Royce 1970 and the Waterfall Misreading

Overview

A paper arguing against waterfall became waterfall's founding citation through three ordinary mechanisms. None of them required anyone to misread it on purpose. Winston Royce drew the single-pass sequence early in his August 1970 WESCON paper, called it "risky" and said it "invites failure", then spent the remaining pages patching it. In 1976 two colleagues from his own firm, TRW's Bell and Thayer, credited him with introducing "the concept of the 'waterfall' of development activities" and reproduced the simple diagram rather than the patched ones. A decade later the US Department of Defense wrote phase-gated documentation into DoD-STD-2167, and a respectable 1970 citation suited a process that auditors and fixed-price contracts already favoured.

Sources: Hebrew University of Jerusalem course materials (1970) (↗); dwheeler.com (2018) (↗); Kallokain blog (2023) (↗); arXiv (2023) (↗); IEEE Computer (2003) (↗)

An honest assessment of Royce contradicts both popular stories. He did not invent the sequential model. Barry Boehm noted in 1987 that Benington and Hosier had earlier approximations. Nor was Royce a proto-agilist. His remedy was more design up front, substantial documentation, a pilot build ("do it twice"), planned testing and customer reviews, with routine iteration confined to adjacent phases. That is a sequential process with feedback and a prototype. It sits much closer to Boehm's 1988 spiral than to Scrum.

Sources: Wikipedia (2026) (↗); martinfowler.com (2018) (↗); IEEE Computer / ACM Digital Library (1988) (↗)

The past 18 months moved the story on two fronts. In March 2025 a memo from Defense Secretary Pete Hegseth made the Software Acquisition Pathway the preferred route for Pentagon software. This followed a 2023 GAO finding that some weapons programmes still ran on "waterfall". The institution that turned Royce's diagram into contract structure is now formally dismantling it. Meanwhile the "Royce was misread" correction, once an agile insurgent's talking point, has become orthodoxy. A 2025 ProjectManagement.com explainer, a March 2025 podcast episode and an October 2025 arXiv history all retell it, and sceptics in the Laurent Bossavit tradition now argue the correction is itself folklore.

Sources: Reuters (syndicated by Arizona Digital Free Press) (2025) (↗); Bloomberg Law (2023) (↗); ProjectManagement.com (2025) (↗); Oddly Influenced (podcast) (2025) (↗); arXiv (2025) (↗); Leanpub (Laurent Bossavit) (2013) (↗)

The evidence is uneven. The textual facts about Royce's paper are short and checkable. The causal story linking his paper to defence standards rests on a handful of secondary sources, chiefly Larman and Basili, plus personal testimony from his son.

Timeline

1956

• Benington describes stagewise production for SAGE

1957

• Incremental development already in use on early projects

1970

• Royce warns that single-pass development invites failure

1976

• Waterfall label attached to Royce's development activities

1985

• DoD-STD-2167 codifies phased, document-driven development

1987

• Royce paper reprinted at ICSE 9
• Boehm concedes prior art and Royce's prototyping

1988

• DoD-STD-2167A aims at life-cycle neutrality
• Spiral model declares waterfall dead

1994

• MIL-STD-498 replaces 2167A with incremental and evolutionary options

1998

• Timeboxed iterations from CCPDS-R written up in RUP-style framework

2001

• Agile Manifesto published

2003

• Iterative practice traced back to the mid-1950s

2013

• The misreading story itself challenged as folklore

2023

• GAO finds weapons programmes still on waterfall

2025

• Pentagon makes Software Acquisition Pathway the preferred route

Key Findings

Figure 2 was a warning label, not a recommendation

The core reading is consistent across practitioner and academic lanes. Royce presents the plain sequence, then immediately calls the implementation "risky" and says it "invites failure". He never uses the word "waterfall". His Figure 3 adds feedback, but only between adjacent steps. He identifies the real hazard as long backtracks, where testing exposes a design flaw and forces a redesign. Close readings by Sam Hogarth, Andreas Zwinkau, the Pragtob blog and Codemanship all agree on this.

Martin Fowler supplies the counterweight agile retellings tend to drop. Royce called his corrected approach "fundamentally sound", and that approach remains recognisably sequential.

