Northrim BanCorp, Inc. NRIM

NRIM — Northrim BanCorp, Inc. — Deep-value analysis (REFRESH)

Analysis date: 2026-09-02 · Price: $25.38 (data pack yfinance + tracker, 2026-09-02) · Market cap: $564.6M · Sector: Financial Services / Banks – Regional · Exchange: NASDAQ · HQ: Anchorage, Alaska · CIK: 0001163370

REFRESH mode. The reference analysis is the one from 2026-07-28 (rapoarte/deep/2026-07-28-deep-NRIM.md, verdict MONITOR at a price of $26.74). The deep_delta.py triage flagged a new filing (10-Q Q2 2026, filed 2026-07-31) and one unverifiable falsifier (the NIM engine). What I inherit and what I re-derive is stated explicitly at the start of each chapter.

New primary sources for this refresh: 10-Q Q2 2026 (filed 2026-07-31, accession 0001163370-26-000029, document nrim-20260630.htm) — not in the local file; edgar_10k_downloader.py only downloads 10-Qs for a completed fiscal year, so I pulled it directly from EDGAR and converted it with filing_text.py. The rest: 10-K FY2025 (2026-03-06), DEF 14A (2026-04-14), the Q2 2026 results 8-K and the PBCO merger 8-K (both 2026-07-23). Deterministic data pack: data-pack-NRIM-20260902.md. Research brief: research-NRIM-20260902.md. Simulation: mc-NRIM-20260902.json.

---

Executive summary

What I inherit unchanged from the reference analysis: the business description and three-segment structure, the nature of the regional moat, the PBCO deal terms (1.160 ratio, ~$167.3M, ~21% dilution, closing Q4 2026/Q1 2027), management's capital-allocation profile, the fact that the "302.9% dilution" in the data pack is an artifact of the 4-for-1 split on September 18, 2025, and the $57.44M owner-earnings base. None of these have moved: the new filing covers exactly the same quarter (Q2 2026) the July report had already analyzed from the July 22-23 8-K press release. The refresh doesn't bring a new quarter of profit — it brings the exhibits behind the press release.

What I re-derive, and why it matters. The four tables in the 10-Q that the press release didn't contain partially reverse the reference thesis's most emphasized red flag:

1. Criticized loans FELL, they didn't rise: $86,318 thousand (12/31/2025) → $68,996 thousand (06/30/2026), −20.1%; net of government guarantees $43,979 → $42,606 thousand, −3.1%. As a share of the portfolio: 3.76% → 2.89% [10-Q Q2 2026, Note 3, risk-rating-by-origination-year tables].
2. Potential problem loans fell by half: $21.2M (12/31/2025) → $11.5M (06/30/2026), and the 10-Q states explicitly why: "primarily due to the transfer of three loans to a single borrower to nonaccrual status, as well as paydowns" [10-Q, MD&A, "Potential problem assets"].
3. So the nonaccrual jump from $11.97M to $22.98M is not new deterioration — it's the migration of an already-identified exposure from the watchlist into nonaccrual. The Community Banking segment's net adversely classified assets are $28.9M at 06/30/2026 versus $32.1M at 06/30/2025: down 10% year-over-year, after rising 75% in Q2 2025 [10-Q, MD&A, Community Banking segment].
4. Actual losses are near zero: gross loan charge-offs $394 thousand in H1 2026, recoveries $71 thousand → $323 thousand net, i.e. 0.014% of the portfolio over six months. The loan provision was $2,047 thousand — 6.3× realized losses. The bank is building its reserve, not harvesting it.

What's worsening, on the other hand — and this is a new finding, not inherited: the purchased receivables (factoring, Specialty Finance) book. It grew 20.8% in six months ($101.6M → $122.8M) while revenue from it grew 4.6% ($12,047 → $12,605 thousand), and its provision was zero for all of 2025 and all of Q1 2026, rebuilt only in Q2 2026 to $610 thousand = 0.49% of balance — on a book that in 2025 posted a net charge-off ratio of 2.16% of average balance [10-K FY2025, purchased receivables ACL table; 10-Q Q2 2026, Note 4]. Management's own stated reason for rebuilding it is itself a signal: "an increase in concentration of these assets."

Model correction — mandatory. The Monte Carlo simulation in the delta-triage pack (mc-NRIM-20260901.json) is invalid: it received --oe 57443 (thousand USD) alongside --sh 22.11 (millions of shares), a 1000× unit error. That's why it reported a median intrinsic value of $26,756.51/share and a median MOS of +103,849% — figures absurd on their face, but which would have passed unnoticed as a marker line. I reran it with correct units (--oe 57.443 --sh 22.544): median IV $26.24, median MOS +3.4%, undervaluation probability 54.0%. The same error existed in the reference report (its DEEPMC line was just as inflated) — so this isn't a new regression, it's an inherited error the refresh fixes.

The thesis, updated. Northrim remains an Alaska community bank with top-tier profitability for the sector (ROAE 17.7%, TTM adjusted ROTE 20.5%), trading at 8.2× TTM GAAP profit, 9.8× adjusted profit (excluding the $14.5M one-off gain from the Pacific Wealth Advisors sale) and 1.90× tangible book value. The growth engine — net interest margin expansion — is confirmed and more durable than I thought: of the +29 bp of NIMTE year-over-year (4.72% → 5.01%), only part comes from deposit repricing; the rest comes from the asset mix (loans rose from 75.19% to 79.29% of interest-earning assets) and from the securities portfolio rolling upward (yield 3.07% → 3.79%) — two favorable winds independent of the Fed's next move.

Estimated value (5 triangulated models): from −8.7% (EPV at an 11% cost of capital) to +34.3% (excess-return P/TBV with a sustained 17% ROTE), with the base (median) at +3.0%. Monte Carlo across 20,000 scenarios: P10 −30.8%, P50 +3.4%, P90 +54.6%, P(undervalued) 54.0%. A wide range, centered on zero.

Verdict: MONITOR — unchanged from July, but with the reasons reweighted. Credit risk, which was reason #2 for caution in July, has materially eased in light of the 10-Q. PBCO execution risk remains intact and unmitigated. A new, concrete risk has appeared (under-provisioning of the factoring book). The stock is close to normalized intrinsic value; it doesn't justify a firm conviction position at $25.38, but the qualitative thesis is more solid than it looked five weeks ago.

---

The business and its moat

Inherited in full from the reference analysis (nothing in the new filing contradicts the business structure), with two factual corrections I verified directly in the 10-K this time, plus updated segment data from the 10-Q.

Correction 1 — market position. The July analysis wrote "~18% market share, #2, after Wells Fargo Alaska 42.2% and First National Bank Alaska 22.4%" — a self-contradicting statement. The FDIC table reproduced in the 10-K (Summary of Deposits, 06/30/2025) is unambiguous: Northrim Bank has 17.5% of Alaska's bank deposits, with 20 branches and $2,812,725 thousand of deposits, in 3rd place, after Wells Fargo Bank Alaska (42.2%, 37 branches) and First National Bank Alaska (22.4%, 27 branches). Next come Key Bank 6.8%, First Bank 5.0%, Mt. McKinley 3.2%, Denali State 2.9% — seven commercial banks total in the state, $16.04 bn of deposits [10-K 2026-03-06, "Competition" section]. Sub-markets are far more concentrated: 64% in Nome, 23% in Juneau, 22% in Matanuska-Susitna and Sitka, 20% in Anchorage, but only 8-9% in Kodiak and Ketchikan.

Correction 2 — the research brief. The brief (the "Competitive position" section, sourced from aggregator umbrex.com, flagged there as unverified) gives the deposit share as "~15%, among the top 3." The primary-filing figure is 17.5%, and the rank is clearly 3rd. The brief correctly flagged its own uncertainty; its figure is off by ~250 bp. Likewise, the brief gives the dividend as "$0.68/share, 2.68% yield" (source stockanalysis.com) — the 10-Q's statement of changes in equity and cash flow give $0.16/quarter, i.e. $0.64 annualized and a yield of 2.52% at $25.38; the tracker independently gives 2.56%. The brief is ~16 bp too optimistic on yield.

How it makes money, by segment — figures from 10-Q Q2 2026, MD&A "Analysis of Business Segments":

Segment	Q2 2026 net profit	Q2 2025	Δ	H1 2026 net profit	H1 2025	Δ
Community Banking	$11,745 thousand	$7,743 thousand	+51.7%	$22,245 thousand	$18,531 thousand	+20.0%
Home Mortgage Lending	$1,607 thousand	$1,929 thousand	−16.7%	$2,693 thousand	$2,733 thousand	−1.5%
Specialty Finance	$1,990 thousand	$2,106 thousand	−5.5%	$4,079 thousand	$3,838 thousand	+6.3%
Total	$15,342 thousand	$11,778 thousand	+30.3%	$29,017 thousand	$25,102 thousand	+15.6%

Profit concentration is greater than the reference analysis suggested: all of Q2 2026's profit growth comes from Community Banking (+$4.0M), while the other two segments both declined. The "diversification across three segments" is more a revenue diversification than a profit one — 76.6% of quarterly profit comes from classic Alaska commercial banking.

