After Action Report  |  Proof of Concept 5

SPCX: Navigating Event Windows #1 and #2
with AI Agentic Analysts and Auditors

Window 1 asked whether the fee would keep rising into a release with no precedent. Window 2 asked by how much it would fall after one. The record answers both, and it does not answer them alike.

Ed Blount, Tom Daniels, Dan Hammond, and David Schwartz
Advanced Securities Consulting LLC

Section 1Executive Summary

Between July 30 and August 20, 2026, ASC's governed AI inference vector models issued twenty-two dated editions on the securities-lending market in SpaceX (SPCX) across the first two IPO lockup releases of August 6 and August 20. Proof of Concept 5 tested whether an AI-assisted securities-lending inference process could remain useful and controlled when model history was insufficient, the event had little precedent, data quality changed intraday and human judgment remained necessary. Each distributed edition stated its market-state read in writing, with the settled observable that would prove it wrong, before the resolving data existed. This report prints every edition in the order it was made, including the internal alphas that were never sent, the edition withheld as late, the two editions held on the auditor's data question and the two release-day editions that were held, and scores the distributed editions against the September 3 settled record; internal, withheld and held editions are printed and marked as such rather than scored.

The working hypothesis the experiment began with was that successive releases could overwhelm demand and crush the stock. They did not. The stock fell to an all-time low close of 108.27 the session before the first release and rallied off it; through the first countdown the editions read HOTTER against a stock still searching for a floor, then turned to COOLER at T+1 on the day the re-rates came down. Into the second release the stock eased, closing at 134.00 on August 20 against 143.34 two sessions earlier. The second window was less clean. The editions carried a fixed inference box with a direction, a range and a stated confidence, the settled fee moved less than five basis points on most days, and when an anomalous trade appeared in the real-time blotter (2mm shares at 58%) the agentic auditor held that day's editions until the print was corrected from a new trade print to an internal adjustment, while the release-day edition was held pending reconciliation.

The distributed editions were produced by a governed framework: written instructions, specialist agents applying a human-defined mandate, an independent agentic auditor (independent here meaning a separate instance with its own context and a written adjudication protocol, not an outside party), and a human who decided what could be claimed. The quantitative models entered that framework as an input it evaluated each session. SpaceX priced on June 11, listed on June 12, and the daily lending file for the name began on July 28. The eight-model deep ensemble therefore reached the first countdown with a settled fee record on the name that began on June 15, about thirty-three sessions, of which ASC had held the daily file for two days, and the framework held its output below the validation gate rather than publishing it, as the design had anticipated for a name this new. The same framework carried the five Window 1 reads, wrapped the Window 2 quantitative output in an inference box with a stated range and confidence, held the August 19 editions when the agentic auditor questioned an anomalous blotter trade, and they were scored in retrospect against the adjusted point, and held the August 20 edition on a tape whose manifest did not reconcile. Models of the same families, trained on a sufficient record, had carried the earlier proofs of concept, KLAR among them; the governance introduced here is what let the line run on a name where no such record yet existed.

What the two windows produced, beyond the scored record, is a working division of labor between the quantitative AI models and the language models, which this report calls the reasoning model. The quantitative rungs ran in ASC's shop. The agentic analysts held the corporate-action taxonomy, the Project Instructions and the context files of earlier events, read the model output against them, and wrote each edition's market state with its pre-committed conditions. The agentic auditor tested every edition for provenance, stale data, arithmetic and overstatement before a human released it. That division is portable. The instruction set and the context files can be furnished to a tester's own Claude instance; ASC runs the linear models and the non-linear models in its own shop and delivers the results through the inference state record (ISR), so that the analyst and the auditor do their reading inside the tester's walls, against the tester's own tickets or a vendor's aggregates, under the same rules that governed these twenty-two editions. How the analysts read, what the auditor refuses, and what the ISR carries are the subject of the technical report and of the tester program that follows it.

