Fit

Fit

Outside the framework's universe (U2 not met); contested: X1, X2, X3, X4, S1, P5

At $17.29 the equity is worth $8.53 billion, 14.7% below the framework's $10 billion line; U2 not met resolves the tab before any pillar is weighed. Confidence is low on the tally's stated basis — name-mask divergence or load-bearing probability divergence above 0.20 — and the mask probe did differ, on X1 through X4. No exclusion was hit, the watchlist-only overlay is not set, and six criteria are recorded contested.

The confidence basis matters as much as the tier. Two model families sat on the jury and agreed on every pillar verdict except one; what pushed the tier to low was the name-masked probe returning a different label from the unmasked jury on the four exclusion tests, which sets the prior_driven_risk flag. The tally's own arithmetic gap between masked and unmasked probabilities is 0.02.

Universe and exclusions

Source: derived — 493,551,000 shares (FY2025 income statement basis) at the $17.29 close of 24 July 2026, per fit_features.market_cap; the $10 billion line is the framework's own.

U2 — Scale: not met. This is not a valuation judgment and not a quality judgment. It is a size test, and the company is below it: 493,551,000 shares at $17.29 computes to $8,533,496,790 against a $10,000,000,000 line, a shortfall of $1.467 billion, or 14.7%. The share price that would clear the line is $20.26. The counter-fact in the same breath, and it is a real one: the miss is a price event of recent vintage, not a shrinking business. The last close at or above $20.26 was 4 June 2026, at $21.03 — seven weeks before the reference date — and over the fiscal year just reported revenue grew 18.5% to $2,896.3 million, operating cash flow grew 34% to $992.7 million and gross spend grew 11% to $13,394.7 million [1]. All three jury seats that voted on U2 returned not met, and the masked seat agreed. The framework's rule is a level, measured on the date; the tally applies it as U2 not_met -> out_of_universe.

U1 — Listing: met. Class A common stock has traded on the Nasdaq Global Market under TTD since 21 September 2016 [2]; the issuer is Nevada-incorporated and headquartered in Ventura, California, is not an ADR and is not a Chinese issuer. Three seats, unanimous. The one structural caveat sits outside the test: unlisted Class B stock carries ten votes per share [3], which bears on control rather than listing venue.

The exclusion screen is clean of hits. The tally records exclusion_hits: []. Every seat that voted found the auto-manufacturer test (X1), the promotional-CEO test (X2), the structural-decline test (X3) and the consensus-saturated-story test (X4) unhit, and the China sensitivity flag (S1) unraised — but each of those five is recorded contested because the seats returned different labels for that finding, and the aggregate carries that as a split. The evidence behind each:

• X1 — Auto OEM. One operating segment, the advertising technology platform; no manufacturing operations disclosed; property, equipment and lease assets of $738.9 million are offices and hosting infrastructure [4]. Automotive appears in the record only as an advertiser vertical.
• X2 — Promotion pattern. The test needs both prongs. The ownership prong fails outright: the founder-CEO holds 11.2% of the equity, about $926 million at $17.29, and bought 6,000,000 shares in the open market for $148.1 million on 2–4 March 2026 at an average of $24.68. Against that, on the same page of the record: those purchases are 29.9% underwater, the Class B structure gives 49.7% of the vote on 11.2% of the economics, four directors resigned in five weeks in March–April 2026, the chief financial officer's chair has held four occupants in eighteen months, and a consolidated securities class action carries a Section 20A insider-trading count against the CEO, then-CFO and CSO. Full treatment in Self-Help.
• X3 — Structural decline. Revenue rose in each of the ten reported fiscal years, from $202.9 million in FY2016 to $2,896.3 million in FY2025; consecutive_decline_years is 0 and the three-year high-single-digit decline flag is false. The counter-fact is deceleration rather than decline: gross-spend growth fell from 25.3% to 11.2%, Q1 FY2026 revenue grew 11.8%, the Q2 FY2026 guide of at least $750 million implies 8.1% at its floor [5], and Google Network revenue — the closest recurring measure of third-party publisher advertising — fell in each of 2023, 2024 and 2025, from $31,312 million to $29,792 million [6].
• X4 — Consensus-saturated story. Tested at today's price, not the peak: 2.95 times FY2025 sales against 28.6 times at the 4 December 2024 peak, 23 of 36 ratings at hold or lower, and an 88.0% drawdown. The skeptic pass weakened this claim on one leg — the Viant Technology 2.08x peer multiple rests on a share price the corpus does not carry — so the peer comparison is set aside and the three corpus-supported tests stand on their own. What survives as a residual observation, not a screen result: a business that carried 28.6 times sales nineteen months ago is cheap because of an active share-loss narrative, not because investors stopped looking.
• S1 — China dependence (flag only, not raised). The company discloses revenue as United States $2,476.7 million (85.5%) and International $419.6 million (14.5%) [7], with international property, equipment and lease assets at 11.7% of the total [8]. China is named twice in the FY2025 10-K, both times inside a list of growth-opportunity markets alongside the United Kingdom, Germany, France, Japan, India and Australia. The ceiling on China revenue is therefore 14.5% by construction and almost certainly a small fraction of that — but the corpus carries no China line, so the figure cannot be sized, and absence of disclosure is evidence of immateriality under SEC thresholds rather than of absence.

Pattern match

Against the reader contract's four recognition setups, this fits none of them cleanly, and the nearest one fails on its own check.

The closest is quality tech monopoly or duopoly on a fear dip — the precedent shape is a franchise whose average top-to-bottom swing runs far wider than any change in its value, bought when the fear is specific and testable. Two of that pattern's markers are present: the fear is specific (an inventory-owning rival taking connected-television share) and the drawdown is deep enough to qualify on any reading. The structure check is where it fails, and it fails on the issuer's own words: "Our industry is highly competitive and fragmented. We compete with other demand-side platform providers, some of which are smaller, privately held companies and others are divisions of large, well-established companies such as Google and Amazon" [9]. That sentence has appeared in every annual report since FY2021, with the named rivals changing from Google and Adobe [10] to Google and Amazon. Scale runs the same way: FY2025 gross spend of $13.4 billion is 19.5% of Amazon's $68.6 billion of advertising services revenue [11] and 4.5% of Google's $294.7 billion of advertising revenue [12]. A duopolist does not describe itself as one of many in a fragmented field.

The other three patterns are absent on their entry conditions. Cyclicals at the bottom, large banks only — not a bank, not a cyclical with a monthly loss-rate series to read against macro fear. High dividend plus high FCF yield — no dividend has ever been declared or paid, none is anticipated, and the credit agreement restricts payment [13]. Healthcare or insurance forecasting error — there is no regulated repricing mechanism here; advertising budgets have no annual rate filing that mechanically readjusts the book, which is precisely what the Clock tab finds when it looks for a repricing cycle and finds an execution clock instead.

The pillar ledger

Source: ruchir/fit_tally.json — per-criterion verdicts, vote splits, trimmed-mean probabilities and spreads, as recorded by the deterministic tally.

Year-10 durability (P1) — not met

The gate is binary by construction, and any proper doubt resolves it. All four seats returned not met, at a trimmed-mean probability of 0.585 with a spread of 0.04 — the tightest agreement anywhere in the ledger.

The deciding arithmetic is that none of the framework's five conviction sources is present on the company's own filed description. Market structure: the issuer calls its industry highly competitive and fragmented [14], and $13,394.7 million of gross spend is 1.9% of the $700 billion digital advertising market the company itself sizes [15]. Regulatory entry barriers: no licensing regime is disclosed anywhere in the 10-K. Capital intensity: FY2025 capital expenditure of $197.0 million is 6.8% of revenue, and net property and equipment of $396.8 million is 6.4% of $6,153.2 million of total assets. Essential products: master services agreements "do not contain any material commitments on behalf of clients" and are "terminable at any time upon 60 days' notice by either party" [16]. Operating history: incorporated November 2009, listed September 2016 — about 16 years old and under ten years public, against a 30-to-50-year reference. What is left protecting year-10 cash flow is execution plus a 21.6% take rate, and the framework does not count execution as a moat.