Home Mortgage Lending. Sold production was $239.1M in Q2 2026 versus $249.7M in Q2 2025 (−4.2%); total Residential Mortgage production $222M vs $216M. Geographic expansion is genuinely advancing: the Arizona, Colorado and Pacific Northwest markets produced 27% of Q2 2026 production, versus 22% a year earlier. The servicing portfolio reached $1.66 bn. (6,657 loans), +7% year-over-year. Profit fell, though, from higher operating expenses ($8.1M vs $7.6M, mainly group medical costs) and lower servicing income. It's a segment growing geographically but not growing in profit — useful as a fee and deposit source, not as a value engine.

Specialty Finance. Average purchased-receivables and loan balances were $141.5M in Q2 2026 versus $124.1M in Q2 2025 (+14.0%), but the segment's net profit fell 5.5%. The implied yield on purchased receivables compressed: H1 2026 annualized revenue $25.2M on an average balance of ~$112.2M = ~22.5%, versus $25.8M on $101.0M = 25.5% in 2025 [10-K FY2025, "Purchased Receivables"; 10-Q Q2 2026]. Nearly 300 bp of yield compression in a year, on a book growing 20%. The reference analysis cited yields of "28-31%" from the 10-K — those were SCF's 2025 figures, including one-time fees; the 2026 trajectory is visibly downward. See the O'Glove chapter.

The moat, re-evaluated. It remains what it was: local knowledge, credit decisions made in Alaska, a cheap and sticky deposit base. Non-interest-bearing deposits reached 27.66% of total average deposits (vs 26.65% a year earlier), and the cost of interest-bearing deposits fell to 1.71% from 2.04% [10-Q, "Components of Net Interest Margin" table]. The average loan/average deposit ratio rose from 78.54% to 83.71% — the bank is putting its cheap deposits to harder work, which is the real source of margin expansion. What the new filing doesn't change: the moat is regional and non-transferable. In Oregon (PBCO) and in nationwide factoring (SCF), Northrim competes without the advantage sustaining its Alaska margins. The combined company's ~$4.2 bn of assets will contain ~18.5% assets from a market where the franchise has no history.

---

Management and capital allocation

Inherited: the Schierhorn → Huston transition, the PBCO deal terms, the buyback pause, the subordinated debt issuance. Re-derived from the 10-Q and DEF 14A: the real cost of the subordinated debt, the post-issuance regulatory capital structure, and quantifying the buyback contrarianism (the reference analysis mentioned it qualitatively, without a price).

Buybacks — quantified for the first time. The company repurchased 834,692 shares in 2023, 60,136 in 2024, zero in 2025, and the board "has not presently authorized any repurchases of its common stock for 2026"; at 12/31/2025 no shares remained available under the program [10-K FY2025, "Repurchase of Securities"]. Placed alongside the XBRL cash outflows (PaymentsForRepurchaseOfCommonStock: $9.0M in 2023, $789 thousand in 2024, $0 in 2025), this yields an average repurchase price of ~$10.79/share in 2023 (post-adjustment for the 4:1 split) — the year of the regional bank panic. The stock is at $25.38 today: +135% versus the price the bank bought at, at a moment almost nobody was buying regional banks. And the program's complete halt in 2025-2026, with the stock at 1.90× tangible book value, is the other correct half of the same discipline. This is the strongest evidence of rational capital allocation in the entire file.

Subordinated debt — the real cost. The $60.0M of Fixed-to-Floating Subordinated Notes issued in November 2025 carry 6.875% fixed until December 1, 2030, then 3-month SOFR + 3.48%, maturing December 1, 2035; $1.4M of issuance costs amortize to maturity [10-Q, Note 8]. The cost in Q2 2026 was $1.0M, and $2.1M in H1 2026 — versus zero in H1 2025. The effect is visible in the margin table: the average cost of borrowings jumped from 4.14% to 6.08% (+194 bp) year-over-year. In other words, the +29 bp margin expansion happened despite a funding headwind already fully absorbed, not ahead of it. On a full-year basis, the run-rate is ~$4.1M pre-tax; the TTM used as the owner-earnings base only includes ~$2.5M, so the $57.44M base is ~$1.2M after-tax too generous if we look purely at cost — but the $60M was also invested in interest-earning assets, so what matters is the spread, not the gross cost. I leave it unmodified and note it as a reverse-conservatism element.

What that capital bought. The Company's regulatory ratios at 06/30/2026: total risk-based capital 14.46% (minimum 8.00%, "well-capitalized" threshold 10.00%), Tier 1 risk-based 11.26%, CET1 10.90%, leverage 9.32%. The Bank (the operating entity) sits at 13.39% / 12.38% / 12.38% / 10.26% [10-Q, "Capital Requirements and Ratios"]. The 10-Q explains the gap: the $10M of trust preferred plus the $60M of subordinated notes are regulatory capital at the Company level but not the Bank level — $70M of additional regulatory capital. The issuance moved the company ~107 bp on total risk-based capital and produced a measurable collateral benefit: FDIC insurance expense fell "due to improved regulatory capital ratios", by $937 thousand for the half-year (from $1,773 to $836 thousand). In other words, part of the subordinated debt's cost self-funds from the insurance premium saved.

The PBCO transaction — reread through the capital-allocation lens. Note 15 of the 10-Q confirms the terms unchanged: 1.160 NRIM shares per PBCO share, a two-step structure (PBCO → Whitewater Sub → Company), a subsequent bank merger, PBCO shareholders reaching ~21% of the company, a combined company of ~$4.2 bn of assets, approvals needed from the Fed, FDIC, Oregon DCBS and Alaska DCCED plus both shareholder meetings. The capital-allocation element: Northrim is paying with its own paper, valued at 1.90× tangible book, for a bank valued at ~1.67× book value ($167.3M / $100.2M of PBCO equity). This is the correct arithmetic of a disciplined buyer — you use expensive currency to buy cheaper assets. The risk isn't price, it's execution: system conversion is scheduled only for Q4 2027, i.e. 12+ months after the estimated closing (Q4 2026 / early Q1 2027), an unusually long window in which the two banks stay operationally separate.

The dividend. $0.16/quarter, unchanged Q1 → Q2 2026; $7,117 thousand paid in H1 2026 versus $7,069 thousand in H1 2025 (+0.7%). Payout on H1 profit: 24.5%. Yield 2.52% at $25.38. There's no debt contracted to fund distributions — the $60M went explicitly into regulatory capital and growth initiatives, and distributions are fully covered by profit.

Skin in the game — the weak point, unchanged. All named directors and executive officers (16 people) held 536,875 shares = ~2.4% of the equity combined as of March 31, 2026 [DEF 14A 2026-04-14]. There is no controlling shareholder, no founding family. The only reported institutional holder above 5% is Dimensional Fund Advisors (~6.1%), a passive quant fund, not a strategic investor.

---

What changed in the last 4 quarters (Q3 2025 → Q2 2026)

Inherited: the high-level quarterly evolution table and the explanation of the Q3 2025 one-off gain. Fully re-derived from the 10-Q and data pack: the balance sheet position by position, which in the reference report appeared only aggregated.

Balance sheet, position by position: 12/31/2025 → 06/30/2026 (thousand USD, 10-Q Q2 2026)