Section 2Event Background

SPCX combined minimal trading history, an IPO of unprecedented size, a rare staggered release structure and observable stock-loan demand. Nine scheduled steps would pass before all shares became eligible to trade, stretching the supply transition across roughly a year. SpaceX priced on June 11 and listed on June 12; every tranche date on this page is anchored to the June 11 pricing. SpaceX's first earnings report, for the second quarter, came on Tuesday, August 4, and under the amended S-1 the first tranche released on the second full trading day after it, Thursday, August 6, 2026; the second release fell on Thursday, August 20. The later tranches followed the calendar, on September 9, September 24, October 9 and October 24 (a Saturday, rolling to the next business day), with the 180-day lockup expiring December 8 and the insider block under the 366-day lockup releasing on June 12, 2027; only the bonus tranche and the third-quarter slug were conditioned on price and earnings tests. Index inclusion followed the listing: SPCX's addition to the Nasdaq-100 created mechanical borrow demand from index-tracking funds that the deluge thesis did not weigh. The July 31 edition sized the first tranche at approximately 920 million shares, about 1.4 times the IPO float of 638 million; the August 20 release covered roughly 319 million shares. Before any settled lending data existed for the name, the reasoning model, the qualitative layer of the production line, worked through the prospectus mechanics and produced the supply ladder that every later edition carried: which shares released on which date, by what mechanism, to which holder class, and which conditional windows would and would not open. ASC's earlier posts on the name set out that ground: SpaceX IPO: What Shorts and Stock Lenders Cannot Afford to Miss (March 31, 2026), SPCX: The Microeconomics of Short Selling into the Largest IPO in History (May 28), SPCX Day One: $160.95 and the Short Book Nobody Was Ready For (June 14), SPCX Window One: Three Borrowers, One Lendable Pool (June 22), and, at CSFME, Information Asymmetry and the Staggered Lockup: Analyzing the SPCX E1 Supply Shock (August 8).

The data history is the fact that governs everything that follows. Live access to the daily securities-lending feed for SPCX began on July 28, when the team also requested backfill to May 1. Training runs on the name began the week the first countdown began. A contemporaneous note from the modeling lead on July 29, reporting the first model output, put the caution in the model's own terms.

Even the model itself cautions that it is not doing a good job with this thin a sample.Modeling lead, internal note, July 29, 2026

The countdown method, written down in Proof of Concept 4 and demonstrated on KLAR in March 2026, ran from T-5 to T-0 with a fixed cadence: each session's edition stated a market state, a pre-committed condition under which the state held, and the settled observable that would falsify it.

Section 3Model Performance Evaluation

Which rung may publish: the fitness decision, session by session

The governance enhancement this proof of concept introduced to the quantitative models is a rung-fitness decision made in writing every session: which rung of the model ladder has enough history behind it to publish, and what the edition may say when the answer is none of the upper rungs. On SPCX the answer was known before the first window opened. No one expected a name that priced on June 11, with a daily lending file that began on July 28, to supply enough observations to train the eight-model deep ensemble, and the regression rungs were squeezed by the same short span. The ensemble was therefore held below its validation gate in both windows and its output entered no distributed edition. On August 13 the first inference box built from its output reported a direction confidence of 14 percent, and the two editions carrying it stayed internal. The July 31 edition capped its confidence at 75. The ladder, whose rungs run from momentum through the closed-form regression models to the eight-model deep ensemble, was stopped at momentum, and the editions said so: the July 31 edition led with the option-implied forward borrow rather than a model forecast, and the August 3 edition printed sixty tickets of re-rate direction rather than a point estimate. This was controlled abstention, the system declining unsupported complexity rather than dressing thin history in fluent prose.

Was abstention pre-specified?

The gate, the observation counts and the consequence as the record states them. The requirement and its date precede both windows; the decision follows from the count, not from the ensemble's later error record, which Section 3 reports separately as evidence the decision was sound.

The evidence the fitness decision rested on is in the record. On the four sessions where a one-step forecast could be scored against the settled fee, the ensemble's errors exceeded the naive carry's, with mean errors up to 4.10 percentage points of fee (410 bp, on the four Window 1 sessions, when the fee ran between 300 and 700 bp) against carry errors of 0.05 percentage points; on a fee whose settled record was about thirty-three sessions old and that barely moved between sessions, the carry's no-change forecast is the expected winner and the comparison says more about the sample than about the models. The linear rungs read the same way. A walk-forward replay of seven closed-form models on the SPCX new-fee series over 40 origins from June 17 to August 19 finds the best direction edge, Drift5 at 0.600 against a base rate of 0.525, inside its own binomial standard error, with persistence carrying the lowest level error at 40.3 bp (mean over the 40 origins) and no stable ranking across data variants. No trailing-window model anticipates the August 7 reversal, which is the mechanism the editions carried in their pre-committed conditions and the reason the reasoning model, not a rung of the ladder, produced the Window 1 reads.

