover

Non-R&D innovation, and any firm that does not report R&D separately — which includes most of the European utility sector.

In our process

The argument to have ready when "let's just acquire a startup" arrives. Also an uncomfortable mirror: our own effectiveness is measurable in principle, and we have not defined the measure. Defining it before someone else does is a governance act.

• No link recorded yet — see CONTRIBUTING.

Claim

Three failure classes recur: inventions adopted and later found harmful; inventions that dominated and were then discarded; and inventions permanently a few years away.

Evidence

Historical case analysis in Smil's usual quantitative register — dates, volumes, costs, and an allergy to narrative.

Doesn't cover

Organisational method. Smil has no interest in how you run a lab, and says so.

In our process

Our calibration instrument. Run every technology we are excited about against the three classes before it enters the trend register. Note that some of our own live entries sit uncomfortably close to the third class, and that our records already flag the unresolved offtake question that puts them there.

• Publisher page, MIT Press purchase

Claim

Decentralisation produces exploration; bureaucracy produces scale; progress requires both, in sequence. Societies stagnate when the institutions that must scale a technology cannot adapt to it — and nothing, including AI, makes progress automatic.

Evidence

Comparative economic history across roughly a thousand years. Synthesis and argument rather than identification; recent, and the critical response is still arriving. Unusually decorated for a book this new — winner of the PROSE Award in Economics (Association of American Publishers); shortlisted for the Financial Times and Schroders Business Book of the Year Award and for the Lionel Gelber Prize; a Financial Times, Five Books and Bloomberg Best Book of the Year; a Foreign Policy Best Book of the Summer; one of OODALoop's Top 10 Technology, Security and Business Books of 2025; longlisted for the Non-Obvious Book Awards. Note what that list measures: how well the book argues, not whether it is right. It stays rated C.

Doesn't cover

Firm-level tactics. It operates two levels above anything we decide.

In our process

The cleanest available statement of our structural position: the lab is the decentralised exploration function, the operating organisation is the scaling bureaucracy, and Frey's point is that neither is the hero. Useful vocabulary for conversations that would otherwise become lab-versus-line.

• Publisher page, Princeton University Press purchase
• Princeton Best of 2025 — the accolades list free
• Review, LSE Review of Books free
• Review, Foreign Affairs paywalled
• Author's page free

Claim

Failures divide into basic (known process, avoidable), complex (multiple causes in familiar territory) and intelligent (new territory, hypothesis-driven, right-sized, acted upon). Only the third should be tolerated, and surfacing any of them requires psychological safety.

Evidence

The taxonomy is a practitioner framing, but it rests on Edmondson's psychological-safety research, which is among the better-replicated findings in organisational behaviour.

Doesn't cover

Safety-critical engineering, where loose application is actively dangerous. A grid-connected hardware failure is a basic failure. It does not become intelligent by being interesting.

In our process

Gives us defensible language for "fail fast" inside a regulated utility. The operational test: an intelligent failure must have been pre-declared as a hypothesis with a kill threshold. Anything else is a failure with good public relations.

• Author's book page free
• Publisher page, Atria purchase

Claim

Firms that industrialise controlled experiments — thousands of A/B tests under statistical discipline, with the organisational authority to act on losing results — systematically out-learn firms that run occasional pilots.

Evidence

Two decades of Thomke's field research at HBS plus documented programmes at Booking.com, Microsoft, Amazon and LinkedIn, where experiment volume and decision discipline are observable. Grounded in real programme data; the step from "these firms experiment at scale and prosper" to causation is not closed, and the firms studied are digital, high-traffic and self-selected.

Doesn't cover

Low-traffic businesses. The statistics of A/B testing need sample sizes a B2B infrastructure firm never sees; for us the transferable content is the organisational discipline, not the testing machinery.

In our process

The evidence base for the standing rule that a losing experiment result kills the variant regardless of whose idea it was. Cite it for experimentation governance (S7), not for any specific test design.

• No link recorded yet — see CONTRIBUTING.

Claim

Teaching founders to treat their idea as a theory — explicit hypotheses, deliberate tests, evidence-based decision rules — measurably changes outcomes: more termination of weak ideas, more decisive pivots, and performance gains consistent with more efficient search.

Evidence

A randomized controlled trial (116 Italian startups, Management Science 2020) in which the treatment group received the same entrepreneurship training as control but framed scientifically, followed by a 2024 Strategic Management Journal replication programme of four RCTs totalling 759 firms. Effects on termination and pivoting replicate. This is the rare spot in the innovation literature with genuine causal evidence.

