Spectra Systems Corporation SPSY

Deep-value analysis — SPSY (Spectra Systems Corporation) — RERUN / REFRESH

Date: September 11, 2026 · Exchange: LSE / AIM · Yahoo ticker: SPSY.L · Reference price: … Market cap: GBP 75.54 mil. = US… at GBPUSD 1.34966 (48,424,151 shares) Reporting currency: USD · Trading currency: GBp — DIFFERENT; all yields relative to market cap are converted explicitly. Reference thesis: rapoarte/deep/2026-08-17-deep-SPSY.md (08/17/2026, verdict AVOID at …) · Delta pack: rapoarte/deep/delta-SPSY-20260908.md (REFRESH triage, 6 unverifiable falsifiers)

Price convention, stated because different sources give different figures today. yfinance currentPrice = … (retrieved 09/11/2026 06:32, identical to the data pack); last close in the series = … (09/09/2026); stockanalysis.com shows … at 09/10/2026 16:05 GMT; the tracker has … at 09/10/2026. The real band is 153-…. I use … in ALL models and in Monte Carlo, since it's the price in this session's data pack — a MOS calculated on one price and a simulation run on another would be incomparable. At …, every MOS in the report falls by ~2.6 percentage points.

---

What I inherit and what I re-derive (REFRESH regime note)

The rule for this rerun: no new filing exists. The check is direct and I state it as such — the full list of press releases on Investegate for SPSY shows the last entry as "New Licensing Agreement," August 4, 2026, i.e. an announcement already fully analyzed in the reference report. There's no release after August 4, no "Notice of Interim Results" for H1 2026, and no interim published as of this report's date. The latest financial statements remain those audited at December 31, 2025, published March 30, 2026 — today they're 8.4 months old. The quarterly balance sheet in the data pack mechanically confirms the same thing: the available columns are 2025-12, 2025-06 and 2024-12, exactly as in August.

I inherit, without re-deriving (the factual base is identical, nothing was published to move it):

Chapter	What I inherit	Why it's legitimate
Business and moat	the three-tier structure, FY2025 segment margins, the post-sensor cliff arithmetic, the pipeline inventory	all figures come from the FY2025 release (03/30/2026) and RNS filings through 08/04/2026
Management and capital allocation	the 3-year allocation table, dividend history, absence of buybacks, the failed Cartor acquisition	same; updated with the actual July 2026 dividend payment and the CFO's identity
Quality of Earnings (O'Glove)	all 5 sections, over 2023-2025 + H1 2025	there's no new reporting; the … unbilled receivables is still the last published balance
CEO profile	the Thorndike grid, 2.5/5 score	updated with the insider ownership movement
Accounting red flags	the 14 points, with their figures	same — plus two additions that didn't exist in August
The FCF bridge	the definition, the table, the 3 owner-earnings routes (…-7.4 mil.)	there's no new cash flow statement; the DCF base remains …
DCF assumptions (oe/g1/r/gt/nd)	unchanged: 7.4 / 6% / 11% / 2% / −…	an opinion move I don't make: I haven't received new information justifying a change; keeping the assumptions, the only variable between August and today is price, so the MOS difference is interpretable

I fully re-derive (something verifiable moved here):

1. The intrinsic value and MOS of all 5 models, recalculated with the same DCF function the tracker uses (dcf_buffett.intrinsic_value) at a price of … and today's exchange rate (…/pence versus 0.01355546 in August). The five intrinsic values reproduce within 0.6% — which retroactively validates the August report's arithmetic — and the MOS moves entirely due to price.
2. Monte Carlo, rerun from scratch. The September 8 delta pack's simulation is unusable: it reports a MOS median of ** …** and an intrinsic value of 171,261 against a price of 171. The cause is a unit error, not a fundamentals shift: the stored thesis gives oe = … and nd = −… in thousands of USD, and mc_dcf.py reads them as millions — an owner earnings base of 7.4 billion. The reference report had run correctly (--oe 7.4 --nd -…), but hadn't written the DEEPIN; line, so its assumptions didn't survive in the form the code reads. Today's rerun fixes exactly that: runs correctly AND declares the line.
3. The comparison with the tracker's GBL score, since the tracker was re-scored: 86.7% (26/30) on 08/17/2026 → 65.8% (19/30) on 09/06/2026, via rescor_dovezi. The 21-percentage-point divergence that was chapter 10's conclusion in August closed on its own.
4. The falsifier set, entirely rewritten. The triage returning "6 unverifiable falsifiers" wasn't accidental and isn't temporary: it's a structural property of the issuer, demonstrated below, and a falsifier set that can never be evaluated defends no thesis.
5. The yields and multiples relative to market cap (dividend, FCF, FCFE, P/E on each profit definition, the Graham Number), all at the new price.
6. The targeted external checks the brief flagged as gaps: the dividend payment, the CFO's identity, the 2026 profit consensus.

The price move is context, not thesis. … (08/17) → … today, …. Put in series, the move is older and larger: the 12-month high is … on June 5, 2026, so the stock has lost 24.3% in three months, of which ~… (≈5%) is mechanically the dividend going ex on July 2. The rest is drift on thin volume (3-month average: 82,872 shares/day, ~GBP 130,000/day) in a window with no press release at all. I build nothing on this move; I use it only as the entry price in the models.

---