Position	06/30/2026	12/31/2025	Δ	Explanation
Cash and due from banks	41,948	36,042	+16.4%	Normal operating fluctuation
Interest-bearing deposits at other banks	130,265	109,864	+18.6%	Parked liquidity; average balance jumped from $27.2M to $85.1M year-over-year, but at a falling yield (7.60% → 3.69%)
Trading equity securities	11,315	8,392	+34.8%	$3,016 thousand of purchases in H1; small position
Available-for-sale investment securities	411,644	420,661	−2.1%	Purchases of $127.3M vs maturities/calls of $135.5M — the portfolio rolls into higher yields (3.07% → 3.79%)
Held-to-maturity securities	31,750	26,750	+18.7%	$15.0M purchased, $10.0M matured
FHLB stock	7,187	6,764	+6.3%	Proportional to FHLB borrowings
Loans held for sale	83,272	100,323	−17.0%	The mortgage warehouse drained: $410.7M of sales vs $362.5M of originations in H1. The most important item for the "FCF bridge"
Loans (portfolio)	2,386,328	2,295,499	+4.0%	+$90.8M in six months (~8% annualized); the organic growth is real
Allowance for loan losses (ACL)	(25,461)	(23,737)	+7.3%	The reserve grows faster than the portfolio: ACL/loans 1.067% vs 1.034%
Purchased receivables, net	122,792	101,642	+20.8%	The fastest-growing item on the balance sheet — see red flags
Mortgage servicing rights, at fair value	28,475	27,474	+3.6%	Additions $2,056 thousand, minus a fair-value change of $1,055 thousand
Other real estate owned (OREO)	1,224	—	new	The first OREO in the analyzed period; a $1,224 thousand transfer from loans in H1 2026
Net fixed assets	41,486	39,692	+4.5%	New Palmer, AK branch (the 21st); H1 capex $3,474 thousand vs $524 thousand in H1 2025
Right-of-use assets (operating leases)	11,263	5,911	+90.5%	New non-cash lease liability of $6,325 thousand in H1 2026 (zero in H1 2025) — renewals and/or new locations
Goodwill	49,874	49,874	0%	Unchanged (SCF, 2024); no impairment test triggered
Other intangibles, net	950	950	0%	Unamortized in the period
Other assets	81,074	84,172	−3.7%	A $4,530 thousand decline in the cash flow statement
Total assets	3,415,386	3,290,273	+3.8%
Demand deposits (non-interest-bearing)	826,271	721,925	+14.5%	The best news on the liability side: free funding grows twice as fast as the balance sheet
Interest-bearing demand deposits	1,281,777	1,242,546	+3.2%
Savings	246,617	250,006	−1.4%
Money market	194,665	195,793	−0.6%
Certificates of deposit	369,458	402,759	−8.3%	The most expensive funding (2.78%) is contracting — the source of the repricing
Total deposits	2,918,788	2,813,029	+3.8%
Borrowings	81,574	81,729	−0.2%	$10.3M trust preferred + $58.7M net subordinated + rest
Operating lease liabilities	11,459	5,941	+92.9%	The mirror of the right-of-use assets
Other liabilities	55,951	63,030	−11.2%	A $9,054 thousand decline in the cash flow statement
Equity	347,614	326,544	+6.5%	Retained earnings $331,394 thousand (+7.0%); AOCI went from +$619 to −$98 thousand

The four large swings and their explanations. (1) Non-interest-bearing deposits +14.5% — the most valuable move on the entire balance sheet; it explains why the margin is widening even with the rising cost of borrowings. (2) Certificates of deposit −8.3% — the favorable repricing is visible here, at a 2.78% average cost versus 3.21% a year earlier. (3) Purchased receivables +20.8% — the only asset growth that outpaces its associated revenue growth by a factor of 4.5×. (4) Leases +91% — not a quality issue (it's non-cash, $6.3M), but it signals new fixed commitments ahead of a merger that will bring another 11 branches.

Margins, NIM and cash conversion

Indicator	Q2 2026	Q2 2025	Δ	Source
Net interest income	$37,136 thousand	$33,592 thousand	+10.6%	10-Q, income statement
Reported NIM	4.96%	4.66%	+30 bp	10-Q, MD&A
NIMTE (fully tax equivalent)	5.01%	4.72%	+29 bp	10-Q, margin table
Yield on interest-earning assets	6.35%	6.27%	+8 bp	ibid.
— of which loan yield	6.94%	6.99%	−5 bp	ibid.
— of which long-term taxable securities	3.79%	3.07%	+72 bp	ibid.
Cost of interest-bearing liabilities	1.88%	2.12%	−24 bp	ibid.
— of which interest-bearing deposits	1.71%	2.04%	−33 bp	ibid.
— of which borrowings	6.08%	4.14%	+194 bp	ibid.
Average loans / average interest-earning assets	79.29%	75.19%	+410 bp	ibid.
Non-interest-bearing deposits / total deposits (average)	27.66%	26.65%	+101 bp	ibid.
Efficiency ratio (cost/revenue)	59.44%	64.7%¹	improvement	8-K Q2 2026
Total operating expenses	$32,022 thousand	$32,488 thousand	−1.4%	10-Q
Total other operating income	$16,737 thousand	$16,640 thousand	+0.6%	10-Q

¹ The efficiency comparative is Q4 2025's (the press release doesn't give Q2 2025 on the same basis).

Margin decomposition — the answer to the "unverifiable" falsifier. The delta-triage pack flagged with a ? the claim "NIM expansion is the real growth engine," because the mechanical check needed two consecutive quarters of revenue growth and was missing one. The 10-Q's margin table answers directly, no proxy needed: of the +$4,456 thousand increase in pre-tax profit (Q2 2026 $20,224 vs Q2 2025 $15,768), +$3,544 thousand (79.5%) comes from net interest income, +$466 thousand from lower expenses, +$349 thousand from a lower provision, and only +$97 thousand from other operating income. The falsifier HOLDS, verified directly in the filing.

More important is what the expansion is made of. The reference report's hypothesis was that it's almost entirely deposit repricing — hence transitory. The decomposition shows otherwise: the cost of liabilities fell 24 bp (a weighted ~17 bp contribution to NIM), but asset yield rose 8 bp, and the mix shifted 410 bp toward loans. And the rise in asset yield doesn't come from loans (whose yield fell 5 bp) — it comes from the securities portfolio rolling from 2020-2021 coupons into current yields (+72 bp, from 3.07% to 3.79%, on an average balance of $450.4M). This is a tailwind independent of what the Fed does with the policy rate: it continues as long as low-coupon old paper remains in the portfolio. Conclusion: margin expansion is roughly 60% cyclical (deposits) and 40% structural/mix — less fragile than the reference Pre-mortem assumed.

---

Balance sheet analysis — Quality of Earnings (Thornton O'Glove method)

NEW chapter — the 2026-07-28 reference report didn't contain it (it had 8 ## chapters, not 10). Written entirely from filings: 10-Q Q2 2026, 10-K FY2025, 10-K FY2024. O'Glove for a bank requires translating the categories: "receivables" become loans and accrued interest receivable, "inventory" becomes loans held for sale and purchased receivables, and the central test becomes provision vs. actual losses.

1. "Receivables vs. revenue" — loans and accrued interest

Metric	H1 2026	H1 2025	Δ
Portfolio loans (end of period)	$2,386,328 thousand	—	+4.0% vs 12/31/2025
Interest and fees on loans	$82,305 thousand	$77,989 thousand	+5.5%
Loan yield (Q2, tax equivalent)	6.94%	6.99%	−5 bp
Growth in accrued interest receivable	$1,968 thousand	$1,072 thousand	+83.6%
Total interest and dividend income	$92,050 thousand	$86,360 thousand	+6.6%

The classic O'Glove test — "receivables growing much faster than revenue" — doesn't trigger at the loan level: the portfolio grew 4.0% in six months while loan revenue grew 5.5% year-over-year, with a practically flat yield. There's no revenue booked without economic substance in the loan book.

A finer signal does trigger, though: the growth in accrued interest receivable was 83.6% larger than the same period last year, while interest revenue grew only 6.6%. In absolute terms it's small ($1.97M on interest revenue of $92.1M, i.e. 2.1%), but the direction is exactly what O'Glove tracks: booked revenue not yet converted into cash, accelerating faster than revenue itself.

The correlated signal, and the most directly O'Glove-type in the whole file: $468,000 of interest on nonaccrual loans was reversed through the income statement in Q2 2026 — versus zero in Q2 2025 and zero in H1 2025 [10-Q, Note 3]. This is exactly the "profit is an opinion" pattern: revenue recognized in prior quarters, withdrawn now. The effect is small ($468 thousand pre-tax ≈ 0.6% of Q2 profit) and already in the reported figures — but it's the first time in the cycle this has appeared.

2. "Inventory vs. COGS" — the mortgage warehouse and purchased receivables

Loans held for sale (the mortgage segment's inventory): $83,272 thousand at 06/30/2026 versus $100,323 thousand at 12/31/2025, −17.0%. Against sold production of $239.1M in Q2, turnover is ~31 days. Inventory is falling while sales continue — the opposite of a bad signal. No channel inflation.

Purchased receivables (the factoring segment's inventory) — this is where the problem is:

Metric	06/30/2026	12/31/2025	12/31/2024
Purchased receivables, net	$122,792 thousand	$101,642 thousand	$74,100 thousand
Growth	+20.8% in 6 months	+37.1% in 2025	—
Associated revenue (period)	$12,605 thousand (H1)	$25,800 thousand (year)	$7,100 thousand (year)
Revenue growth	+4.6% vs H1 2025	+263% (full SCF year)	—
Implied annualized yield	~22.5%	25.5%	~18.3%
ACL — purchased receivables	$610 thousand	$0	$3,649 thousand
ACL / purchased receivables	0.49%	0.00%	4.92%
Net charge-off / average balance (year)	—	2.16%	0.00%

This is a textbook divergence: the asset grows 20.8% in six months, the revenue it produces grows 4.6%, the yield compresses ~300 bp, and the loss reserve was zero for all of fiscal 2025 and all of Q1 2026, rebuilt only in Q2 2026 to 0.49% — on a book that posted net charge-offs of 2.16% in 2025. A year earlier, at 06/30/2025, the same book carried a reserve of $3,432 thousand on a smaller balance, i.e. coverage on the order of 3.8-4.0%. Coverage fell roughly eight-fold.