The strongest surviving counter-fact belongs in the same treatment, because it is substantial. Revenue has grown in all ten reported fiscal years and stands at 14.3 times its FY2016 level; the client retention rate exceeded 95% in each of the eleven years to FY2024 [17]; the platform draws inventory from over 430 directly integrated exchanges, publishers and supply-side platforms and more than 370 third-party data vendors [18]; the balance sheet carries no funded debt against $658.2 million of cash and $644.9 million of short-term investments [19]; and consensus free cash flow rises from $780 million in FY2026 to $1,098 million in FY2029. The gate did not fail because the business is weak. It failed because the sources of conviction the framework requires — structure, regulation, capital, essentialness, longevity — are not the sources of this company's strength. Full treatment in Durability and Business.

Scale asymmetry frames the doubt: Amazon's advertising services revenue of $68,635 million is 23.7 times The Trade Desk's revenue and grew 22.1% in FY2025 [20], while the chief executive told the Q3 FY2025 call: "Looking ahead to the next ten years, I doubt Amazon will develop a DSP as we understand it" [21]. Two companies made incompatible claims in the same quarter, and neither discloses demand-side-platform revenue, so the primary record cannot settle it.

FCF consistency (P2) — not met

Four seats, unanimous. The framework's basis is adjusted free cash flow — reported free cash flow less stock-based compensation less the trailing five-year average of acquisition spend — and the feature file returns not_computable for every year FY2016–FY2025 because stock compensation is absent from the structured cash-flow feed. Recomputed from the filed statements [22], the rolling five-year averages are $33.4 million (FY2019–23), $73.5 million (FY2020–24) and $89.9 million (FY2021–25). Only three complete windows exist, and the average nearly tripled across them — rising, not stable — while adjusted free cash flow itself was negative in three of the seven years to FY2025 (FY2019, FY2021, FY2022). Neither the stability the criterion asks for nor the business-model-inherent 5-to-8-year negative episode it tolerates is what the series shows.

The counter-fact carried in the same treatment: those negative years are stock-compensation artefacts rather than operating shortfalls. Reported free cash flow was positive and rising in every year on record, from $24.5 million in FY2019 to $795.7 million in FY2025, and stock compensation has been flat in dollars for four years ($498.6M, $491.6M, $494.7M, $490.6M) while revenue grew 83.6%, taking it from 31.6% to 16.9% of revenue and lifting adjusted free cash flow 5.3 times in two years, from $57.3 million to $301.6 million.

Dislocation and yield (P3a, P3b, P3c, P3d)

Source: derived — adjusted FCF = reported FCF minus stock-based compensation minus the trailing 5-year average of acquisition spend, computed from the filed cash-flow statements [23], over the $8,533.5 million market capitalisation; the 8.5% line is the framework's fortress bar.

P3a — Identifiable event: met, four seats. Two dated adverse events produced the two largest single-day falls: the Q4 2024 revenue miss reported on 12 February 2025 [24], followed by a 32.98% decline the next session, and the guidance reset [25] with a simultaneous chief financial officer departure on 7 August 2025 [26], followed by 38.61%, the largest one-day fall in the company's history. The counter-fact that sits inside this reading, and it is the one that separates this case from the framework's canonical setup: there was no near-term earnings cut in level terms. FY2026 consensus earnings per share of $1.85 is 4.6% above the FY2025 actual of $1.77, and consensus revenue of $3,177.1 million is 9.7% above FY2025. What was cut is the growth rate — guided quarterly growth stepped from 17.0% to 8.1% across six consecutive guides — and the price fell 87.6% against it.

