461300 — i-Scream Media

i-Scream Media (아이스크림미디어) — KOSDAQ 461300 — deep-value analysis (REFRESH)

Analysis date: 09/15/2026 · Reference price: … (data pack 09/15/2026 01:03; the tracker gives … as of 09/14/2026) · Single currency: KRW for reporting AND price (no conversion risk) Regime: REFRESH on the thesis of 08/23/2026, triggered by deep_delta.py on 09/13/2026 with the reason “3 unverifiable falsifiers.” Primary sources: the H1 2026 semiannual report filed with DART on 08/14/2026 (rcpNo 20260814000516), the FY2025 annual report (rcpNo 20260319001077), and — new versus the prior analysis — the entire 2026 DART filing register (34 filings), from which I fully read the mass-holdings reports (rcpNo 20260827000272 and the six prior ones from 2026), the interim dividend declaration (20260807900270), and the corrected stock-option declaration (20260729000460).

Executive summary (1 page: thesis, estimated value, verdict)

What I explicitly inherit. The 08/23/2026 analysis was based on the H1 2026 semiannual report, filed with DART on 08/14/2026. The DART register checked today confirms no new periodic filing exists — the next quarterly report (Q3 2026) is expected around 11/14/2026. I therefore inherit unchanged: the business and moat description, the segment table from note 36, the control structure, the O’Glove analysis of receivables/inventory/debt, the annual margin series, the FCF bridge, and the owner-earnings definition (…). Nothing in these chapters has been invalidated by new facts.

Why the file reopened and what I fixed. The mechanical triage didn’t find deterioration; it found that three of eight falsifiers of the thesis can’t be evaluated at all: pe and pb are None in yfinance for 461300.KQ (verified today with teza_deep._serie_metrica: both return [None]), and revenue_growth_yoy required two consecutive quarters, one of which is missing from the yfinance series (Q1 2025). A thesis whose falsifiers can’t be read isn’t a testable thesis — it’s an opinion with decoration. The final chapter of this report reconstructs the falsifier set exclusively on metrics I’ve tested to actually populate.

What is truly new and changes something. Three things, all found today in filings the prior analysis hadn’t opened:

1. The controlling family cut its share pledges by 39% over eight months, without selling a single share. Pledged shares fell from 1,831,798 (13.69%) in January to 1,120,537 (8.35%) as of 08/27/2026, while the group’s holding stayed fixed at 8,182,149 shares throughout 2026. The latest move, from 08/25/2026, is explicitly “기존 주식담보 일부 상환” — a partial repayment of Park Ki-suk’s share-collateralized loan — and it occurs exactly on the day of the interim dividend payment. This directly contradicts the darkest hypothesis from the prior pre-mortem (the i-Scream Edu precedent, the family selling after listing) and mechanically explains why the 40% payout policy is sustainable: it funds the family’s personal deleveraging.
2. Dilution through options is 2.7x larger than I estimated in August. The corrected declaration of 07/29/2026 gives 440,500 unexercised options outstanding, 3.30% of capital — not ~162,000 (1.2%), as had resulted from a partial reading of the semiannual report. The 03/27/2025 tranche has an exercise price of … and is settled through new share issuance (신주교부), not from treasury: it’s real dilution, already 46% in the money.
3. The minority holder’s access to the portfolio is now modeled, not ignored. The prior report had flagged its own defect (“Monte Carlo gives 100% probability of undervaluation because it doesn’t randomize exactly the variable that matters”). I ran the simulation with λ = 0.40 on net cash, calibrated on the declared payout policy. The median MOS falls from … to …, and the probability of undervaluation from …% to …%.

The most important correction of reasoning. The August thesis was built on “81% of the price is covered by net financial assets.” Today’s arithmetic shows the thesis doesn’t need the portfolio to work. The DCF on the business alone (owner earnings …, g1 4%, r 12%, gt 0.5%, zero credit for the portfolio) gives an enterprise value of …, i.e. …/share — 53.6% above the price of …. For the current price to be correct with zero credit for the portfolio, normalized owner earnings would need to be 23,438 mil., i.e. half of TTM FCFE of …. The portfolio is optionality, not the thesis’s foundation. This makes the thesis significantly more robust than the prior report described it — and moves the central risk from “will I ever get the money?” to “how much does the publisher produce after the curricular wave?”.

Estimated value. Five models, all recalculated at the price of … and with the access judgment made explicit: from liquidation with the business at zero (…, …) to a bull DCF with the 2028 middle-school wave (…, …). Median: …. Monte Carlo over 20,000 scenarios, with λ = 0.40: median intrinsic value …, P10–P90 range from … to ….

Verdict: BUY, an option-type position, not a core one — the same verdict as in August, but now supported on a different leg. The cheapness is verified in the filing and doesn’t depend on the portfolio. Three risks remain: the curricular cycle at its peak with development cut 65%, a … portfolio of AI/semiconductor stocks that can reverse the press headlines in a single half, and a board with nobody competent to oversee that portfolio. Position sizing should be set on model 1 (…) as the loss threshold, not on the Monte Carlo median.