The counter-argument, for fairness. The $3,649 thousand of reserve as of January 1, 2025 was largely an acquired credit mark from the Sallyport acquisition (10/31/2024), related to an identified PCD pool. In 2025, $1,513 thousand was reversed against a liability to sellers (contractually owed collections, with no income-statement benefit) and $2,211 thousand was actually charged off. So the move from 4.92% to 0% is substantially the normal mechanics of an acquisition-mark run-off, not a deliberate relaxation of provisioning standards. And the Q2 2026 rebuild ($625 thousand provision) is explained by management as caused by "an increase in the quantitative allowance ... on pooled purchased receivables due to an increase in concentration of these assets" [10-Q, MD&A] — i.e. the model reacted to concentration, which is the correct behavior.

My verdict on this point: the signal stays valid, but it's a forward-looking signal, not retrospective manipulation. The book with the highest yield and the highest historical loss rate in the company ($122.8M, +20.8%, ~22.5% yield, 2.16% realized losses last year) is covered today at 0.49%. If the 2025 loss rate repeats on the current balance, the annual loss would be ~$2.65M against a $0.61M reserve — a difference of ~$2.0M pre-tax, ~$1.5M after-tax, i.e. 2.7% of TTM adjusted profit. Not existential. To watch at Q3.

3. Debt — gross, net, maturity schedule, cost, covenants

Item	Balance 06/30/2026	Rate	Maturity
Junior subordinated debentures (Trust 2)	$10,310 thousand	90-day SOFR + 0.26% + 1.37% = 5.30%, swap-fixed at 3.72% on $10.0M	March 15, 2036
Subordinated Notes 2035	$58,684 thousand (net of costs)	6.875% fixed until 12/01/2030, then 3-month SOFR + 3.48%	December 1, 2035
Other borrowings	~$12,580 thousand	—	—
Total borrowings	$81,574 thousand	average cost 6.08% (Q2 2026)	—

12-24 month maturity schedule: empty. Nothing matures before December 2035. Refinancing risk in the relevant window is zero — a remarkably clean position for a small bank making an acquisition. The first rate reset isn't until 12/01/2030.

3-year evolution. Total debt (LongTermDebt tag, data pack): $23.0M (12/31/2024) → $81.7M (12/31/2025) → $81.6M (06/30/2026). The jump is entirely the November 2025 issuance. Against equity of $347.6M, holding-company-level leverage is 23.5% — modest.

Debt used for dividends or buybacks? No. The $60M went into regulatory capital (Total RBC 14.46%, +107 bp) and growth initiatives. H1 2026 distributions ($7,117 thousand) are covered 4.1× by half-year profit. No shares were repurchased in 2025 or 2026.

Covenants. The subordinated notes are "unsecured, subordinated obligations" with acceleration only in limited bankruptcy/insolvency cases — no financial maintenance covenants. The real constraint is regulatory: a 2.5% conservation buffer above minimums, below which dividends, buybacks and discretionary bonuses are restricted. At 14.46% Total RBC versus 10.5% (8.0% + 2.5%), the margin is ~396 bp.

4. Discretionary expenses — what's falling and what's rising abnormally

Expense (thousand USD)	Q2 2026	Q2 2025	Δ	H1 2026	H1 2025	Δ
Salaries and personnel	20,784	20,854	−0.3%	40,290	38,077	+5.8%
Data processing	3,482	3,366	+3.4%	6,787	6,470	+4.9%
Occupancy (rent/utilities)	1,825	2,104	−13.3%	3,929	3,993	−1.6%
Professional and outside services	1,428	1,113	+28.3%	2,587	2,228	+16.1%
Marketing	686	1,042	−34.2%	1,587	1,714	−7.4%
Compensation — Sallyport acquisition payments	500	600	−16.7%	1,000	1,200	−16.7%
Insurance (including FDIC)	432	756	−42.9%	836	1,773	−52.8%
OREO expenses, net	102	2	×51	114	5	×23
Other expenses	2,783	2,651	+5.0%	5,514	5,199	+6.1%
Total	32,022	32,488	−1.4%	62,644	60,659	+3.3%
Fixed-asset amortization (cash flow)	—	—	—	1,680	1,780	−5.6%
Capex (fixed-asset purchases)	—	—	—	3,474	524	×6.6

Three O'Glove observations, in order of severity:

(a) Amortization falls 5.6% while capex rises 6.6-fold and the fixed-asset base grows 4.5%. This is the classic pattern O'Glove tracks — falling amortization expense on a growing asset base. A benign and probable explanation is available: the Palmer branch went live during Q2 2026, so new assets contribute less than a quarter's amortization, while fully depreciated old assets exit the base. The profit effect is small (~$100 thousand for the half-year). Not a red flag today; something to watch if the divergence persists into Q3-Q4 — if amortization stays flat at an annualized capex of ~$7M, the question about useful lives becomes legitimate.

(b) Marketing −34.2% for the quarter. A real discretionary cut, confirmed by the MD&A as one of the causes of lower Community Banking expenses. Not large ($356 thousand, 0.8% of the quarter's pre-tax profit), but exactly the category O'Glove calls "borrowing from the future": deferred customer-acquisition spending is paid for in tomorrow's growth. In a quarter where the bank is opening a new branch and preparing to integrate 11 branches in Oregon, cutting marketing is counterintuitive.

(c) Insurance −52.8% for the half-year — NOT a discretionary cut. The MD&A explains it twice, explicitly: "a decrease in FDIC insurance expense due to improved regulatory capital ratios." It's the mechanical consequence of the subordinated debt issuance, which raised capital ratios and lowered the FDIC assessment. Structural and repeatable, not borrowed from the future. I classify it as a real improvement, not a red flag — and it's important to separate it, because the data pack's automated check ("SG&A +5.6% vs. revenue +2.9%") can't make this distinction.

(d) An ambiguous signal: Community Banking salaries fell $942 thousand in Q2, from "lower group medical claims expense and lower accruals for profit sharing and related taxes." Lower profit-sharing accruals in a quarter where segment profit grew 51.7% is a combination that deserves a question. Group medical expenses are genuinely volatile at a bank with ~450 employees and plausibly explain a good part of it; the profit-sharing accrual depends on an annual formula that can adjust quarterly in either direction. I note it as ambiguous, not manipulation.

(e) Professional services +28.3% — the start of consulting costs for the PBCO merger. Expected to accelerate in Q3-Q4 2026.

Accounting estimate changes: the 10-Q explicitly states "There have been no material changes to the valuation techniques or assumptions within the models that affect our estimates during the second quarter of 2026." The ACL sensitivity analysis confirms the model's robustness: under long-term-average unemployment and GDP assumptions, the ACL would rise by only $2.2M (9%). There's no hidden fragility in the model.

Effective tax rate: 24.00% in H1 2026 versus 24.72% in H1 2025 (Q2: 24.14% vs 25.30%). A decline of ~72-116 bp, contributing ~$0.27M to half-year profit — under 1%. Not a source of EPS growth.

5. Summary — net profit vs. CFO (accruals) divergence over 3 years

Year	Net profit	CFO	CFO / Net profit
2023	$25.4M	$38.8M	1.53×
2024	$37.0M	−$8.7M	−0.24×
2025	$64.6M	$139.3M	2.16×
H1 2026	$29.0M	$67.3M	2.32×

The data pack's Sloan ratio — accruals (NI − CFO) / assets = −3.7% — formally looks very good (negative = cash above profit). I refuse to use it as evidence of earnings quality, and it's important to say why: at a bank with a mortgage-origination arm, CFO swings entirely with the loans-held-for-sale warehouse. In 2024 CFO was negative ($−8.7M) because the warehouse was filling up; in 2025 it was $139.3M because it was draining. Neither figure says anything about earnings quality. An investor who read −0.24× in 2024 as "catastrophic accruals" and 2.16× in 2025 as "exceptional quality" would have drawn exactly the wrong conclusions.

The valid accruals test for a bank is provision versus actual losses:

Period	Loan provision	Net charge-offs	Ratio	ACL / loans
H1 2026	$2,047 thousand	$323 thousand	6.3×	1.067%
H1 2025	$671 thousand	$106 thousand	6.3×	1.010%
Q2 2026	$760 thousand	$111 thousand	6.8×	—

The bank is provisioning more than six times the losses it actually suffers, and the reserve is growing as a percentage of the portfolio (1.034% → 1.067%) while the portfolio itself grows 4%. This is the exact opposite of reserve harvesting. Reported profit is not inflated through under-provisioning at the loan-book level.