Doesn't cover

Brand-versus-brand questions. The trials test the scientific core that Lean Startup and Testing Business Ideas operationalise — they do not show any named framework beats another, and the outcome measures are early-stage (termination, pivots, early revenue), not long-run survival or returns.

In our process

The citation that carries the experiment-before-build gate rule. When someone asks why we demand pre-declared kill thresholds, this is the answer — an A-rated one.

• No link recorded yet — see CONTRIBUTING.

Claim

Defines innovation for statistical and policy purposes, including the product/process distinction and the threshold that something must have been made available to users or brought into use to count at all.

Evidence

An international standard by construction, not a finding. Its authority is institutional.

Doesn't cover

Anything about how to succeed. It is a measurement instrument.

In our process

Whatever we call innovation internally, this is the definition Brussels and the statistical offices are already using. Its "brought into use" threshold quietly disqualifies most of what organisations call innovation, including some of ours — a useful test to apply before a portfolio review rather than during one.

• Oslo Manual 2018, OECD, full text free
• OECD publication page free

Claim

A readiness assessment is only meaningful if it identifies critical technology elements, states the evidence required at each level, and produces a maturation plan. The number alone carries nothing.

Evidence

Expert-consensus best practice built on programme reviews and documented failures. Not experimental; unusually well-audited for a guidance document.

Doesn't cover

Integration and manufacturing readiness, market risk, and — the recurring criticism — the fact that a naked TRL label on a slide is exactly the practice it warns against.

In our process

Already the backbone of our feasibility work. The discipline lives entirely in the evidence requirements, not the scale. Keep the CTE register; keep refusing bare TRL numbers in gate papers.

• GAO Technology Readiness Assessment Guide free

Claim

Innovation can be managed through a certifiable management system, in the same structural family as ISO 9001.

Evidence

None on outcomes. No study demonstrates that certification improves innovation results, and given the measurement problem it is not obvious what such a study would look like.

Doesn't cover

The possibility that the auditable version of a process and the effective version are different processes.

In our process

Read it so we are not surprised when group functions propose it, and so we can adopt the parts that improve traceability while resisting a wholesale certification programme. The right posture is informed and specific, not reflexive.

• ISO/TC 279, Innovation management free
• ISO 56001:2024 purchase

Claim

Innovation studies is an actual research field with accumulated findings, and most popular frameworks were either tested there years ago or never entered it at all.

Evidence

The field itself, surveyed chapter by chapter by the people who built it.

Doesn't cover

Readability and speed. It was not designed for a Tuesday afternoon, and it is now two decades old — check the journals for anything post-2005.

In our process

The standing check before adopting any framework: spend twenty minutes finding out whether the field already knows about it and what it concluded. That habit alone would have saved us the worksheet formats we later discarded.

• No link recorded yet — see CONTRIBUTING.

Claim

Claims made in the trade-book layer are testable, and the journals are where they get tested.

Evidence

Peer review, replication attempts, and the accumulated record of frameworks that did not survive contact with data. Research Policy in particular is where the popular canon goes to be checked and usually fails.

Doesn't cover

Anything actionable on its own timescale. Paywalls are a real constraint; check for author preprints and open-access mirrors before assuming a paper is unreachable.

In our process

The escalation path whenever a framework is about to influence a gate decision. If a claim matters enough to change a decision, it matters enough to spend twenty minutes in the literature first.

• Research Policy, Elsevier paywalled
• R&D Management, Wiley paywalled
• Technovation, Elsevier paywalled

Claim

Innovation management is a learnable organisational capability with recurring patterns — search, select, implement, capture — that hold across sectors, firm sizes and technology generations.

Evidence

The standard academic textbook, now in its eighth edition, synthesising several decades of innovation-management research with references at chapter granularity. The synthesis is broad and systematic; the underlying studies vary widely in strength, and the book inherits that variance. Rated B for the synthesis, not for any single claim inside it.

Doesn't cover

Depth. As a survey it flattens live controversies into settled-sounding frameworks, and its process models describe how firms organise innovation, not whether those arrangements cause success.

In our process

The reference desk's first stop: before citing any single-framework book on a question, check what the textbook literature already says about it. A framework that contradicts the survey without argument is carrying marketing, not knowledge.

• No link recorded yet — see CONTRIBUTING.