P3b — Capitulation: met, four seats. The maximum 20-day average volume during the fall was 8.89 times the pre-peak median, against a 2x reference line, with single sessions at 36.1x (8 August 2025) and 28.3x (5 March 2026). The counter-fact is a timing problem rather than a level problem: that 8.89x window ended 24 March 2026, at closing prices of $22.34 to $29.79, which is 29% to 72% above today; at the 23 July 2026 low the trailing 20-day average was 4.48x, the lightest reading in five months. Peak fear was four months ago and the price is lower. Full treatment in Dislocation.

P3c — Yield versus the bar: not met, four seats, at 497 basis points short. The balance sheet is net cash — $1,303.1 million of cash and short-term investments against zero drawn debt at FY2025 year-end [27] — which selects the fortress reference line of 8.5%, the most favourable of the three bars available. On that line: FY2025 adjusted free cash flow of $301.6 million over the $8,533.5 million market capitalisation is 3.53%, the three-year average is 1.96%, and the trailing twelve months to 31 March 2026 is 4.31%. The single number that carries the most weight in the other direction is the unadjusted one: reported free cash flow of $795.7 million is a 9.32% yield, above the bar. The gap between 9.32% and 3.53% is stock-based compensation of $490.6 million [28], and the framework's basis subtracts it. Adjusted free cash flow has risen every year since FY2022, from minus $36.7 million to $301.6 million, because that expense has been flat in dollars while revenue grew 83.6% — the trajectory is toward the bar, from a long way below it.

P3d — Forward path: not met, four seats, at a trimmed-mean probability of 0.18 with a spread of 0.08. Consensus free cash flow clears the bar on the vendor's own definition — 9.14% for FY2026 and 10.35% for FY2027 on today's market capitalisation — but that definition is cash from operations less capital expenditure, which adds stock compensation back. Restated onto the framework's basis with stock compensation held flat, the same consensus gives 3.35% in FY2026 rising to 7.08% by FY2029; banking the chief executive's performance option roll-off in full lifts the series to 4.17% and 7.95%. Neither path reaches 8% anywhere in the estimate horizon. Reaching the bar by FY2028 requires $682.7 million of adjusted free cash flow, which needs reported free cash flow about 14% above consensus, or a 43% cut to stock compensation against $290 million a year of already-committed amortisation. The skeptic pass weakened the point estimate and it is carried as a range: roughly 10% to 30%, with 20% the working figure. Full treatment in Yield.

Balance sheet and self-help (P4a, P4b, P4c)

P4a — Outlast and allocation headroom: met, four seats. At 31 March 2026 the company held $878.4 million of cash and $527.5 million of short-term investments [29] against zero funded debt and an undrawn $750 million revolver maturing 14 April 2031; the maturity ladder is leases only, cash interest paid in FY2025 was $0.993 million, and working capital is $1,971.2 million. Debt paydown cannot become the competing use of cash, because there is no debt to pay down. The counter-fact in the same treatment: cash and short-term investments fell from $1,921.5 million to $1,303.1 million over FY2025 because repurchases of $1,380.4 million exceeded that year's free cash flow of $795.7 million by $585.1 million — the balance sheet is funding the buyback, not the other way round.

P4b — Repurchase engine: not met, on the ledger's only genuinely split vote: one seat met, three not met, and cross-family agreement false. The program is executed rather than announced — $2,445.8 million of repurchase value against $2,761 million authorised, 45,944 thousand shares retired program to date — but 57.6% of it was absorbed by reissuance to employees, so the share count fell 4.0% from 490,468 thousand at end-2022 [30] to 470,999 thousand at 31 March 2026 [31], and the filings describe the program's purpose as offsetting dilution. The average price paid across the program is $53.23 against $17.29 today. The dissenting seat's case is on the record too: the count fell 4.1% in FY2025 alone and 5.06% over five quarters from its 2024 peak, and March 2026 repurchases were struck at $24.31, the lowest of the program. Full treatment in Self-Help.

