The Roadmap Is a Promise: Osborne, Star Citizen, and the $39,900 Cybertruck

A publicly traded developer of an unfinished video game once explained, in writing, why it was deleting most of its own roadmap. Not because the plans had changed. Because of how the plans were being read. Cloud Imperium Games, the studio behind Star Citizen, had spent years publishing one of the most detailed open-development roadmaps in the industry — feature by feature, quarter by quarter, visible to every backer. In February 2022 they stripped it back to a single quarter, and the reason they gave is the most honest sentence I’ve ever seen a company write about this artifact: players interpret anything on the release view as a promise.

Not “some players.” Not “dated items.” Anything on the release view. A studio with a decade of data on how audiences actually consume roadmaps — more data than any product team will ever have, because their audience was obsessive, paying, and vocal — concluded that the caveat doesn’t exist. The “subject to change” footer, the confidence column, the disclaimer read aloud at the start of the stream: none of it survives contact with a reader who wants the thing on the roadmap. I wrote in the build manual that a date on a slide reads as a promise no matter how many footers you attach. This post is the evidence file: three stories, forty years apart, all documenting the same missing organ. Audiences do not perceive confidence intervals. They perceive commitments.

Osborne, 1983: the story that’s half wrong, and dangerous anyway

The canonical telling goes like this. In early 1983, Osborne Computer — maker of the Osborne 1, one of the first commercially successful portable computers — pre-announced its next-generation machines while the current one was still on shelves. Customers, being rational, stopped buying the machine that had just been declared obsolete by its own manufacturer. Dealers froze orders. Revenue collapsed onto a company with no cushion, and by September 1983 Osborne filed for bankruptcy. Announcement in April, dead by autumn. The “Osborne Effect” entered the vocabulary: pre-announce the future and you kill the present.

Here’s the complication, and it matters: the clean version is partially apocryphal. Later accounts — including from people inside the company — point to co-causes that the legend edits out. Kaypro was eating Osborne’s position with a better-value competitor. Inventory management was a mess, with the company mismanaging stock of the current machine in ways that would have hurt regardless. The announcement didn’t act alone; it landed on a company already stumbling, and the tidy causal story got written afterward because tidy causal stories always do. If you tell the Osborne story as “pre-announcement, therefore death,” you’re repeating folklore.

But here’s what makes this case worth its slot in the trilogy: the debunked version still governs behavior. Ask why Apple announces products days before you can buy them, or why release comms teams treat forward-looking statements as radioactive. The Osborne Effect is cited by people who have never read a primary account, and the fear of it — accurate or not — has shaped four decades of announcement discipline. There’s something almost recursive about it: even the cautionary tale about promises got received as a promise. Whatever the true causal weights were in 1983, the industry concluded that telling customers about the future is a loaded transaction, and it has behaved that way since. The industry is right, even where the story is wrong — and the next two cases show why, with cleaner data.

Star Citizen, 2022: the best-instrumented roadmap experiment ever run

If you wanted to design an experiment to find out whether caveats work — whether an audience can be trained, through repetition and goodwill, to read a roadmap as intentions rather than commitments — you could not do better than Star Citizen. Crowdfunded, so the audience is financially and emotionally invested. Open development, so the roadmap is public and granular. Years of iterations, so the studio could try every mitigation: disclaimers, tentative markers, blog posts explaining what the roadmap does and doesn’t mean. If any audience on earth could learn to apply a discount factor to a roadmap row, it was this one, and no studio ever had a stronger incentive to teach them.

The mitigations failed. Every slipped item on the long-range view generated the anger of a broken promise, not the shrug of a revised forecast, and it kept happening no matter how the rows were labeled. So in February 2022 CIG did the only thing that had a chance of working: they changed the structure. The public release view now shows one quarter — the work actually in flight — and everything beyond it was removed from the promise surface entirely. Not re-labeled. Removed. Their stated reasoning was the line this post opened with: players interpret anything on the release view as a promise.

Read that as a product manager and the finding is stark. A decade of the best-resourced expectation-management effort in software, and the conclusion was that no amount of labeling converts a visible commitment-shaped object into a non-commitment. The only variable that works is what’s structurally on the page. It’s the same conclusion I reached in the sales room by a shorter road: enterprise buyers screenshot, and whatever artifact enters the room resurfaces later with commitments read into it that you never made. CIG just proved it at population scale, against the most sophisticated roadmap-reading audience that will ever exist.

Cybertruck: $39,900 is not a price, it’s a memory

November 2019: Tesla unveils the Cybertruck on stage and announces a starting price of $39,900, with production slated for 2021. The number does exactly what numbers on stages do — it detaches from every qualifier around it and becomes the product’s identity. A sub-$40K electric pickup. Reservations pile up against that number.