Sources: Hebrew University of Jerusalem course materials (1970) (↗); Sam Hogarth's blog (2023) (↗); Zwinkau blog (2015) (↗); Pragtob blog (2012) (↗); Codemanship blog (2023) (↗); martinfowler.com (2018) (↗)

Spacecraft, not software products

Royce framed the paper as "personal views" from nine years of spacecraft mission planning, commanding and post-flight analysis software. He admitted uneven success on time and cost. That context explains his choices. Late discovery of a design flaw in a mission system on 1960s hardware was ruinously expensive, so front-loading design and building a pilot made economic sense. Customer involvement meant formal reviews with a government sponsor, not a product owner in the room.

One small trap: some web summaries place him at Lockheed, where he worked later. The 1970 paper came out of TRW, the same firm as Bell, Thayer and Boehm.

Sources: Hebrew University of Jerusalem course materials (1970) (↗); BibSonomy (1970) (↗); Wikipedia (2026) (↗)

The name came from down the corridor

Bell and Thayer's 1976 ICSE-2 paper, "Software requirements: Are they really a problem?", is the earliest documented use of "waterfall" for Royce's activities. A 2023 arXiv simulation study and the 2025 Saravanos history both confirm it. Blog analyses of the passage say the authors reproduced only the early sequential diagram.

Bossavit's citation archaeology adds a twist. Early literature often wrote "Boehm's waterfall", so Royce's name became attached gradually. Boehm's own 1976 IEEE Transactions on Computers survey, which helped popularise the staged diagram, was not retrieved in this sweep.

Sources: Kallokain blog (2023) (↗); arXiv (2023) (↗); arXiv (2025) (↗); Leanpub (Laurent Bossavit) (2013) (↗); Leanpub (2013) (↗)

Royce did not invent the thing he warned against

Boehm's 1987 remark is the best contemporary hedge. He described Royce's paper as the one generally thought to define the stagewise model. In the same breath he credited Benington and Hosier with earlier approximations and noted that Royce already treated prototyping as essential. The paper was reprinted in the ICSE 9 proceedings in March 1987 (pp. 328 to 338), which explains why so many citations carry a 1987 date.

The bawiki essay and a 2013 Hacker News thread push from the other side. Single-pass development predated Royce, and "waterfall" simply names what Figure 2 depicts. Both "Royce invented waterfall" and "Royce was the first to oppose it" lose ground here.

Sources: Wikipedia (2026) (↗); BibSonomy (1970) (↗); bawiki.com (2015) (↗); Hacker News (2013) (↗)

The standard allowed incremental builds; the auditors read sequential phases

DoD-STD-2167 (4 June 1985) listed sequential phases. Larman and Basili report that the DoD suffered significant failures in the late 1980s under the strict, document-driven, single-pass model it required, and that the principal author of 2167 later told them he regretted it. 2167A (29 February 1988) was meant to be life-cycle neutral. Its formal reviews and audits still read as a waterfall preference. MIL-STD-498 replaced it in late 1994 and offered grand-design, incremental and evolutionary strategies, accompanied by Maj. George Newberry's article "Removing the Waterfall Bias".

The awkward detail is that 2167 itself, in section 4.1.2, already permitted overlapping iterations and an "incremental build". Mosaic Projects therefore calls 2167 an iterative standard, while ScienceDirect calls it a "grievous misreading" of Royce. The gap between the text of a standard and how contractors applied it is the real story. MIL-STD-1679, the Navy weapon-system standard that preceded 2167, did not surface in any lane's retrieved sources.

Sources: IEEE Computer (2003) (↗); web3us.com (Larman and Basili text) (2003) (↗); Wikipedia (2026) (↗); Wikipedia (2026) (↗); Mosaic Project Services (2024) (↗); ScienceDirect Topics (n.d.) (↗)

Gantt charts do not read footnotes

Boehm's 1988 spiral paper states that the stagewise model became the basis for most acquisition standards. The incentives explain why.

• Fixed-price contracts need payment milestones that map onto discrete phase deliverables.
• Phase-gate approvals give auditors and lawyers a document trail.
• Gantt charts render sequence cleanly and iteration poorly.
• The civil-engineering metaphor made "design before build" look like professional discipline rather than a guess.