An important nuance, cutting both ways: of the $22,982 thousand of nonaccrual loans, $19,281 thousand (83.9%) carries no specific provision, and the ACL on nonaccrual is only $1,339 thousand. At 12/31/2025 the unprovisioned proportion was 53.4% ($6,392 of $11,968 thousand). CECL methodology permits this when collateral fully covers the exposure — and the single borrower's three loans are described as "well-collateralized" (commercial real estate plus commercial loans). But concentrating 84% of nonperforming loans in the "no reserve, relying on collateral" category means a single unfavorable collateral re-appraisal could produce a large point-in-time provision. This is the most concrete short-term risk channel.

Quality of Earnings verdict: MEDIUM

Arguing for HIGH: provisioning at 6.3× actual losses; the loan-loss reserve rising as a percentage of the portfolio; the criticized-loan pool down 20%; potential problem loans down 46%; an empty 24-month debt maturity schedule; zero debt contracted for distributions; no accounting policy or estimate change; a stable tax rate; a draining mortgage warehouse; an ACL resilient to sensitivity testing ($2.2M under long-term-average assumptions).

Arguing for LOW: 0.49% coverage on the factoring book that lost 2.16% last year and is growing 20.8%; 83.9% of nonaccrual loans with no specific provision; the cycle's first interest reversal ($468 thousand); falling amortization with capex ×6.6; marketing cut 34%; reduced profit-sharing accruals in a quarter with 52% profit growth.

MEDIUM is the honest verdict: reported profit is real and backed by cash where cash matters (margin, deposits, loan provisioning), but there are three points where accounting optimism is visible, all concentrated in the credit area, and all quantifiable at 2-5% of annual profit — not 30%.

Signal to watch at the October 28, 2026 report (Q3): (1) the ACL on purchased receivables — if it stays under 1.0% with the balance above $130M, the signal worsens; (2) whether the single borrower's three loans stay "no ACL" or receive a specific provision — a provision allocation on them would confirm the collateral was reappraised downward; (3) whether fixed-asset amortization stays below $900 thousand/quarter with capex at the current pace.

---

The CEO profile — Outsider traits (William Thorndike method)

NEW chapter — the reference report didn't contain it. Written from the DEF 14A (2026-04-14), 10-K FY2025 and 10-Q Q2 2026.

Who he is. Michael G. Huston, 58. Joined the bank in May 2017 as Executive Vice President and Chief Lending Officer, after 25 years at First Interstate Bank (last role: EVP and Chief Banking Officer). Promoted to President of the Bank in March 2022, then President, CEO and COO of the Company on April 6, 2024. Since January 1, 2026 he also holds the Chairman role at both the Company and the Bank, following Joseph Schierhorn's retirement.

A methodological caveat, stated before the grid: Huston has been CEO for roughly 2 years and 5 months. The Thorndike grid measures a capital-allocation pattern over 10-20 years. Any score on such a short tenure is an extrapolation, and the strongest evidence of Outsider behavior in the file — the 2023 contrarian buybacks — occurred under Schierhorn, not Huston. What follows is an evaluation of the institution and of Huston's early decisions, not a verdict on the man.

1. Capital allocation as priority #1 — mixed evidence, leaning favorable. The three major capital decisions of Huston's tenure are internally coherent and logically sound: (a) fully halting buybacks in 2025 and authorizing none for 2026, with the stock at 1.90× tangible book value; (b) issuing $60M of subordinated debt at 6.875% in November 2025, which brought $70M of additional regulatory capital at the holding level and collaterally lowered the FDIC premium by ~$1.9M a year; (c) the PBCO acquisition, paid entirely in stock — own currency at 1.90× TBV for an asset at ~1.67× book value. All three are cost-of-capital arbitrage decisions, not growth-for-growth's-sake decisions. I found no explicit stated allocation framework in communications (a return threshold, a hierarchy of uses), though — so it's observed coherence, not declared doctrine.

2. Contrarian buyback — YES at the institutional level, untested at Huston's level. 834,692 shares repurchased in 2023 ($9.0M, an implied average price of **$10.79/share** post-split) — the year of the regional bank collapse. 60,136 shares in 2024 (~$789 thousand). Zero in 2025 and zero authorized for 2026, with the program exhausted. The stock is at $25.38 today. This is exactly the pattern Thorndike looks for: aggressive when the paper is cheap, fully stopped when it's not. The antithesis of mechanical buying at highs. Credit for the 2023 decision belongs to the prior management; credit for not resuming the program at today's prices belongs to Huston and the current board, and is equally important.

3. Decentralization and frugality — strong. The 10-K states that all Community Banking segment credit decisions are made in Alaska, by a leadership team "intimately involved with serving customers and making credit decisions." In the PBCO deal, the People's Bank branches take the Northrim name but remain managed by the current staff, with Julia Beattie (PBCO CEO) as market president for Oregon and one PBCO director on each board. On frugality, the numbers are clear: Huston's total 2025 compensation was $1,491,789 — 2.3% of the company's net profit and 0.26% of market cap. Company-wide stock compensation was $1.7M in 2025, i.e. 2.6% of profit. No options were granted in 2023, 2024 or 2025. The Company's board met eight times in 2025. No corporate jet, no exotic compensation structures.

4. Independence from Wall Street — probable, but weakly documented. The stock is covered by only 2-4 analysts (KBW only initiated coverage in July 2026, with "Market Perform" and a $30 target). I found no company-issued quarterly forecast in the filings. Press releases systematically report both GAAP figures and adjusted reconciliations, and they openly reported the $14.5M one-off PWA gain instead of blending it into "core profit." A positive signal, but with low informational volume — a company with 2 analysts doesn't have much pressure to manage.

5. Skin in the game — WEAK, the clearest unfavorable point. All 16 named directors and executive officers held 536,875 shares = ~2.4% of the company at 03/31/2026. There's no controlling shareholder. Huston's 2025 compensation structure: salary $572,039, stock $358,133 (24.0% of total), cash profit sharing $386,126, SERP contribution 20% of base salary, 401k $21,000, deferred compensation plan $117,700. In other words, over 60% of pay is short-term or deferred cash, under a quarter is equity. Long-term alignment is weak by design. Progression: total compensation $682,997 (2023) → $1,055,960 (2024) → $1,491,789 (2025), doubling in two years, though tracking his promotion to CEO and the company's doubled profit. The 2025 advisory "say-on-pay" vote passed with 98%.

6. Proven contrarian behavior — partial. The best case for Huston isn't a buyback, it's the PBCO deal structure: he chose to pay entirely in stock at a moment his own paper is at 1.90× TBV, instead of using cash or debt. This is the decision a disciplined allocator makes and an empire-builder avoids (because it dilutes). Counter: it's the company's first interstate acquisition ever, made 20 months after taking office, with system conversion scheduled only 16+ months after closing — a riskier execution profile than an orthodox Outsider would choose.

Outsider score: 3 / 5

The institution has the right DNA — a proven contrarian buyback, real frugality (CEO at 2.3% of profit, zero options), authentic decentralization, paying in expensive paper for cheap assets. What's missing for a higher score: weak skin in the game (2.4% for 16 people, under a quarter of compensation in stock), a declared capital-allocation framework, and — decisively — time. Two and a half years in the role, with the tenure's defining decision (PBCO) still unclosed and unintegrated.

Key-person risk

Role concentration, not succession. Since January 1, 2026, Huston simultaneously holds Chairman + President + CEO + COO of the Company and Chairman + President + CEO of the Bank — seven titles, one person, at a $3.4 bn-asset company executing its first interstate expansion. Counterweights exist and are real: every other board member is independent, and John C. Swalling is the lead independent director with explicit duties (chairing executive sessions of non-employee directors, leading annual director self-assessments, recommending committee composition). The board makes its case for this choice in the proxy. But concentration remains concentration, and its timing — simultaneous with PBCO — is the least favorable possible.

A second critical person, less obvious: Mark Edwards, Chief Credit Officer, the only named executive whose contract provides smaller severance (1× salary, not 2×, on change of control; 9 months of benefits, not 12). At a company where the central near-term risk is credit migration and where 84% of nonaccrual loans are unprovisioned on the basis of collateral valuation, continuity in the credit function matters disproportionately.

Related parties: the proxy explicitly states there are no family relationships among directors, nominees or executives. I identified no significant related-party transactions.

---

Accounting red flags

Inherited and confirmed: the "302.9% dilution" artifact (the 4:1 split on 09/18/2025 — the real diluted share count rose from 22,446,232 to 22,544,448 over 12 months, i.e. +0.44%); the non-recurring nature of the $14.5M PWA gain. Re-evaluated or new: points 2, 3, 5 and 6.

1. The 302.9% "dilution" in the data pack is an artifact — CONFIRMED. A 4-for-1 forward split, effective September 18, 2025 (authorized shares 10M → 40M, par value $1.00 → $0.25). Instruments comparing the unadjusted pre-split balance (~5.5M) with the adjusted post-split one (22.2M) misreport. Real organic dilution is under 0.5%/year. The real future dilution to watch remains: +21% from PBCO.

