Startup marketing has an evidence problem.
The advice most founders read is not false. It is worse than false. It is incomplete in a way that is hard to see from inside the market. The failures have been removed before the lesson is written.
Aviation faced a version of this problem fifty years ago. It did not solve it by asking pilots to be more candid. It built a system that made candor less costly.
Six weeks
In October 1974, a United Airlines crew flying the approach to runway 12 at Dulles interpreted its clearance as permission to descend. The aircraft came close to the terrain near Mount Weather.
United investigated the incident internally. It found that some of its pilots were routinely descending below minimum altitudes when cleared early for an approach. United issued a notice to its own pilots, telling them to check minimum altitudes for their route first.
United did not tell the FAA.
On 1 December 1974, TWA Flight 514 flew the same approach, with the same misunderstanding, and struck Mount Weather. All 92 people aboard were killed. The FAA record and later summaries of the accident describe the approach-clearance confusion and the earlier United incident as central to what followed, see the FAA Lessons Learned record, the ALPA account of the legacy of TWA Flight 514, and the accident entry at Aviation Safety Network.
The information that could have mattered existed. It was current. It was specific. It was inside one company.
That is the part founders should notice.
What they built
After the accident, the FAA proposed a national reporting program. NASA accepted the operating role. The Aviation Safety Reporting System began operating on 15 April 1976.
The program is described in three words in NASA's own briefing, voluntary, confidential, non-punitive. Those words are not slogans. They are the architecture.
By the end of 2025, the ASRS had received 2,321,050 reports. In 2025 alone, it received 121,128 reports, about 10,094 per month and 485 every working day, according to the NASA ASRS Program Briefing.
These are not only accidents. They are near-misses, confusions, procedure gaps, communication failures, and moments where someone almost got it wrong.
That distinction matters. Accident investigations are necessary, but they arrive late. Near-miss reporting catches the system while it is still capable of correction.
The lesson was not that people needed to care more. The lesson was that important information dies when the cost of sharing it is too high.
Three design choices
The ASRS worked because it addressed three different reasons people stay quiet.
Each design choice solved a specific behavioral problem. Remove one, and the system becomes theater.
Voluntary reporting
Voluntary reporting defeats the idea that an organization should only learn when an event becomes unrecoverable.
If reporting depends on formal investigation, the system sees only the visible wreckage. It misses the confusion that almost became wreckage, and that confusion is often the most useful signal.
In a young company, this is the difference between documenting a failed quarter and documenting the moment a team first realized the launch was not behaving as expected. The second record is less polished, but more valuable.
Confidential reporting
Confidentiality defeats the fear of being identified.
The ASRS process is physical and procedural, not merely cultural. NASA removes identifying information, returns the identification strip to the reporter as proof of filing, and destroys the original report, both physical and electronic, according to the ASRS briefing.
That detail matters because anonymity is not the same as trust. A founder saying, “we want candor,” is not a system. A process that separates the reporter from the lesson is closer to one.
The insight is simple. People will report more of what happened when the report does not become a permanent label attached to them.
Non punitive reporting
Non-punitive reporting defeats the fear of consequence.
Here again, the ASRS is not relying on good intentions. FAA Advisory Circular 00-46F, as summarized in the NASA briefing, provides waiver of civil penalty and certificate suspension when specific conditions are met. The violation must be inadvertent and not deliberate. It must not be criminal and must not involve an accident. The reporter must have no prior violation finding in the previous five years. The report must be filed within ten days.
The ten-day window is underrated.
It forces the report to be written while the reporter can still remember being confused. Not after a month of explanation. Not after the board deck has turned uncertainty into a clean narrative. Not after everyone has agreed which story is safest.
Most post-mortems are written too late. By then, the facts have been sanded into a lesson.
Publication mattered
Collection was only half the system. The other half was publication.
Since 1979, NASA has published CALLBACK, a monthly lessons-learned newsletter built from de-identified ASRS report excerpts. By the end of the 2025 briefing period, CALLBACK had published 551 issues. In 2025, it had 33,765 email subscribers and 1,799,748 views, according to the ASRS Program Briefing.
NASA also reports 8,069 safety alert messages issued to organizations with authority to fix problems, and 416,361 public database queries since July 2006.
The structure is important. The reports do not sit in a vault. They become readable operating knowledge.
That is the move startup marketing usually skips. Founders either say nothing, or they publish a retrospective so polished that the actual failure is no longer visible.
A useful failure record is not a confession. It is a de-identified operating note that lets someone else recognize the pattern sooner.
The model travelled
The ASRS is not an aviation curiosity.
NASA's briefing lists other fields that copied or adapted the model, including rail, medicine, security, firefighting, public utilities, maritime, and structural engineering. It also references international confidential aviation reporting systems connected through ICASS.
So the relevant question is not whether the model can transfer beyond aviation. It already has.
The question is what portion of it a small software company can copy without pretending it is NASA.
Startup evidence gap
Startup marketing has the public investigation layer. It does not have the confidential near-miss layer.
Founders can read how a company got its first thousand users. They can study the launch thread, the pricing pivot, the cold email sequence, the founder-led LinkedIn motion, the community build, the Product Hunt spike, the SEO compounding curve.