None of these depends on Royce's argument, which is why the model outlived its evidence. The GAO finding reported by Bloomberg Law in 2023 shows the pattern persisting decades after MIL-STD-498.

Sources: IEEE Computer / ACM Digital Library (1988) (↗); Michigan State University course copy (1988) (↗); Bloomberg Law (2023) (↗)

Practice was never as pure as the paperwork

Larman and Basili trace iterative and incremental development to the mid-1950s. They argue that thought leaders in each decade supported it and that many large projects used it successfully. Gerald Weinberg recalled incremental work in 1957 and colleagues who thought waterfalling a huge project "rather stupid". In his account, the waterfall label mainly made practitioners realise they were doing "something else".

This suggests waterfall dominated contracts, textbooks and procurement more than it dominated engineering floors. No systematic survey of 1970s or 1980s practice exists to test the claim.

Sources: IEEE Computer (ACM Digital Library) (2003) (↗); Weinberg blog (2011) (↗); XP 2008 (University of Southampton ePrints) (2008) (↗)

The son's testimony is generous and second-hand

Walker Royce told Larman and Basili that his father "was always a proponent of iterative, incremental, evolutionary development". He then described TRW's CCPDS-R project in 60 pages: six timeboxed iterations of roughly six months each, a pattern consistent with the later Rational Unified Process. His own 1998 textbook calls the 1970 paper the most quoted source of the waterfall model.

This is a family witness relayed through a co-author's book excerpt. It is not archival evidence, and Winston Royce died in 1995 without, as far as the sweep found, publishing a correction of his own.

Sources: web3us.com (Larman and Basili text) (2003) (↗); Flylib (book excerpt) (2004) (↗); Addison-Wesley (via dokumen.pub) (1998) (↗); Wikipedia (2026) (↗)

Spiral is the heir; agile is a distant cousin

Royce's foresight is real but specific. Boehm's spiral paper explicitly credits the 1970 model's adjacent-stage feedback and "build it twice" step, then generalises them into risk-driven process selection with waterfall as a special case. Larman and Basili find "hints of iterative development, feedback, and adaptation" in Royce, but say the paper is clearly not iterative and incremental development in the classic sense.

Three lines of descent are traceable:

• Early risk discovery runs from the pilot build through spiral risk analysis to RUP's elaboration phase.
• Test planning as a first-class activity predates modern test-first practice.
• Customer involvement is a precursor, though Royce meant formal reviews rather than XP's on-site customer.

Genuinely new in the agile wave are short fixed timeboxes, working software as the primary progress measure, welcoming requirement change late, self-organising teams and lightweight documentation. The last one runs directly against Royce, who wanted documentation heavy enough to manage by.

Sources: IEEE Computer / ACM Digital Library (1988) (↗); Wikipedia (2025) (↗); IEEE Computer (2003) (↗)

Evidence & Data

This history has few numbers, so its dates carry most of the evidential weight, and secondary sources often get them wrong. The paper ran to pages 1 to 9 of the August 1970 WESCON proceedings and was reprinted at pages 328 to 338 of the ICSE 9 proceedings in March 1987. MIT Press anthologised it again in 2021. Royce based it on nine years of spacecraft software work.

Sources: BibSonomy (1970) (↗); ResearchGate / MIT Press (2021) (↗); Hebrew University of Jerusalem course materials (1970) (↗)

The procurement chronology, with the conflicts the sweep surfaced:

Item	Best-supported date	Conflicting claims in sources
DoD-STD-2167	4 June 1985	Mosaic Projects reads it as iterative
DoD-STD-2167A	29 February 1988	Smartsheet gives 1985
Brooks-chaired Defense Science Board report	1987	Cited only second-hand, via an arXiv thesis
Defense Science Board report urging iteration	June 1994	Larman and Basili only
MIL-STD-498	Late 1994	November or December by source; Mosaic Projects gives 1996

Sources: Wikipedia (2026) (↗); Wikipedia (2026) (↗); Mosaic Project Services (2024) (↗); web3us.com (Larman and Basili text) (2003) (↗)

Larman and Basili's 2003 IEEE Computer article has roughly 2,500 Google Scholar citations. That makes it the most influential single source on this question and a single point of failure for the procurement narrative. The CCPDS-R figures (six iterations of about six months each, written up in 60 pages) come from Walker Royce via that article. Error rates in popular sources are high: a 2007 Serena Software agile primer says Royce coined "waterfall", which the paper itself disproves.