2. Credit quality deterioration — DOWNGRADED in severity. The reference report classified it as the second reason for caution, based on the single figure then available (NPA net of guarantees $11.9M → $23.0M over 12 months, ACL/NPL coverage 290% → 117%). The 10-Q tables give the full picture and it's better:

Metric	06/30/2026	12/31/2025	12/31/2024	Direction
Criticized loans (total)	$68,996 thousand	$86,318 thousand	—	↓ −20.1%
Criticized loans, net of govt. guarantees	$42,606 thousand	$43,979 thousand	—	↓ −3.1%
Criticized / portfolio	2.89%	3.76%	—	↓
Potential problem loans	$11.5M	$21.2M	$1.6M	↓ −45.8%
Adversely classified assets, net (Community Banking)	$28.9M	$29.4M	$6.3M	↓ (and −10% vs $32.1M at 06/30/2025)
Nonaccrual loans (gross)	$22,982 thousand	$11,968 thousand	—	↑ +92.0%
Nonaccrual net of guarantees	$21,811 thousand	$11,968 thousand	$7,533 thousand	↑
— of which no ACL	$19,281 thousand (83.9%)	$6,392 thousand (53.4%)	—	↑
Net charge-offs (half-year)	$323 thousand	—	—	0.014% of the portfolio
ACL / loans	1.067%	1.034%	—	↑

The correct reading: the problem-loan cohort formed in 2025 (adversely classified from $9.6M to $33.5M, potential problem loans from $1.6M to $21.2M) and peaked at 12/31/2025. In H1 2026 that cohort is resolving: three loans of a single borrower migrated to nonaccrual, the rest paid off. The total problem pool is shrinking. Doubling NPAs is the lagging indicator of already-consumed deterioration, not the leading indicator of new deterioration. The leading indicators (special mention, potential problem, adversely classified) are all falling.

What still deserves worry: concentration. A single borrower, three loans, ~$10.5M — 45% of net nonperforming loans. And 83.9% of nonaccrual has no specific provision, relying exclusively on collateral valuation.

3. NEW — under-provisioning of the purchased receivables book. Detailed in the O'Glove chapter. In short: a $122.8M balance (+20.8% in six months), $610 thousand ACL (0.49%), zero for all of 2025 and Q1 2026, against net charge-offs realized at 2.16% in 2025 and coverage of ~3.8-4.0% a year earlier. Mitigating: the early-2025 reserve was largely an acquired credit mark from SCF, run off through normal PCD mechanics. Quantified exposure if 2025's loss rate repeats: ~$2.0M pre-tax, ~2.7% of TTM adjusted profit.

4. Falling effective tax rate — negligible effect, reconfirmed. 24.00% (H1 2026) vs 24.72% (H1 2025). Contributes ~$0.27M for the half-year, under 1% of profit. Not a material EPS growth source. (The data pack's automated check reports the reverse — "24.1% now vs. 21.7% four quarters ago," a rise — because it uses TTM windows distorted by the PWA gain. The income statement figures from the 10-Q are the correct ones.)

5. NEW — the cycle's first interest reversal. $468,000 of accrued interest on nonaccrual loans reversed through the income statement in Q2 2026, versus zero in Q2 2025 and zero in H1 2025 [10-Q, Note 3]. Small in absolute value, but it's the pure O'Glove signature: revenue previously recognized, withdrawn now.

6. NEW — amortization falling on a growing asset base. Fixed-asset amortization $1,680 thousand in H1 2026 vs $1,780 thousand in H1 2025 (−5.6%), while net fixed assets grew 4.5% and capex grew 6.6-fold ($3,474 thousand vs $524 thousand). A benign explanation is available (the Palmer branch went live during the quarter). To be reconfirmed at Q3.

7. The $14.5M PWA gain remains non-recurring — reconfirmed. Any extrapolation of GAAP growth 2024→2025 (+74.8%) overstates earning power. The tracker uses TTM GAAP EPS of $3.04, which includes the gain: that's why it shows a 28.8% discount to the Graham Number. On adjusted EPS ($2.62 estimated for 2026) the Graham Number is $30.35 and the discount is +19.6% — still positive, but 9 points smaller.

8. What is NOT a red flag, despite appearances. The data pack's Sloan accruals ratio (−3.7%) and 33.0% FCF yield are both inapplicable to this company, for the same reason: CFO is dominated by the loans-held-for-sale warehouse. Detailed in the valuation chapter.

---

Triangulated valuation

The FCF bridge and the chosen definition

A targeted check in the consolidated cash flow statement (10-Q Q2 2026, page 8). The filer is US GAAP, and interest paid appears in operating: "Interest paid | $20,185 | $21,590" in the supplemental information, and the XBRL InterestPaidNet confirms it ($42,555 thousand for FY2025). There's no interest in investing or financing, no lease-principal lines (all leases are operating; the $6,325 thousand non-cash lease liability appears only in supplemental information). There are no minority interests. So the AFYA-type error doesn't apply here — there's nothing additional to subtract, and the data pack is right that the divergence between definitions is under 10%.

The problem is different, and more serious: the metric itself measures nothing at this company.

Term (H1 2026, thousand USD)	Value	Source
Net cash from operating activities (CFO)	67,298	10-Q, Cash Flow Statement
— of which: proceeds from sale of loans held for sale	+410,668	ibid.
— of which: origination of loans held for sale	−362,522	ibid.
— net contribution of the mortgage warehouse	+48,146	calculated
— remaining CFO, excluding the warehouse	19,152	calculated
Capex (fixed-asset purchases)	−3,474	ibid.
"Simple FCF" (CFO − capex)	63,824	calculated
Same-period net profit	29,017	10-Q, income statement

71.6% of reported CFO is the mortgage inventory movement — working capital, not a flow to the shareholder. Annualized, "simple FCF" would be ~$128M at a company earning $58M; the data pack correctly reports an FCF yield of 33.0%, which would make NRIM the cheapest stock in the market if the figure meant anything. In 2024 the same formula gave negative CFO ($−8.7M), so a negative "FCF," at a bank earning $37.0M. The metric oscillates with the direction of the warehouse filling and has no connection to earning power.

The chosen definition: Earnings Power Value (Greenwald) on normalized net profit = $57,443 thousand. Composed of net profit over the last four reported quarters, with Q3 2025 adjusted to remove the pre-tax $14.5M gain from the Pacific Wealth Advisors sale:

Quarter	Reported net profit	Adjustment	Normalized base
Q3 2025	$27,065 thousand	−$10,870 thousand (PWA gain, net of tax)	$16,195 thousand
Q4 2025	$12,231 thousand	—	$12,231 thousand
Q1 2026	$13,675 thousand	—	$13,675 thousand
Q2 2026	$15,342 thousand	—	$15,342 thousand
Normalized TTM			$57,443 thousand

This is exactly the same base as in the reference analysis — and it should be, since the new filing doesn't bring a new quarter. I checked that Q2 2026 in the 10-Q ($15,342 thousand) matches the figure from the press release used in July. The base is conservative on two counts and generous on one: it contains the weak Q4 2025 quarter ($12.2M) and is below the half-year annualization ($29,017 × 2 = $58,034 thousand) and well below the Q2 annualization ($61.4M); conversely, it contains only ~$2.5M of the $4.1M annual cost of the subordinated notes issued in November 2025 ($1.2M after-tax under-count). Net-net, I consider it slightly conservative.

Mandatory external check. The earnings yield implied by my base: $57.443M / $564.6M market cap = 10.17%, i.e. an adjusted P/E of 9.83×. What the market publishes: stockanalysis.com gives a TTM P/E of 8.35× (on GAAP EPS of $3.04, which includes the PWA gain) and a forward P/E of 9.88×; my independent calculation on GAAP gives 8.24×. My base (9.83× adjusted) is practically identical to the published forward P/E (9.88×) — my figure is neither far from the market nor a discovery. The 33% FCF yield in the data pack, by contrast, has no counterpart at any broker, confirming that it's the artifact, not my figure. The 2-analyst 12-month price target consensus is $30.50 (+20.2% versus $25.38), and the tracker independently confirms this discount (disc_consens = 0.2017).

Net debt = 0, intentionally. mc_dcf.py warned that the balance sheet implies net debt of −$256.3M (cash plus short-term investments minus total debt). This normalization is wrong for a bank: a bank's cash and securities are operating assets funded by deposits, and deposits aren't counted as debt by the balance-sheet heuristic. Subtracting a "net debt" of −$256M would add $11.37/share of phantom value. EPV on net profit attributable to shareholders already fully reflects the funding structure — interest on deposits and on subordinated debt already flows through the income statement before reaching my base. So nd = 0 is the correct choice and is declared here, per the script's requirement.