What the reads show beyond a simple rule

A reader will ask what the record would have shown without the framework. A one-line rule, HOTTER while new loans print above the seasoned book, reproduces every Window 1 mark, because the intrinsic spread stayed positive through the countdown; the Window 1 marks are therefore not evidence that the reasoning model saw what a rule could not. What the rule does not reproduce is on the record in three places: the August 3 withdrawal of a morning shift when sixty live tickets contradicted it, the August 7 turn to COOLER on 142 of 148 re-rate tickets while the spread was still positive at settlement and the rule would still have read HOTTER, and the August 19 and 20 holds, which no rule on the fee series could have produced because the fault was in the input, not in the signal. In Window 2 a naive carry on direction would have regretted on the same flat sessions the distributed editions regretted on. This comparison is a reconstruction, stated as a limit on what the marks show rather than as a measured advantage.

This is a finding about data on one name, and the same model families have a record on names where the data existed. In Proof of Concept 1 (January 2025, reported on the ASC website in Traders' AI-Boosted Edge in Securities Lending of January 29 and Deep Learning vs. ARIMA of February 23, and in the CSFME article How Deep Learning is Revolutionizing Securities Lending of February 25, 2025) the deep-learning ensemble outperformed ARIMA on public lending data, with more than 72 percent of its correct forecasts at turning points and traders acting on AI-predicted reversals accounting for over 60 percent of $155,476 in net savings against the Goldilocks benchmark. Proof of Concept 2 (January to May 2025) tested the same model class against a custodian's own records on an agent lender's live book. Proof of Concept 3 (April to December 2025; final report of December 15, 2025) captured $12.2 million on Lennar in three sessions beginning November 25, 2025. In Proof of Concept 4, on the KLAR lockup of March 2026 (after-action report of March 10, 2026, published as KLAR: The Squeeze The Market Missed), the T-Series predictive models, benchmarked, calibrated and trained against the RPT corpus of confirmed historical events, identified a structural shortage when the market expected a supply deluge, and the outcome confirmed it: zero shares available and a 68.62 percent cost to borrow at the close of March 9. The models' measured record on the seasoned five-thousand-name universe the production platform serves every day is set out in the technical report; it is not summarized here because those figures await verification against the run manifests. What SPCX added was the governance that decides, on the record, when that history is present and when it is not.

Where the agentic layer wrapped the quantitative output

Gating was the smaller part of the enhancement. In four dated instances the reasoning model took quantitative product and returned it governed. On July 31 a second AI instance independently reproduced the option-parity extraction, 790 plus or minus 126 bp on the front segment and a 666 bp front-to-back spread at 4.7 standard errors, so that the figure the edition carried had been replicated before it was published, and the agentic auditor required that it be stated as ASC's inference rather than as observed fact. On August 3, before any number was quoted, an agent's review of the option chains rebuilt overnight (from 935 kilobytes to 86 megabytes, and from between four and fourteen strikes per snapshot to between 72 and 193) found the zero-days-to-expiry rows contaminated and 186 of 2,456 rows with null open interest clustered on nine dates where a vendor endpoint had timed out, and specified the fix; a synchronized quote capture was running on SPCX every sixty seconds by 13:44. On August 13 the inference box, with its direction, range, stated confidence and pre-committed conditions, was specified to be built automatically from the daily prediction file, so that in Window 2 the quantitative output entered every edition already wrapped in the terms on which it would be scored. And on August 14 a simple tape estimator assembled from the screened ticket feed placed within 1.15 bp of the settled fee against the carry's 4.12, the one scored observation (one session, August 14) in either window on which any instrument beat the carry. Each is the reasoning model inferring from, checking, or restating quantitative output; none is the layer producing a forecast of its own.

Where the auditor and the build refused

Beside that work sat the controls. The agentic layer enforced the validation boundary, tested evidence against human-defined instructions, capped confidence, and prevented unsupported output from entering the guidance. On July 31 the agentic auditor's first finding was that a sentence describing the options market as having raised its event-week borrow price stated an inference as observed fact; the edition was changed. On August 8 the production instance found a conflict between an accuracy claim and the governing instructions and would not continue until the Executive Director ratified. On August 19 the agentic auditor identified an anomalous trade in the real-time blotter (2mm shares at 58%), and both editions prepared that day were held rather than distributed until the print was confirmed to be an internal adjustment rather than a new loan. On August 20 the build refused its own output. Each of these is the control working, and none of them is a forecast.