P4c — Dividend cover: not applicable, four seats. No dividend has ever been declared or paid, none is anticipated, and the credit agreement restricts payment [32]. No part of the return case depends on one.

Diagnosis (P5) — contested

The temporary-or-permanent question was decided by the profile's adversarial trial: two opposing cited briefs, three judges reading in different orders. The result is a probability that the impairment is temporary of 0.41, with a mean of 0.4633 and a spread of 0.30 across per-judge readings of 0.34, 0.41 and 0.64. The tally records it as contested, and it is not order-stable: the seat that read the temporary brief first came out at 0.34, while the two that read the permanent brief first averaged 0.525 — a reading-order gap of 0.185. All four jury seats carried the trial result rather than re-deriving it.

The cash-flow arithmetic behind the two readings, on identical assumptions of a 10% discount rate and 3% terminal growth from year eleven: on reported free cash flow the shares at $17.29 sit 52% below the permanent-scenario value of $35.76; on the framework's adjusted basis, once the $462.2 million of stock compensation consensus carries for FY2026 is removed, the same permanent scenario values the equity at $16.11 and the shares sit 7% above it. The skeptic pass weakened this claim on sensitivity and the range is carried: across a 9%–11% discount rate and 2%–4% terminal growth, the adjusted permanent value spans $13.64 to $20.48, so at 9% and 3% the price flips back below value. The counter-fact from the trial itself: the 0.64 seat discredited the permanent brief's only attempt to show declining earning power — Q1 FY2026 net income of $40.0 million against $50.7 million — on the ground that pretax income rose and operating income rose 22%, making the fall a tax-rate artefact. Three quote checks failed across the two briefs, one against the temporary case and two against the permanent case. Full treatment in Damage Math.

Instrument context (I1) — not verifiable

Four seats, unanimous, and this criterion never blocks a pillar verdict. Listed options on TTD do extend to 21 January 2028 — 545 days, or 17.9 months, from the reference date, clearing the framework's 12-month test and reaching its 18-month preference — with intermediate expiries at 15 January 2027, 19 March 2027 and 17 June 2027. Implied volatility reads 84.63% on the 30-day mean and 72.54% on the 180-day mean, both dated 24 July 2026, against the framework's reference lines of roughly 55 acceptable and 60–70 elevated. The reason the criterion returns not verifiable rather than exists: every one of those figures is web-sourced, the corpus carries no option chain, and the named missing datapoint is aggregate open interest and bid-ask spreads on the long-dated contracts from a dated citable source — three vendors returned HTTP 403 to automated retrieval. Because the tally could not establish the instrument facts, the watchlist-only overlay is not set; that flag reads false. Full treatment in Clock.

What a 3x-in-3-years would require

The tally returns no re-rating arithmetic. Its note, in full: Re-rating math unavailable because the applicable bar or normalized adjusted FCF is missing.

The reason is specific and worth stating plainly. fit_features.adjusted_fcf, adjusted_fcf_yield, yield_baseline, fcf_stability and float_retirement_years all return not_computable because stock-based compensation is absent from the structured cash-flow feed for every fiscal year FY2016–FY2025, and balance_sheet_class returns unknown because debt or cash is missing for FY2025. Every adjusted figure on this tab was recomputed from the filed statements instead, and flagged as such — but the deterministic re-rating engine takes its inputs from the feature file, and it had none. No price-at-bar and no upside-to-bar can be published from this run.