The five models

Base data at 06/30/2026 / 09/02/2026: price $25.38 · shares outstanding 22,244,766 · average diluted Q2 shares 22,544,448 · equity $347,614 thousand · goodwill $49,874 thousand · intangibles $950 thousand → tangible book value $296,790 thousand = $13.342/share · book value $15.627/share · TTM normalized net profit $57,443 thousand · TTM adjusted ROTE 20.47% (on average tangible capital of $280.58M) · TTM adjusted ROE 17.33% · beta 0.897.

#	Model	Explicit assumptions	Value/share	MOS vs $25.38
1	EPV Greenwald — bear	$57.443M capitalized at an 11.0% cost of equity (a premium for Alaska geographic concentration + PBCO integration risk), no growth	$23.16	−8.7%
2	EPV Greenwald — base	The same $57.443M at 9.75% (beta 0.897, community-bank earnings with sticky deposits), no growth	$26.13	+3.0%
3	Peer P/E	Estimated 2026 diluted EPS = $2.62 (H1 actual $1.29 + estimated H2 $1.33, assuming a NIM plateau and slightly higher provisions) × 11.0× (the average for small regional banks, $2-10 bn of assets)	$28.82	+13.6%
4	Excess-return P/TBV — bear	Normalized ROTE 14.5% (fading from the current 20.5%), 10% cost of capital, 5% growth → justified P/TBV 1.90× × TBV $13.342	$25.35	−0.1%
5	Excess-return P/TBV — bull	Sustained ROTE 17.0%, 10% cost of capital, 5.5% growth → justified P/TBV 2.56× × TBV $13.342	$34.10	+34.3%

Summary: min −8.7% · base (median of the 5) +3.0% · max +34.3%.

What I changed from the reference report and why. The fundamental assumptions (OE, discount rates, multiples, ROTE bands) are identical — a refresh doesn't rewrite the model if the facts haven't moved it. Three technical differences: (a) the reference price fell from $26.74 to $25.38 (−5.1%), which mechanically lifts every MOS by ~5.4 points; (b) I moved from ~22.3M to 22,544,448 diluted shares (the effective Q2 2026 average per the 10-Q), which slightly lowers the EPV values; (c) I raised the estimated 2026 EPS from $2.60 to $2.62, based on the realized half-year figure of $1.29. I didn't raise the growth rate, didn't lower the cost of capital, and didn't extrapolate the current 20.5% ROTE — even though the new data on margin durability would have justified an argument in that direction. I prefer to let the improvement in information quality show in the narrative chapters, not injected into the model.

Graham Number (independent check): √(22.5 × $2.62 × $15.627) = $30.35, with the current price 16.4% below it (MOS +19.6%). The tracker shows a discount of 28.8% because it uses TTM GAAP EPS of $3.04, inflated by the PWA gain — the correct figure on a normalized basis is the one here.

Transaction comparable: PBCO bought at ~1.67× book value ($167.3M / $100.2M). NRIM trades at 1.62× book value and 1.90× tangible book. The regional bank M&A market is therefore paying multiples very close to what NRIM trades at — neither an overvaluation nor a large-undervaluation signal.

Monte Carlo — 20,000 scenarios

A correction needed before interpretation. The simulation in the delta-triage pack (mc-NRIM-20260901.json) had received --oe 57443 (thousand USD) alongside --sh 22.11 (millions of shares) — a 1000× unit mismatch. That's why it reported a median intrinsic value of $26,756.51/share against a price of $25.74, with a median MOS of +103,849.1% and a 100.0% undervaluation probability. The figures were absurd, but the marker line would have passed unnoticed. I reran it with consistent units: --oe 57.443 --g1 0 --r 0.0975 --gt 0 --nd 0 --sh 22.544 --px 25.38. The result below is copied as-is from the script's output and verified against mc-NRIM-20260902.json.

Monte Carlo — NRIM (20000 scenarios)

Reference price: 25.38 · median intrinsic value: 26.24

Percentile	P5	P10	P25	P50	P75	P90	P95
MOS (%)	-38.2	-30.8	-16.4	+3.4	+27.4	+54.6	+73.6

• Probability the stock is undervalued (MOS > 0): 54.0%
• Probability of a margin of safety above 30%: 23.0%
• Central assumptions: OE 57.443 · g1 0.00% · r 9.75% · gt 0.00% · net debt 0 · shares 22.544 mn · FX 1
• Dispersions: OE ±25% (lognormal), g1 σ=3.0%, r σ=1.0%, gt σ=0.50%
• Terminal value weight: 39.4% — below the fragility threshold, so the result isn't a small-denominator artifact

Interpretation. The simulation's median (+3.4%) and the triangulated five-model median (+3.0%) converge almost perfectly — a sign the mechanical model and analytical judgment describe the same object, which didn't happen at GSL or AFYA. The convergence is also an implicit argument that the owner-earnings base is correctly placed.

What widens the range. The P10→P90 distance is 85.4 percentage points (from −30.8% to +54.6%) — very wide for a bank with predictable revenue. The dominant cause is the ±25% dispersion on owner earnings, which in a no-growth EPV model translates almost one-to-one into value (value is OE/r; ±25% on OE = ±25% on value). The second contributor is the 100 bp sigma on the discount rate: at r = 9.75%, a one-point downward move (8.75%) raises value 11.4%, and an upward one (10.75%) lowers it 9.3%. With g1 = 0 and gt = 0, growth assumptions contribute almost nothing to the dispersion — which is intentional: EPV is by definition a no-growth model, chosen precisely so I don't extrapolate a 20.5% ROTE as structural.

What it means for position sizing. P(undervalued) = 54.0% is practically a coin flip. Only 23% of scenarios offer a margin of safety above 30%, while 25% of scenarios imply a value loss above 16%. At these probabilities, a conviction position is unjustified: the asymmetric ratio a deep-value investor looks for (P(MOS > 30%) well above P(MOS < 0)) is simply not present — 23% vs 46%. The correct sizing is a small, watch-level position, adding conditional on a price below ~$21-22 (where the simulation's P10 turns positive) or on evidence that the 20% ROTE is sustained for another two-three quarters.

---

Pre-mortem — why the thesis could be wrong

Scenario 1 — The single borrower isn't "well-collateralized," and the unprovisioned 84% materializes. My re-evaluated thesis rests decisively on the interpretation that the doubled nonaccruals represent the migration of an already-identified cohort, not new deterioration — an interpretation supported by the 20% decline in criticized loans and the 46% decline in potential problem loans. But $19,281 thousand of the $22,982 thousand of nonaccrual loans (83.9%) carries no specific provision, and the justification is exclusively the collateral valuation: commercial real estate plus commercial loans, described as "well-collateralized." If the valuation of those Alaska commercial properties proves optimistic — Anchorage's CRE market is small, illiquid, and $10.5M of exposure concentrated on a single borrower doesn't liquidate without moving the price — the bank would need to book a specific provision quickly. A 40% provision on the uncovered exposure would mean ~$7.7M pre-tax, ~$5.9M after-tax: 10.2% of TTM adjusted profit in a single quarter, plus the immediate withdrawal of the multiple premium the market pays for "bank with clean credit." The most likely trigger: a mention in the Q3 10-Q (report date October 28, 2026) that the ACL on nonaccrual has risen above $1.3M. This is the scenario with the highest probability × impact of the three.

Scenario 2 — PBCO derails, and derails precisely because it doesn't integrate. The deal terms haven't changed, and Note 15 of the 10-Q reconfirms them: 21% dilution, ~$4.2 bn of combined assets, approvals needed from four authorities (Fed, FDIC, Oregon DCBS, Alaska DCCED) plus both shareholder meetings. The risk element isn't the price — it's the 16+ month window between the estimated closing (Q4 2026 / Q1 2027) and system conversion (Q4 2027), during which two banks with separate cultures, systems and credit teams operate under a single ownership. The risks the 10-Q itself lists in the forward-looking statements section are exactly these: deposit attrition after the announcement, loss of customers or staff, management distraction, unrealized or delayed synergies, above-expectation merger expenses. Professional services expenses already grew 28.3% in Q2 and will accelerate. The adverse-alignment signal: Huston simultaneously holds seven executive and board titles at a moment he needs maximum managerial bandwidth. If PBCO's ROTE is structurally below Northrim's — plausible, given it was bought at 1.67× book value when NRIM trades at 1.90× tangible book — the deal will be dilutive to combined ROE for 2-3 years, even if accretive to nominal EPS, and models 4-5 above (which depend entirely on sustained ROTE) would lose their support.