What they usually cannot read is the four hundred companies that ran the same play into the same mountain.
That absence changes decision quality.
A founder does not copy a tactic from the full market. They copy it from the visible market. The visible market is biased toward companies that survived long enough, grew enough, and benefited enough from the tactic to make the story worth publishing.
Then AI systems ingest that published record and return it as confident guidance. The missing failure data does not become less important. It becomes laundered at scale.
This is why much marketing advice feels true and still fails in use. The advice may describe what happened. It just does not describe how often the same move failed under adjacent conditions.
Why advice fails?
Marketing advice fails founders because it is usually detached from base rates.
A case study can tell you that one technical founder grew through founder-led content. It rarely tells you how many founders posted for six months and produced no qualified pipeline. A teardown can show you a launch that spiked. It rarely tells you which launches created attention without trust, trials without activation, or meetings without urgency.
That does not make case studies useless. It makes them incomplete.
The better question is not, “Did this tactic work for someone?” It is, “What conditions had to be true for it to work, and what happened when similar teams misread those conditions?”
Startup content rarely answers the second question because the people with the answer have incentives to stay quiet.
Why nobody shares?
There is no ASRS for startup marketing because there is no immunity.
A founder who publishes what did not work is speaking in front of investors, candidates, customers, competitors, and current employees, all at once. They publish under their own name. The company is identifiable. The numbers are sensitive. The downside is immediate and personal.
So the public version becomes safer.
The founder removes the damaging figures. The failure is reframed as a growth lesson. The timing shifts until the company has recovered enough for the story to signal maturity instead of weakness.
This is not moral failure. It is incentive design.
Aviation did not fix the reporting problem by asking pilots to be braver. It reduced the penalty for telling the truth quickly.
What founders copy
A two-person company cannot build a national confidential reporting institution. It should not pretend otherwise.
But it can copy four narrow mechanisms.
Create a filing window. When a campaign, channel, launch, sales motion, or positioning test underperforms, write it up within ten days. Keep it short. What did we expect? What did we do? What happened? What did we misread?
Make no blame written. The person who files the report is never blamed for filing it. If that promise is not credible inside the company, do not start the system. It will produce either silence or self-protection.
De-identify before publishing. Publish the pattern, not the person. Remove client names, employee names, deal names, and details that create avoidable exposure. The goal is learning, not spectacle.
Publish a callback. Create a recurring, short lessons-learned note, monthly or quarterly. Use three or four paragraphs per incident. Describe the expectation, the miss, the signal that was ignored, and the operating change.
The fourth item is the one with strategic value.
Almost any company can publish opinions. Far fewer can publish an accurate, de-identified record of how work actually went wrong. That record is harder to copy because it is made from lived operating data.
A practical format
A small-company report does not need a template complex enough to discourage use.
It can fit on one page.
Date filed, within ten days of the event.
Expectation, what we thought would happen.
Action, what we actually did.
Observed result, what happened, in plain numbers where possible.
Misread, what signal we interpreted incorrectly.
Change, what we will do differently next time.
Publication version, what can be shared after de-identification.
The format should be factual before it is reflective. Reflection can come later. The first job is to preserve the confusion while it is still visible.
What not copy
Do not turn the system into a performance review input.
Do not use it to prove who was right. Do not publish client-sensitive detail. Do not inflate four incidents into a benchmark. Do not call a single anecdote a trend.
Most importantly, do not write the public version first.
If the external story determines the internal record, the system has already failed. The point is to learn before reputational editing begins.
The honest limits
Four reports are not a data set.
A company of six cannot produce statistical significance about a channel, audience, message, or sales motion. It should not pretend that a few internal notes have the weight of market research.
The ASRS has limits too. Its reports are self-selected. NASA treats them as complementary to other safety monitoring, not as a substitute for it. The value is not perfect measurement. The value is early, candid signal from people close to the work.
That is enough.
A small company can produce something the highlight reel structurally cannot, an accurate record of how the thing went wrong while it was going wrong.
What changes?
The immediate benefit is better internal memory.
Founders stop rerunning the same campaign because the previous failure has been converted into folklore. Marketers stop inheriting vague warnings without context. Sales and marketing stop arguing from selective recollection.
The external benefit is trust.
Not because failure content is noble. Because useful, de-identified failure content gives readers something rare, a view of the conditions under which a common tactic did not work.
That is materially more useful than another success story stripped of denominator, timing, and context.
The takeaway
United investigated. United wrote the finding. United warned its own pilots. Each step was competent, and it still was not enough, because the learning stopped at the company line.
Startup marketing has a quieter version of the same structural problem. The market sees the winners. It rarely sees the near-misses. Then founders make channel decisions from a data set with the failures deleted.
Aviation's answer was not more bravery. It was a system, voluntary, confidential, non-punitive, with fast filing and regular publication.
Founders cannot copy the institution. They can copy the mechanism.
Write the report within ten days. Protect the filer. De-identify the lesson. Publish the pattern. Better marketing starts when the record gets closer to what actually happened.