Everything that has happened on a covered name since its note published. Notes themselves are never rewritten; new information is published here and dated, and the note it belongs to carries the same entry.
Monitor New information. The thesis in the note still stands.
Revision A specific claim in the note is corrected or superseded. The thesis stands.
Position change What we hold in the name has changed. Opened, added to, reduced or closed, with the figures.
Stance change The framework view has moved, with the reasoning attached.
We added to Kodiak on Friday’s twelve-month low. Average cost is now $3.77.
What we did. We added to the KDK position on Friday 18 September, into the decline. Blended average cost moves from $3.80 to $3.77. The entry date is unchanged at 31 August 2026 at $3.83. The disclosure block above this log now carries the 18 September as-of date.
The add was small, and the arithmetic says so without our having to characterize it. A three-cent move in a blended average, off a starting average of $3.80 with a fill somewhere in Friday’s range, implies an addition of roughly a tenth of the existing position. We are spelling that out because there is a real difference between adding and doubling, and because at a Speculative stance our own framework makes position sizing, not conviction, the operative decision. Adding a tenth is a sizing decision. It is not an upgrade, and nothing below this line changes.
What the stock did. Kodiak set a new twelve-month low of $3.45 on Friday and traded around $3.48, putting market capitalization near $699 million against the 200.6 million shares in this note’s key data. Volume was roughly 886,000 shares, essentially in line with the 872,000 average. Two sessions earlier, on 16 September, the stock fell 6.2% to $3.575 on 527,000 shares, about 40% below average. Neither day looks like institutional distribution. It looks like drift on ordinary volume, which is what a micro cap does when nobody is defending the bid.
We could not find a company-specific catalyst, and we looked for one. No filing, no announcement, no lost contract, no equity offering and no change to guidance in that window. What is visible instead is a ratings book that has split along methodological lines. Wall Street Zen and Zacks now both carry the stock at Strong Sell, and both run quantitative processes that key off reported earnings, which for Kodiak means a $0.50 second-quarter loss against a $0.17 estimate on $3.5 million of revenue. Against them sits a fundamental book that is still mostly positive: Needham at Buy with an $8.00 target, Morgan Stanley at Equalweight with $9.00, Oppenheimer at $9.00, Chardan at $12.00, Craig-Hallum at $12.00. Consensus target is $10.89 on a $3.48 stock.
A consensus target at three times the price is not an endorsement. It is a sign that the published models and the tape have stopped speaking to each other. On a company with a going-concern flag and revenue measured in single-digit millions, that gap is not going to close because the models are patient. It closes when the operating milestone lands or when the financing arrives, and one of those is a good outcome. We do not read a twelve-month low as validation of the bear case. We are also not going to call it noise.
What did change, and it is the reason we added. On 8 September Kodiak reported its Autonomy Readiness Measure at 93% for August, up from 91% in July, and reaffirmed that driverless long-haul commercial service launches by the end of 2026. Management described the remaining work as final engineering verification and validation, and said it expects to complete the safety case before year end. As of the close of the second quarter, 35 driverless trucks were operating commercially in the Permian Basin with no human in the cab, and Atlas Energy has committed to taking the Kodiak-powered fleet to 100 trucks by 2027. The operating record moved forward in the same month the equity made a twelve-month low. That divergence is the whole reason to add, and it is equally the reason the position stays sized as an option rather than a holding.
What did not change, and we are not hiding from it. Roughly fourteen months of runway. Negative free cash flow of $147 million on the last read, corroborated by Morgan Stanley’s own initiation. Dilution remains our base case rather than a risk factor. The June resale registration put 30.8 million shares in front of the market, around 15% of the count, half of it warrant shares struck at $6.00. At $3.48 that strike sits 72% above the stock, so that specific overhang is out of the money and correspondingly less pressing today, but the registered shares beside it are not, and a strike that far away is a statement about where the last institutional money came in. And the question this note was built around, who writes the policy that replaces commercial auto liability on a driverless truck, remains unanswered by every analyst publishing on the name.
The discipline we are holding ourselves to. Averaging down into a going-concern balance sheet is the most reliable way there is to convert a speculative position into a real loss, and we are aware that this is exactly what it looks like from outside. So: the add was small, the stance is unchanged at Speculative, and the next marker is dated and binary. Safety observer removal and driverless long-haul launch by 31 December 2026. If that date slips, the reason to own this changes, and it will be published here as its own entry rather than absorbed quietly into the average.
Thesis unchanged. Framework view remains Speculative. Average cost $3.77 as of 18 September 2026.
Answers agenda item 05, on the long-haul milestone and the next financing from the note.
We are out of the IBM calls. The exit rule did the work, and the thesis was never tested.
We are flat IBM. No shares, no options, no other exposure. This note exists to describe a position, so here is precisely how it ended.
What we did. We sold the majority of the 16 October $260 calls on Tuesday 15 September, against the $4.60 per contract paid on 9 September. We left runners on with a stop at the entry price, and that stop was hit the following session. The trade closed in two parts and never ran to expiry.