The retrospective direction diagnostic

Separately from the record score, a fixed loss was applied after the fact to every issued inference against the settled new-loan fee: the machine call is HOTTER, COOLER or none; the realized sign is the change in the settled new fee with a five-basis-point flat band; and an edition regrets when the sign disagrees or the move is inside the band. It was built on September 8 as an after-action convention, and it is not evidence that the live editions were prospectively pre-committed forecasts under this loss. Of thirteen eligible distributed editions, seven do not regret. In Window 1, one of seven regrets (July 31, when the fee fell on the day of a HOTTER read that nonetheless held on its settled spread condition). In Window 2, six of six regret, because the settled fee moved less than five basis points or the other way on every scored day. The two measures answer different questions. The record asks whether the edition's own stated condition held. The diagnostic asks whether a fixed loss would have paid. On Window 2 the answer to the second question is no, six times, and the page says so.

Speed and error handling inside the trial

Cutoff-to-delivery intervals ran about two hours at T-4 (tape 12:29, export 14:44), two hours forty-six minutes at T-3, about three hours twenty minutes at T-2, and a missed session at T-1, disclosed on the face. By T-0 the two blasts left within four and thirty-nine minutes of their control totals; T+1 went out about fifteen minutes after cutoff. In the second window the August 18 premarket left between 09:23 and 09:27, the first true pre-open delivery of the series, and the August 20 edition was held. The improvement came from moving recurrent work earlier in the session and teaching the agentic analyst and agentic auditor to perform repeatable checks before final review. Four defects, counted as defects rather than as editions, traveled to recipients in distributed editions during the two windows; each is recorded and each was corrected forward, by reissue or by withdrawing a class of figure, and in one case the cure did not reach the recipient. The instances are printed in the technical report so that the criterion for the next window can be tightened if the reviewer so decides.

Section 4Outcome Validation

The scored record follows; the edition-by-edition narrative behind it is in Section 5. Window 1 is recorded edition by edition against the condition each published: whether the stated condition held through settlement or the inference lapsed. Window 2 is presented as issued; the three boxes the September 3 record scores are described, with the range test and the release-quality test shown in separate columns, and the row-level marks are held pending reconciliation of the denominator and the range rule before any is claimed. Between the two windows the settled new-loan fee fell from 697 bp at the first release to 156 bp on August 7 and 28 bp by August 12; every Window 2 range is a level of that fee after the collapse.

Window 1 Pre-committed

Five distributed editions; the stated condition of each held through settlement and none lapsed, September 3 record. The July 30 internal test, scored in retrospect as a control, failed and is not in the denominator; the August 5 premarket alpha, whose condition held when marked in retrospect, is not in the numerator.

Window 2 Pre-committed

The September 3 record scores three Window 2 boxes: one whose range contained the point with direction matched (August 18 intraday, 17.54 bp inside +14 to +20), one whose point fell above the range (August 17, 20.34 bp against +8.6 to +15.3), and one marked a miss on release-quality grounds although its range contained the point (August 18 premarket); the August 14 box is not scored on that record. The Analytical and Production columns separate the range test from the release-quality test, which the September 3 record combined in a single mark. Issued ranges and adjudicated points are levels of the settled new-loan fee in basis points, printed with the leading plus sign the editions used; they are not changes, and the day's settled change appears separately in Section 5. Row-level marks are held pending reconciliation of the denominator and the range rule; the marks printed here are the September 3 record's, reported as that record states them, not this page's own adjudication. The magnitude test written into the acceptance criterion, that more than half of intraday range readings fall within two standard deviations, is reported as not resolvable: nine candidate dispersion constructions exist; taken over the panel's intraday range readings as a whole, the same readings pass at 65.5 percent under one construction and fail at 41.4 percent under another; and the population scoreable inside the countdown windows is two, below any calibration floor. The denominators are stated in the technical report.

Section 5Day-by-Day Analytical Record

Twenty-two editions, in the order they were made, behind the marks summarized in Section 4. Marks are as of the September 3 record and are marks against pre-committed conditions; the Retrospective diagnostic is in Section 3. Internal alphas, withheld and held editions appear in italic. Timestamps are Eastern.

Window 1: countdown T-5 to T+1

I want the human in the loop (me) to have another human (you) backing up my ok-to-send.Executive Director, internal message, July 30, 2026, 10:53 ET

The morning's labeled shift toward PUSH was withdrawn, plainly, when the live session declined to confirm it. The tier table stays HOLD. … Friday settled fired both pre-committed shift conditions; Monday's tape did not confirm them: new loans +3.80% volume-weighted through 12:17 PM, ascending hourly to +4.06%, +160 bp over the book; the heaviest recall wave of the series (at least 2.05M units, panel basis); returns near a tenth of Friday's midday pace.SPCX T-3 Guidance, August 3, 2026, midday edition as issued. Historical record; the labels PUSH and HOLD were the edition's own and are superseded by the desk-neutral HOTTER and COOLER used on this page.

