Software & SaaS — all subsectors

Claim

Swept exhaustively during Act I of the 2026 rout ([[regime]]). A formal subsector taxonomy exists at Output/Discovery/software-sector-taxonomy.md and should be the starting point for any further work rather than a fresh scout pass.

Coverage by subsector

Status	Subsectors
Well screened	ERP · CRM · HCM · Payments · Cybersecurity (network/endpoint) · IT Services · IT Distribution · Data Warehouse
Partially screened	Vertical SaaS (healthcare) · DevOps · Identity · Observability · AdTech · Automation · MarTech · iPaaS · Low-code · GRC/Compliance · Communication platforms
Never screened	GovTech · Capital-Markets software · Contact Center · Construction · E-commerce · Simulation · Data Integration · Automotive software

The "never screened" row is the actionable part of this note — it's where a software screen should go, not back over the well-screened rows.

What surfaced repeatedly

Deep value tier (March): ADBE 9.3x fwd, WDAY 10.7x — the two highest-conviction names across the whole series.
AI-headwind dividend tier: ADP, PAYX, OTEX, BR, SSNC — moats being punished for AI exposure while paying you to wait.
AI-tailwind tier: CTSH, DOX, GEN, NTAP — "legacy" labels on businesses AI was quietly helping.
Vertical / niche: SPSC (EDI network effects, 64% off high), BLKB (nonprofit monopoly, 7.5x), QTWO, QLYS (zero debt, 33% op margins), MANH (zero debt, 29% FCF growth, 44% off high).
July refresh found the field largely picked over — ALRM and IRDM the only names of interest, and "everything smaller that surfaced trades cheap for a reason."

Rejected, with reasons worth keeping

Name	Why
ZS	GAAP-unprofitable at 34x P/FCF
HUBS	Huge drawdown but SBC-adjusted earnings ≈ zero
VEEV	Great moat, 28x P/FCF — not deep value
MDB	Best developer moat in data infra, but SBC exceeds FCF
PD	Value trap — decelerating growth, extreme SBC, buyback funded unsustainably
PLTR	83x fwd — never in the value universe

SBC-versus-FCF was the single most common disqualifier. In this field the headline multiple is routinely misleading and the SBC adjustment is the real screen.

🚩 The unread lesson

Not one of these screens has been checked against what happened next. The March thesis — that AI-displacement fear was indiscriminate and moated names were mispriced — is testable now with four months of hindsight, and testing it would score the method, not just the names. This is the highest-value unrun analysis in the repo. Until it's run, treat the March conclusions as untested hypotheses, which is why confidence here is medium despite the volume of work behind it.

First data point on the retrospective — ADBE, scored 2026-08-04

The March cohort's joint-highest-conviction name (9.3x fwd) is the first to be re-underwritten against real results. The verdict is split, and the split is the lesson.

The March thesis said	What actually happened
The sell-off is indiscriminate AI fear, not a business problem	✅ Half right. Q2 FY26 beat on both lines (+13% rev, +18% EPS); FY26 guidance raised twice above street; the FTC overhang closed
A moated compounder is mispriced at 9.3x	🔻 Half wrong. Adobe voluntarily guided FY26 ARR growth down to 10.2% from a 12.5% exit rate, ~half of it from deferring Creative Cloud price increases
Price would re-rate as the fear passed	⚠️ It de-rated further, then partially recovered. $190 low, now $253.81; the multiple never came back

The mechanism the March screens could not see: a cheap multiple on a business that is about to stop taking price is not the same asset as a cheap multiple on one that will keep taking it. Adobe told the market it could not raise prices on its flagship — and that is a disclosure no screen reads, because it lives in guidance commentary, not in a ratio.

What this scores about the method, not the name:

The screens measured the drawdown correctly and the mispricing only partly — which is the specific failure [[principle-down-a-lot-is-not-cheap]] warns about, arriving here in its subtler form. ADBE genuinely was below its own band. It was still not worth what the screen implied, because the earnings trajectory changed after the screen ran.
The SBC filter held up well. ADBE passed it in March (SBC 19.7% of FCF and falling) and still passes. That filter is the most durable thing the March series produced.
What the screens had no column for was pricing power. Every name in the deep-value tier was scored on multiple, margin and SBC. None was scored on "will this company still be able to raise prices in 18 months?" — which turned out to be the variable that mattered.

Consequence for the open retrospective: run it, but score each name on whether the earnings trajectory held, not on price. Price alone will credit the March screens for the sector rebound and blame them for the sector's de-rating, and neither is information about the method. See Output/Stocks/Technology/ADBE/analyze-2026-08-04.md (conviction 7.5 → 6.5, fair value $380-460 → $290-360, moat 7.5 → 6.0).

One name is not a retrospective. Confidence stays medium.

What would falsify this

The retrospective showing the March picks tracked the sector rather than beating it — which would mean the screens were measuring the drawdown, not the mispricing, exactly the error [[principle-down-a-lot-is-not-cheap]] warns about.

[[regime]] · [[pattern-value-depth-varies-by-software-layer]] · [[pitfall-not-investable-universe]]

History

2026-03-18 → 2026-04-28 — twelve screens across Act I.
2026-07-13 — refresh found the field picked over.
2026-07-30 — consolidated; the missing retrospective flagged as the open item.
2026-08-04 — first retrospective data point. ADBE (the cohort's top pick) re-underwritten: the beat-and-raise validated the "indiscriminate fear" half of the thesis, but a deliberate ARR guide-down on deferred price increases falsified the "mispriced compounder" half. Conviction cut 7.5 → 6.5. INTU analyzed for the first time and added to the roster — it was never swept despite 12+ software screens, because it was inside the portfolio the whole time.