What can be published is the base-rate context the Clock tab established from this name's own history. Seven completed drawdowns of 30% or more have occurred since the September 2016 listing; their median depth is 35.6% and their median round trip 165 days. The deepest, at 64.27%, ran 358 days to its trough and 1,053 days — 2.88 years — from peak back to peak, and it did so while FY2022 revenue grew 31.9%. The current episode is 87.97% deep and 596 days old, with the low set on 23 July 2026, one session before the reference date. Measuring forward from each of the seven prior troughs, the median 12-month gain was 171% and the median 24-month gain 377% — every one of those troughs occurred while revenue compounded 23% to 55% a year, against FY2026 consensus growth of 9.7%, so the recovery base rate is drawn from a materially different business. For scale on the nearer horizon: the sell side's mean target of $24.32 is 40.7% above the close, and the prior peak is 707% above it.

Contested and undetermined

Six criteria are recorded contested. Nothing returned cannot-determine.

X1, X2, X3, X4 and S1 are contested on the same mechanism. Three seats voted each of them; every seat's stated basis is that the exclusion is not hit (or, for S1, that the sensitivity flag is not raised); and the three seats expressed that finding with three different labels — seat a "not_met", seat b "no_hit", seat c "not_hit". The tally counts differing labels as a split, so cross-family agreement is recorded false and each criterion carries forward as contested. The masked seat returned "not_met" on all four exclusions, which is what the mask probe flags as a gate-criterion difference against the unmasked aggregate; the arithmetic distance between masked and unmasked probabilities is 0.02. Both readings, stated plainly: on the substance every seat found no exclusion hit, and the tally records the criteria as contested. The second reading is the one the deploy carries.

P5 is contested on substance, not labelling. All four seats returned contested, carrying the trial's 0.41 with its 0.30 spread and 0.185 reading-order gap. The temporary reading: three quantified, dated drags (the Publicis pause, the tariff-affected consumer-goods comp, the political comp) explain the deceleration, the take rate rose rather than fell, and reported cash flow has never declined. The permanent reading: an inventory-owning competitor 23.7 times the size has taken exclusive positions in the company's largest channel, the take rate carried 7.2 of FY2025's 18.5 points of revenue growth and a holding-company fee audit is aimed at exactly that take rate, and six consecutive guides have stepped down. The trial did not choose between them, and neither does this tab.

Undetermined: no criterion returned cannot-determine. I1 returned not verifiable, with its missing datapoint named above — aggregate open interest and bid-ask spreads on the long-dated contracts, from a dated citable source.

Provenance

Dimension	Reading
Jury composition	Four voting seats plus a name-masked probe. Seats a and b from one model family, seats c and d from a second; the masked seat from the first. Universe and exclusion criteria were voted by three seats; the pillar criteria by four.
Agreement and spread	Cross-family agreement holds on U1, U2, P1, P2, P3a, P3b, P3c, P3d, P4a, P4c and I1. It fails on X1, X2, X3, X4, S1 (label divergence), P4b (one seat met against three not met) and P5 (contested on all seats). Widest probability spread: P5 at 0.30. Tightest: P1 at 0.04.
Trial order-stability	Not stable. The seat reading the temporary brief first returned 0.34; the two reading the permanent brief first averaged 0.525. Gap 0.185.
Name-mask probe	prior_driven_risk is true. The masked jury differed from the unmasked aggregate on X1, X2, X3 and X4; maximum probability gap 0.02.
Skeptic counts	55 ledger claims entered the pass, 16 fully recomputed. Verdicts: 11 survived, 6 weakened, 0 refuted, 1 unverifiable, 37 triaged only.

Source: ruchir/fit_tally.json provenance block and ruchir/refutations.json.

Two sentences on what that means. Nothing in this verdict rests on a single model's reading: two families voted independently on every criterion, a fifth seat voted the same evidence with the company's name stripped out, and every verdict-bearing claim was handed to a skeptic that recomputed the arithmetic and re-opened the cited pages before the jury saw it — six claims came back weakened and are carried on this tab with the weakening stated. The confidence tier is low not because the universe test is doubtful (it is unanimous and arithmetic) but because the mask probe and the unmasked jury labelled the exclusion tests differently, which the tally treats as a signal that prior knowledge of the name may be doing work.