We are not publishing what the trade returned, and that is a standing rule from today rather than a decision about this one. The dates are above, the premium we paid is on the disclosure block, and the closing prices are below, so anyone who wants the arithmetic has everything they need to do it. What we will not do is hand you a percentage. We do not publish position sizes, we have no aggregate record to set it against, and this is the first position we have closed since the site started disclosing at all, which makes a single figure the most misleading shape the information could take. A number on one trade is not disclosure. It is marketing. The policy is now written into how we trade what we cover, and it will apply to the losses on the same terms.
What the underlying did. IBM closed at $239.94 on the day we opened. It ran to $249.09 on Tuesday 15 September, up 2.38% on a day the Nasdaq Composite fell 0.6%, trading as high as $251.37 intraday on about 6.8 million shares. That is the session we sold into. On Wednesday 16 September it handed the move back and more, closing at $237.49, down 4.38%, after management trimmed full-year constant-currency revenue growth guidance to 4–5% from more than 5%, announced alongside a $1 billion CHIPS Act research award to its Anderon subsidiary. That is the session the runners stopped out. It kept going from there: $237.75 on Thursday, then $229.55 on Friday 18 September, down another 3.2%, with the sell side visibly splitting on the name. HSBC now carries a Reduce at $191, against Oppenheimer at $350 and Morgan Stanley at $293. That is a $159 spread on one company.
The honest accounting: whatever the position did, it did not do it for the reason we wrote down. Section 05 listed the four things this trade rested on. The Accenture read-across on 1 October, which we called the clearest in-window catalyst, never arrived while we held the position. A target raise of the kind we speculated about has since landed, with Oppenheimer moving to $350 from $320, and the stock is lower now than when we wrote that section. IBM never traded within $8 of our $260 strike. What the calls did between 9 and 16 September came from delta on a nine-dollar move in the underlying, not from the argument in this note being validated. The exit rule produced the outcome, not the analysis. That distinction is the only part of this worth reporting, and it is the part a percentage would have buried.
The risk we flagged loudest was never tested. Section 04 exists because these contracts expired on 16 October, five days before IBM reports on 21 October, and we published that as the central flaw in our own position. It never came due, because we were out five weeks early. Whether the calendar would have cost us is now unanswerable, and we are not claiming credit for avoiding something we never reached. There is one detail worth recording: the only scheduled event the note put inside our window for that week was the 15 to 16 September FOMC meeting, and the position did in fact resolve across exactly those two sessions. What closed it was company-specific, not the Fed. We had the dates right and the mechanism wrong.
Where the position would stand today. IBM closed Friday at $229.55. That is $30.45 below the $260 strike and $35.05 below the $264.60 breakeven, with under four weeks left. The underlying would need to rise 13.3% to reach the strike and 15.3% to return anything at expiry, inside a window that still ends before the print. Held, this would now be running the losing path described in Section 06. That is not hindsight vindication of the exit. It is the same arithmetic we published on day one, applied to a worse starting price.
What this does to the note. Nothing below this line changes, as always. A trade note carries no framework stance, so there is no stance to move: we have never published a view on whether to own IBM shares, and closing a call position does not create one. We may re-enter. If we do, it publishes here as a dated entry before it is discussed anywhere else on the site.
One result is not a record. A single closed trade tells a reader nothing about a process, and we would be writing this same paragraph if this one had gone the other way. We publish the exits for the same reason we publish the entries, which is so there is no version of this site where only the good ones are visible.
Position closed. A trade note carries no framework stance on IBM, and closing the trade does not create one.
Three corrections to this note: a position we did not disclose, a figure we got wrong, and a business we missed.
We are correcting three things on this note in one entry, because they were found in one pass and they publish on one day. Two are our own errors of omission and one is a figure we reported wrongly. The note body below is unchanged, as it always will be.
One. We hold this name and the note did not say so. We are long KDK. The position was opened on 31 August 2026 at $3.83, five days before this note published on 5 September, and has been added to since at an average cost of $3.80. A position disclosure now sits under the framework stance at the top of this note, giving the direction, the entry date and our average cost. It should have been there on day one.
Alongside it we have published how we trade what we cover, which sets out that we hold only some of the names we write about, that coverage is not a signal of ownership, and that we will disclose any material change to a position including an exit. The generic line in this note's own disclosures footer predates the block above and is superseded by it. We have left the footer as published, because we do not edit a note after the fact.
Two. The revenue split in this note is wrong. The note states that the Q2 2026 revenue split was Driver as a Service 56%, ground autonomy 34%, freight 10%. Kodiak publishes no such breakdown. Those three percentages are the customer concentration disclosure in the Form 10-Q, where unnamed Customer B is 56% of the quarter and Customer E is 34%, and we presented a customer table as a service-line table. There is no revenue disaggregation by service line anywhere in the filing. On the Q2 earnings call management attributed the new revenue directly: slightly over $1 million from a one-time autonomous trucking demonstration with DriveOhio, an initiative of the Ohio Department of Transportation. The 34% is a state transport agency demonstration that management called one-time. It is not a business line and it is not defense.