November 30, 2023: the first Cybertrucks are delivered. The cheapest version you could actually configure cost $60,990 — 53% above the number on the stage — and the promised entry-level variant had been pushed to 2025. Four years late on the timeline, half again over on the price, and the entry price point demoted from “the product” to “a future variant.”

I’m not interested here in whether Tesla should have known better; four years is a long time, and costs moved for everyone. What interests me is the asymmetry in how the number lived. Inside the company, $39,900 was presumably a target — an aspiration with a wide error bar, contingent on production learning curves nobody controlled. Outside, it was a spec. Every headline about the 2023 delivery was structured as promised X, delivered Y; the stage number had become the unit of measurement for the shipped product. That’s the confidence-interval failure in its purest form. Nobody who put down a reservation recorded “$39,900 ± production reality.” They recorded $39,900, and every dollar above it was experienced as a broken promise rather than a resolved uncertainty. The error bar existed in exactly one place — the heads of the people who made the slide — which is the one place it does no good.

The synthesis: encode the uncertainty in the structure, not the footnote

Three cases, three eras, three audiences — 1983 computer buyers, 2022 game backers, 2023 truck reservation holders — and one mechanism. A roadmap is read by people who want the things on it, and wanting collapses probability into certainty. The reader does not apply your discount factor. The reader cannot: they lack your context, they don’t share your incentives, and a named thing with a time attached is the shape of a commitment. Asking a reader to see “Reporting v2 — Q3 — tentative” and perceive a probability distribution is asking them to do your risk management for you, against their own interests. They never will. Those three audiences had nothing in common except that.

Which means the fix cannot live in the caveat layer. It has to live in the structure — the uncertainty has to be encoded in what the artifact physically shows, so that reading it as a promise yields only promises you can keep. This is the deep reason the formats I’ve argued for across this run of posts are shaped the way they are, and I want to name the connection explicitly, because the case studies are the “why” behind mechanics I previously asserted:

• Now/Next/Later works because the columns are the confidence interval. Now is a promise and is allowed to look like one. Next is sequenced but undated — the structure refuses to generate a date for the reader to screenshot. Later is direction only. The reader can’t over-read precision that isn’t on the page.
• Outcomes-over-features works because problems don’t have delivery dates. “We’re going after reporting workflows” can’t be promised-and-broken the way “Reporting v2, Q3” can. When the item is a problem statement, the only commitment transmitted is the one you actually mean: we think this matters and we’re working on it.
• The horizon cut works because deletion is the only caveat that sticks. CIG’s move — show one quarter, remove the rest — is the industrial-strength version. What isn’t on the release view can’t be interpreted as a promise, and nothing else about the release view was true of every row on it.

The same logic runs through the audience-specific views: investors get trajectory and thesis rather than a dated feature list, and marketing announces at the moment of certainty rather than the moment of excitement — the Osborne fear, operationalized, forty years on. In every room the discipline is identical: decide which promises you’re willing to have read into the artifact, then make sure the artifact contains only those. Everything else is a footnote, and the footnote has a documented survival rate of zero.

Put it to work

1. Run the screenshot test on your current external roadmap. Take the artifact you last showed a customer, investor, or conference audience, and reread it with every caveat, footer, and verbal disclaimer deleted — because that’s what the audience retained. Every named-feature-plus-timeframe pair that survives is a promise you’ve made. If you wouldn’t sign that list, the artifact is over-promising structurally, and no disclaimer will fix it.
2. Move the uncertainty from labels into structure. Wherever your roadmap currently uses a confidence tag, a “tentative” marker, or an asterisk to soften a dated item, restructure instead: undate it into Next, generalize it into an outcome, or cut it below the horizon entirely. The rule from the three cases: if being held to it would hurt, it doesn’t go on the visible surface — labeled or not.
3. Log your “promised vs. delivered” gaps the way the press logged Cybertruck’s. For your last three shipped roadmap items, write down what the roadmap said when each first became visible externally, and what actually shipped, when. That delta — not your intentions — is your promise-keeping record, and it’s what your audience uses to price your next roadmap. If the gaps are large, shortening your visible horizon isn’t cowardice; it’s honest pricing.

Further reading

• MassivelyOP’s February 2022 report on Cloud Imperium’s roadmap restructuring — the primary public account of the change, including the studio’s own explanation that players interpret anything on the release view as a promise.
• TechCrunch, November 30, 2023, on Cybertruck delivery pricing — the $60,990 starting figure and the entry-level variant’s slip to 2025, against the $39,900 announced in 2019.
• Bruce McCarthy et al., Product Roadmaps Relaunched — the fullest treatment of roadmaps as communication artifacts with audience-specific views, which is the constructive answer to everything above.
• Janna Bastow’s ProdPad writing on Now/Next/Later — the format whose structure encodes the uncertainty these three stories show audiences will never infer on their own.