The falsifier ledger

These are the standing conditions that would change the readings above. The first five are the framework's own templates, applied to this name; the rest are the trial's dated flip conditions.

Framework condition	Direction	Window
adjusted FCF or EBITDA declines where flat-or-better was underwritten	Breaks the durability underwrite	Any reported period
revenue declines for a third consecutive year	Triggers the P1 disqualifier	Annual; none recorded to date
capital allocation pivots to debt paydown over repurchases	Breaks P4a	Currently inapplicable — no funded debt
share count inflects upward	Hard-fails P4b	Quarterly equity statement
the industry repricing cycle fails to materialize where industry-wide mean reversion was underwritten	Breaks the mean-reversion path in P3d	1–3 years

Condition that would support the temporary reading	Window
Q3 FY2026 (reports November 2026) revenue growth of 13%+ — a full quarter with Publicis spend restored (settled 12 June 2026) and the midterm political comp — which would establish the ~8% Q2 guide as pause-distorted rather than the structural run rate.	November 2026 (estimated)
FY2026 gross spend growth accelerates above the 11% recorded in FY2025 [33] while revenue/gross spend holds at or above 21.6% — showing neither platform volume nor take rate was repriced by the Publicis fee audit.	FY2026 results, February 2027 (estimated)
TTD discloses CTV growing faster than the total business for two further quarters after Amazon's exclusive Roku/Disney footprint is fully live, with total growth recovering — refuting the channel-share mechanism that makes this lasting rather than cyclical.	Two consecutive quarters
FY2026 Adjusted EBITDA margin lands at or above 40% with opex ex-SBC growth back below revenue growth, reversing Q1 FY2026's negative operating leverage (opex ex-SBC +18% on revenue +12%).	FY2026 results
Q3 and Q4 FY2026 revenue reaccelerates to mid-to-high-teens growth after Publicis resolution and political spend returns, while full-year adjusted EBITDA margin remains at least 40%.	November 2026 and February 2027
FY2026 revenue/gross spend stays near the FY2025 take rate and no major holding company or vertical materially reduces gross billings.	FY2026 results
FY2027 revenue and EPS revisions turn net positive for at least two consecutive quarters.	Two consecutive quarters
CTV share and pricing remain stable despite Amazon DSP's Roku and Disney integrations.	Ongoing

Condition that would support the permanent reading	Window
Q2 FY2026 (6 Aug 2026) revenue below the $750m guide, or FY2026 adjusted EBITDA margin guided below 40% — the first actual break in the never-declined cash-flow record that the temporary case rests on entirely.	6 August 2026
Q3 FY2026 growth fails to exceed ~10% despite the resolved Publicis dispute, the lapped tariff/CPG comp and the midterm political comp — that would convert management's three quantified, dated drags into a structural deceleration and validate the consensus curve.	November 2026 (estimated)
FY2026 revenue/gross spend falls below 20% (from 21.6% in FY2025), or TTD discloses a Publicis/holding-company fee concession with revenue impact — direct evidence the take rate, not the cycle, was repriced.	FY2026 results
TTD discloses CTV growing SLOWER than the overall business, or a named CTV supply partner going Amazon-exclusive — the Q4'25 disclosure that CTV grew faster than the company through 2025 is the load-bearing fact separating a vertical-cycle story from an Amazon-share-loss story.	Any quarter
Buybacks suspended or the $750m revolver drawn for operations.	Any quarter

Source: ruchir/fit_tally.json falsifier ledger and ruchir/trial/tally.json flip conditions; the FY2025 gross-spend figure of 11% is at [34].

The nearest of these is the 6 August 2026 print. The FY2026 adjusted EBITDA margin commitment — "at least 40%, approximately in line with 2025" [35] — is unresolved as of this run, because the corpus ends with the Q1 FY2026 call of 7 May 2026.