Window 2: countdown T-5 to T-0

Section 6Market Consensus vs. Model Assessment

With the record in view, the two windows can be set against what was expected of them. They posed different questions.

Window 1: the release of August 6

Window 2: the release of August 20

Section 7Conclusions

For a reader who has skimmed, the record comes to this. Twenty-two dated editions were made between July 30 and August 20, 2026, across the two SpaceX lockup releases: distributed, internal, withheld and held, every one printed above in the order it was made. In the first window, five distributed editions carried a market-state read into and through a release with no precedent, HOTTER through the countdown and COOLER the day after, each with a condition stated in advance that would have lapsed it, and none lapsed: five of five held through settlement, and the turn to COOLER on August 7 came on 142 of 148 re-rate tickets. In the second window, the distributed editions carried a range and a stated confidence on a fee that had collapsed from 697 bp to under 30 bp between the two releases, and under the retrospective direction diagnostic all six regret; the page does not claim a Window 2 edition as a hit. The settled fee moved less than five basis points on every session with a distributed edition, one issued range saw the point fall above it, and two days of editions were held because the auditor would not pass the data.

What the numbers describe is a governed process rather than a forecasting contest. The eight-model ensemble never cleared its 90-session validation gate on a name whose settled fee record began June 15, and its output entered no distributed edition; the reasoning model, the language-model layer, produced every read, wrapped the Window 2 output in the terms on which it would be scored, and stopped the line twice, on August 19 when an anomalous blotter print (2mm shares at 58%) was questioned and on August 20 when a tape did not reconcile. The human contribution became reusable along the way: rulings on conflicting instances, on lateness and on unreconciled tapes became instructions used in later runs, so the human checkpoint got shorter because the system arrived better prepared. The second window's result is printed as plainly as the first's because the first would not be credible without it.

The governance the loop produced is now written where a machine can use it, and that is what makes it transferable. The inference state record (ISR, v1.0, September 10) fixes the fields ASC's models deliver to a client AI: state, score, range, confidence, pre-committed conditions and provenance. The range-construction method and its backtest across both windows (September 6) fix how an inference box is drawn from model output. The daily process guide for the TSMixer refit (September 13) and the linear-model specification (September 10) fix which rungs run in ASC's shop, on what history, and how they are scored. The Project Instructions, version-controlled as prompts-as-code, and the context files of the earlier events (KLAR, CoreWeave and Lennar) and of the SPCX windows themselves are what a tester's agentic analyst would read from on the first morning of its own countdown. The quantitative AI models did the arithmetic; the language models, the reasoning model, did the reading and the checking; a person signed. The next test of the arrangement is whether it holds when the reading is done on someone else's desk. ASC demonstrated a governed inference process that knew when to use a model, when not to use one, when to change its view, and when not to publish at all.

Administrative notes. The population rule scores only editions distributed to beta recipients. The September 7 ruling fixes the panel at fifty-one inferences: twenty-nine produced before the first countdown, of which the twenty-eight dated June 24 to July 30 are excluded as pre-feed, and the twenty-two of the two countdown windows printed on this page. The July 30 date appears on both sides of that line: an annotated note produced before the countdown opened sits in the pre-countdown twenty-nine, and the July 30 internal test printed on this page is the first countdown edition. Window 2 row-level marks await reconciliation of the denominator and the range rule. The magnitude test is reported as not resolvable.

This report is prepared for informational and training purposes only and does not constitute investment advice. It describes market state under ASC's current taxonomy (HOTTER and COOLER) as read in dated editions and marked afterward on a settled record; the editions as delivered carried the labels HOLD and PUSH of the Rate Directive product then in use, mapped on this page to those states. It does not recommend any action to any lender, borrower or intermediary. Ticket totals are floors on a capped extract; sign conventions are assumed, not observed; Window 2 row-level marks are pending reconciliation; no model output that had cleared the validation gate entered any distributed edition. Source: SPCX E1 and E2 inference panel of September 7, 2026; September 3 settled record; ASC internal report of August 25, 2026. © 2026 Advanced Securities Consulting LLC. All rights reserved. Advanced Securities Consulting LLC, 1101 Pennsylvania Avenue, Suite 300, Washington, D.C. 20004.

· ◇ ·