Scenario 3 — The margin turns back, and turns back from the part I declared durable. I argued in this report that NIM expansion is ~60% cyclical (deposit repricing) and ~40% structural (asset mix + the securities portfolio rolling from 3.07% into 3.79%). The "structural" component, though, has an expiration date I haven't quantified: the available-for-sale securities portfolio is shrinking ($420.7M → $411.6M in six months), and the roll into higher yields stops when the old low-coupon paper runs out. At the same time, the favorable mix (loans from 75.19% to 79.29% of interest-earning assets) is by construction a move with an upper bound — it can't exceed 100%, and at 83.71% loans/deposits the bank is approaching the zone where further growth requires more expensive funding, not cheap deposits. If the Fed accelerates cuts, loan yield (already falling: 6.94% vs 6.99%) declines faster than what's left to reprice on deposits, while the $60M of subordinated notes stays locked at 6.875% until December 2030. NIM could revert toward 4.2-4.7% (2023-2024 levels), which would cut ~$8-12M from annual net interest income and pull ROTE toward the 14-15% band model 4 uses as its bear assumption — validating the $25.35 value, i.e. exactly today's price, with zero return over the holding horizon.

---

Verdict compared with the tracker's GBL score

The PRIOR score from the tracker (queried with query_tracker.py scor NRIM, before any re-scoring by me): 65.8% — INTERESTING (Graham 7.5 / Buffett 5.5 / Lynch 7.0, total 20.0/30, F-Score 8.0, denominator 60.0), evaluated 2026-07-28, source "import."

A discrepancy to flag before the comparison. The reference deep report, dated the same day (2026-07-28), had recalculated its own score to 18.0/30 = 60.0%, verdict MONITOR, lowering G8 from 1 to 0.5, B17 from 0.5 to 0, B18 from 1 to 0.5, and L30 from 1 to 0, plus a "cyclical" flag. The tracker, however, retains 20.0/30 = 65.8% INTERESTING as of the same date. In other words, the re-scoring from the reference report never made it into the tracker — on Lynch the gap is a full point (7.0 in the tracker vs 6.0 in the report) and on Graham half a point. The comparison below follows procedure and uses the tracker figure (65.8%), but this is a maintenance item worth flagging: two internal sources give two different verdicts for the same date.

My verdict: MONITOR — unchanged from July, with the reasons rearranged.

Convergence with the tracker (65.8% INTERESTING) — on pure value criteria. The tracker is right on what it measures: P/E 8.3× (GAAP TTM), P/B 1.62×, ROE 21.5%, PEG 0.27, F-Score 8/9, RSI 35.5, dividend yield 2.56%, a simple story and an ignored sector. All independently confirmed by me from the filings. On the pure Graham-Buffett-Lynch grid, NRIM is a quality company at a reasonable price, and 65.8% is a defensible score. Nothing in the new 10-Q contradicts any of these criteria — on the contrary, with credit quality better than believed, the Buffett criterion of "predictable business" is less shaky than in July.

Divergence — three things the mechanical score doesn't see.

1. The denominator is inflated by a one-off gain. The tracker uses TTM EPS of $3.04, which includes the $14.5M pre-tax gain from the Pacific Wealth Advisors sale (Q3 2025). On a normalized basis, TTM EPS is ~$2.55 and P/E is 9.8×, not 8.3×. The discount to the Graham Number falls from 28.8% to 19.6%. Roughly 18% of the "cheapness" the tracker measures is a non-repeating event. No mechanical formula can detect this — it requires reading the press release.

2. The known future 21% dilution is in no cell. The Lynch dilution criterion measures history (+0.44% over 12 months, excellent). But PBCO shareholders will receive ~21% of the company within 2-6 months, with system conversion not until 16+ months after closing. A score calculated on trailing data can't reflect a contracted but unclosed event, and this is the single biggest risk in the file.

3. Credit risk is smaller than the July report thought, but concentration is larger than the tracker sees. Here the divergence runs both ways and it's the most interesting one. The tracker doesn't see credit quality at all (it's not in the GBL grid). The reference report saw it and overweighted it as a risk. I see it and re-rate it downward as a trend — the criticized pool falls 20%, the potential problem pool 46%, actual losses are 0.014% of the portfolio, provisioning is 6.3× charge-offs — but upward as concentration: a single borrower with three loans represents 45% of net nonperforming loans, and 83.9% of nonaccrual has no specific provision. The risk has transformed from "broad deterioration" into "binary exposure on a single relationship." Not smaller, different — and easier to track.

What would move me to INTERESTING with conviction. Three conditions, any two of them: (a) Q3 2026 (October 28) confirms the single borrower's three loans resolve without a material specific provision, and the criticized pool keeps falling; (b) the PBCO deal receives all four regulatory approvals with no imposed conditions and both shareholder meetings approve it, removing the binary closing uncertainty; (c) the price falls below ~$22, where the simulation's P10 turns positive and the margin of safety becomes real instead of median.

What would make me abandon the thesis: a specific provision above $5M on the concentrated exposure; the ACL on purchased receivables staying under 1.0% with the balance above $130M and charge-offs resuming; or regulatory conditions imposed on the PBCO approval.

Conclusion. NRIM remains a good business at a fair price — not a good business at a good price. The median of the five models (+3.0%) and the simulation median (+3.4%) say the same thing with different tools, and when the convergence is this tight, the range matters more than the point: 46% of scenarios lose money. MONITOR, with the next check scheduled at the October 28, 2026 report, and with the three signals to watch listed at the end of the Quality of Earnings chapter.

---

Markers

DEEPFCF;NRIM;EPV Greenwald pe profit net normalizat (nu FCFE — CFO-ul bancar e dominat de warehouse-ul ipotecar);57443;CFO S1 2026 de 67.298 mii contine +48.146 mii din vanzarea neta de credite held-for-sale (71,6%), deci CFO-capex nu masoara fluxul catre actionar; dobanda e integral in operating (20.185 mii S1), deci nu exista corectie de tip AFYA de facut

DEEPDCF;NRIM;-8.7;3.0;34.3;5 DEEPDCFM;NRIM;1;EPV Greenwald bear r=11%;-8.7 DEEPDCFM;NRIM;2;EPV Greenwald baza r=9,75%;3.0 DEEPDCFM;NRIM;3;P/E peer 11x pe EPS fwd 2,62;13.6 DEEPDCFM;NRIM;4;Excess-return P/TBV bear 1,90x;-0.1 DEEPDCFM;NRIM;5;Excess-return P/TBV bull 2,56x;34.3

DEEPMC;NRIM;-30.8;-16.4;3.4;27.4;54.6;54.0;20000

DEEPIN;NRIM;57.443;0;0.0975;0;0

EXPO;NRIM;petrol-alaska;0.4;Cartea de credite e concentrata in Alaska, stat in care circa o treime din economie depinde de petrol si gaze; un soc durabil al pretului titeiului se transmite in pierderi de credit la NRIM simultan cu caderea oricarei pozitii de producator de energie din plan. Sectorul (Banks - Regional), tara (US) si moneda (USD) nu separa acest factor. Pondere partiala (0,4): portofoliul e diversificat pe CRE, rezidential si factoring, iar distributiile Alaska Permanent Fund plus turismul amortizeaza partial ciclul.

---

Sources

• SEC EDGAR — 10-Q Q2 2026, filed 2026-07-31 (accession 0001163370-26-000029, document nrim-20260630.htm); manually downloaded in this session, converted with filing_text.py. The main source for this refresh: the cash flow statement (p. 8), Note 3 (loans and ACL, risk-rating and nonaccrual tables), Note 4 (purchased receivables), Note 8 (subordinated debt), Note 15 (the PBCO merger), MD&A (segments, net interest margin components, nonperforming assets, potential problem loans, regulatory capital).
• SEC EDGAR — 10-K FY2025 (filed 2026-03-06): the "Competition" section (FDIC deposit share), "Repurchase of Securities," the ACL tables for loans and purchased receivables, the history of adversely classified and potential problem loans.
• SEC EDGAR — DEF 14A (filed 2026-04-14): the beneficial ownership table, the summary compensation table, executive employment agreements, board structure.
• SEC EDGAR — the Q2 2026 results 8-K and the PBCO merger 8-K (both 2026-07-23).
• Deterministic data pack: rapoarte/deep/data-pack-NRIM-20260902.md (SEC companyfacts XBRL + yfinance).
• Web research brief: rapoarte/deep/research-NRIM-20260902.md — used for analyst consensus and the event timeline. Two figures corrected against the filings: the Alaska deposit share (brief ~15% from an aggregator vs 17.5% in the 10-K) and the dividend ($0.68 / 2.68% in the brief vs $0.64 / 2.52% from the cash flow and equity statements).
• Simulation: rapoarte/deep/mc-NRIM-20260902.json (20,000 scenarios, generated in this session). The prior simulation mc-NRIM-20260901.json is invalid due to a unit mismatch — do not use it.
• Reference analysis: rapoarte/deep/2026-07-28-deep-NRIM.md. Delta pack: rapoarte/deep/delta-NRIM-20260901.md.
• Internal tracker: query_tracker.py indicatori NRIM and scor NRIM (2026-09-02).

Analysis, not investment advice; verify the figures before trading.