Three. We did not cover the defense business at all, and it explains the revenue history we quoted. Defense is not something Kodiak is building toward. It is a business that already peaked and collapsed.
US military revenue against total, as disclosed
Year
Total revenue
US Army revenue
Share
FY2024
$14.93M
$13.3M
89%
FY2025
$3.80M
$1.0M
26%
H1 2026
$5.33M
not disclosed
below the 10% threshold
Army figures are narrative disclosures in the FY2025 Form 10-K. Cumulative military revenue to date is approximately $30M, against a $49.9M ceiling on a contract that no longer exists.
This changes how the growth rate should be read, and it is why the correction leaves the thesis firmer. We wrote that revenue rose 595% year over year. That is arithmetically true and it is measured off a trough that the defense rolloff created. Total revenue fell 75% from FY2024 to FY2025 as the Army work wound down, and neither the full-year release nor the Q2 release explains the decline. A reader looking at 595% growth without that context is looking at recovery from a collapse and calling it expansion. Against a going-concern flag and roughly fourteen months of runway, that makes the dilution case more acute.
What the defense position actually is. The money came from a single Defense Innovation Unit agreement signed in October 2022 for the Army Robotic Combat Vehicle program, announced at a $49.9M ceiling, of which Kodiak received about $30M. The Army halted RCV development in May 2025 during a budget drill, after the competition had concluded and Textron's Ripsaw had won. In August 2025 the Army awarded three autonomy agreements worth roughly $15.5M for Infantry Squad Vehicles to Forterra, Overland AI and Scout AI. Kodiak was not among them.
What remains is early and mostly unpriced: a US Marine Corps development contract announced 11 February 2026 for autonomy on the ROGUE-Fires missile carrier, with no disclosed value; a strategic collaboration with General Dynamics Land Systems announced 7 May 2026, with no value and no duration; and a Defense Innovation Unit prize challenge worth up to $500,000. No production award, no program of record, no backlog, no guidance, and no separately reported segment. We also found no mention of ITAR or security clearances in any Kodiak filing, so there is no classified moat here, which cuts both ways: no compliance drag, and no barrier protecting the position either.
The organizational investment is real and larger than the revenue. Kodiak added a defense advisory council including the retired lieutenant general who ran the Army's Next-Generation Combat Vehicles effort, and in February 2026 hired a defense vertical lead who previously ran a $157M defense business. That is a serious capture build. It is intent, and it is not yet dollars.
The corrected position. Defense at Kodiak is option value on a franchise that ran down once already, supported by credible hires and a tier-one integrator relationship, with almost nothing booked since. Where the sell side treats government and defense as a diversification adjective, the more accurate reading is that the street is crediting diversification from a business that has shrunk roughly 92% from its 2024 peak. Our framework view stays Speculative, and the reasons for it are firmer than when we published.
Thesis unchanged, and firmer for the work. Framework view remains Speculative.
Answers agenda item on revenue quality and the going-concern case from the note.
Seven analysts now cover Kodiak. None of them price the bottleneck.
Two initiations landed inside three days. Needham started at Buy on 9 September with an $8.00 target, built on 5 times its fiscal 2030 adjusted EBITDA estimate discounted back, and framed removal of the safety observer by year end as the move from technology validation to business scalability. Morgan Stanley started at Equalweight on 11 September with a $9.00 target, citing OEM relationships, lower cash burn and diversification across on-highway, industrial off-highway and government work, against the company sitting behind Aurora in on-highway commercialization and running negative free cash flow of $147M.
Three things in that are worth holding onto.
Morgan Stanley is the first firm not to recommend the stock. Oppenheimer, Citi, Northland, Chardan and Craig-Hallum had all initiated at Buy between June and late August. The first bulge-bracket bank to cover the name declined to, and it did so while carrying a $9.00 target that sits above both Needham's $8.00 Buy and Citi's $8.50 Buy. Ratings and targets across the street now disagree with each other: the most cautious rating on the name implies more upside than two of the bullish ones. That is what a book with no shared framework looks like, and it is consistent with treating the equity as an option rather than a forecast.
Morgan Stanley corroborates the balance-sheet flag. Negative free cash flow of $147M against the $151.1M of cash and securities in our key data, and the roughly fourteen month runway we published, are the same picture from two directions. Dilution stays our base case.
Nobody is pricing the liability transfer. Seven analysts now publish on Kodiak. On the public record, not one of them engages the question our note was built around: whether a driverless truck can be insured at all, and who writes the policy that replaces commercial auto liability. The street is modeling freight economics, utilization and EBITDA multiples. The bottleneck we identified sits outside every published model on the name. We take that as the gap the note exists in, not as a reason to doubt it.
Needham's catalyst is the more useful near-term marker. Safety observer removal by year end is dated, binary and observable, and it goes on our agenda as the next thing to watch.
Answers agenda item on sell-side coverage and the insurance question from the note.
Updates are append-only. An update that turns out to be wrong is answered by a later one, never by editing the earlier entry. Figures carry the date they were reported.