Sources: Google Scholar (2026) (↗); web3us.com (Larman and Basili text) (2003) (↗); Serena Software (University of Pittsburgh copy) (2007) (↗)

Signals & Tensions

The correction is becoming the next myth. Bossavit argues that both camps are wrong: agile revisionists who cast Royce as an iterative advocate, and defenders who say he was simply misrepresented. He locates the split in Boehm's 1987 claim that prototyping is "compatible with" waterfall, which divided the model into a "stupid" half and an "enlightened" half. A 2026 ProjectManagement.com commenter makes the same complaint more bluntly: the misreading explanation is too tidy. "Nobody read past page 2" remains plausible but unproven.

Sources: Leanpub (Laurent Bossavit) (2013) (↗); InfoQ (2011) (↗); Thomas Kappler's blog (2013) (↗)

Agile advocates overclaim Royce in a predictable way. They quote "risky and invites failure" and stop. They skip "fundamentally sound", the documentation demands and the adjacent-phase limit. Bertrand Meyer's critique of agile's self-history is the obvious corrective, but the sweep retrieved only a pointer to it, not its argument.

Sources: martinfowler.com (2018) (↗); ETH Zurich (author site) (n.d.) (↗)

Waterfall persists as a working option, not just a historical villain. Fowler notes enterprise practitioners still describe it as viable, Watermark Learning asks the same question in 2024, and "Wagile" hybrids have their own literature. For regulated or fixed-scope work, the Royce of 1970 is arguably closer to current practice than the Royce of agile slide decks.

Sources: martinfowler.com (2018) (↗); Watermark Learning (2024) (↗); en.wikipedia.org (n.d.) (↗)

The procurement legacy is underreported. Financial-press searches of the FT, WSJ, Economist and Bloomberg found no intellectual history at all. Waterfall appears there only as a cost and acquisition problem, through the GAO finding and the Hegseth memo. The business consequence outlived the scholarly debate by thirty years.

Sources: Bloomberg Law (2023) (↗); investing.com (n.d.) (↗)

Agile's founding myth matters less as agile itself is questioned. A 2026 essay titled "Saying Goodbye to Agile" is a weak signal, not a trend. Still, an origin story built on a villain loses force once the hero is also under review.

Sources: Lewis Campbell's homepage (2026) (↗)

Open Questions

1. Did the 1987 Brooks report change DoD policy? The claim that it prompted the DoD to permit iteration rests on an arXiv thesis citing Larman and Basili. No lane checked the report itself in the DTIC archive.
2. Who drafted 2167, and what did they cite? The principal author's regret is reported second-hand. Drafting records would show whether Royce's paper, Boehm's 1976 survey or neither actually shaped the phases.
3. Where did the textbook diagram come from? Bossavit's "Boehm's waterfall" finding implies the canonical image may descend from Boehm's 1976 survey rather than Royce directly. That needs a page-level comparison.
4. What did MIL-STD-1679 contribute? The Navy predecessor standard is absent from every lane's evidence, despite being part of the brief.
5. Did Winston Royce ever respond publicly? He lived until 1995, through 2167, 2167A and the spiral paper. The sweep found only his son's testimony.
6. How sequential was practice in the 1970s and 1980s? Larman and Basili and Weinberg offer case evidence and recollection, not a survey. Whether waterfall dominated practice or mainly paperwork remains an inference.
7. Does the Software Acquisition Pathway change behaviour? The 2025 memo changes the preferred route. Whether programme offices stop phase-gating in practice is the same text-versus-audit gap that 2167 exposed.

For a technology leader, the transferable lesson concerns institutions more than Royce. A standard can permit incremental builds in section 4.1.2 and still produce waterfall, because the payment schedule, the audit trail and the Gantt chart all read the first diagram and none of them read the caveats.