Data gaps

The tally records 45 data gaps. They fall into seven groups; the ones that constrain this tab's arithmetic come first.

The adjusted-FCF basis is not in the numeric feed. fit_features.adjusted_fcf, adjusted_fcf_yield, yield_baseline, fcf_stability and float_retirement_years all return not_computable because data/financials/cash_flow.json carries no stock-based-compensation field for any fiscal year FY2016–FY2025, and the acquisitions field is absent entirely. Every adjusted figure on this tab and on Yield, Durability and Self-Help was recomputed from the filed cash-flow statements under the profile's definition and labelled as such. This is also why the re-rating arithmetic could not be produced.

The balance-sheet class is not computed. fit_features.balance_sheet_class returns unknown ("debt or cash missing for FY 2025") even though the filed balance sheet reports both. The fortress classification that selects the 8.5% bar was established from the 10-K balance sheet and credit-facility disclosure rather than the feature.

Share-count basis conflicts across sources. fit_features.market_cap uses 493,551,000 shares (a weighted-average diluted figure from the FY2025 income statement); the 10-K cover page reports 475,977,047 outstanding at 31 January 2026 and the Q1 FY2026 10-Q reports 470,999,000 at 31 March 2026. Every yield and multiple on this tab uses the feature file's basis. On the cover-page count the market capitalisation is $8,230 million and the FY2025 adjusted yield 3.66% rather than 3.53% — neither basis clears the $10 billion universe line.

Competitive share cannot be settled from the primary record. Neither The Trade Desk nor Amazon discloses demand-side-platform revenue or market share, and no third-party DSP share data is present in the corpus. Every share statement in this run is a ratio of disclosed revenue or gross spend against the company's own $700 billion market sizing. TTD does not disclose CTV revenue in dollars, so the central feared event has no quarterly metric that can confirm or refute it — channel mix in prepared remarks (video including CTV at a low-50s percent of Q1 FY2026) is the only read available, and it has never been a filed figure in any 10-K. The company reports a single operating segment with no expense detail below the consolidated income statement, so no channel or geographic economics can be derived [36].

China cannot be sized. No China-specific revenue, asset, headcount or risk-factor disclosure exists anywhere in the indexed corpus. The tightest bound is the 14.49% of FY2025 revenue booked as International, a bucket that also contains all of EMEA and APAC.

Market and positioning data are thin or web-sourced. The dedicated web-research provider was unavailable for this run (the API returned a 402 insufficient-credit error), so universe verification, seller composition, the option chain, implied volatility and the Publicis dispute chronology were sourced by general web search and are dated July 2026. No reported short-interest series exists: FINRA returned zero position rows and every file under data/short_interest/ is empty, so the short-interest levels used in the seller-composition claim are not corpus-sourced — the skeptic pass weakened that claim on exactly this ground. Aggregate option open interest could not be obtained from any citable source. No 13F holder-level turnover is staged, so institutional selling is visible only through 5% crossings. Consensus revision history reaches back 180 days for FY2027/FY2028 and 90 days for FY2026, so the estimate path across the first fourteen months of the drawdown cannot be reconstructed, and the company's own six-guide sequence was used as the before-and-after record instead. Whether the company remains an S&P 500 constituent as of 24 July 2026 is not confirmed by any document in the corpus.

Two forward quarters are unobserved. Q2 FY2026 results (due 6 August 2026) are unreported, so the first full quarter carrying restored Publicis spend, a lapped tariff comp and the midterm political comp has not printed — the single largest determinant of whether the temporary or the permanent scenario applies. Q3 and Q4 FY2026 earnings dates are not yet scheduled; the November 2026 and February 2027 windows used above are inferred from the company's reporting pattern and are labelled as estimates. Stock compensation before FY2019 is not in this corpus, so only three complete rolling five-year adjusted-FCF windows exist, and quarterly revenue in the numeric feed contains no fourth quarters, so quarter-level recession testing could not be sourced to a filing page.