Made on
Sep 4, 2026
Price at call
$18.22
6-month call
Bear -7.8%
Target by Mar 2027
$16.80
Great value below
$13.50
Model
v0.6.0
Projection vs Actual (6M history + forecast)
Forecast Change
-7.8% to $16.80
Predicted High
$19.10
at 1 month
Predicted Low
$15.40
at 4 months
Max Drawdown (predicted)
-15.5%
Forecast
The masterbrain trajectory — 7 dated checkpoints. Short rungs (2w, 1m) show the typical range as the primary claim; point calls run from 2 months out. Every claim still gets graded. Predicted Sep 4, 2026 1:05 pm
Bear
SOFI trends lower to
$16.80
(-7.8% from $18.22)
by Mar 2027.
ride-then-fade
ThesisSoFi lacks a clean fair-value anchor and is stuck in a heavy tape (down ~30% YTD, beta ~3) offset by a fresh Kraken/stablecoin narrative and a risk-on regime. Near term the Scotia initiation and crypto story likely lift shares modestly, but weak value/quality/sentiment lenses and rate sensitivity should reassert gravity toward the low-to-mid teens over the winter before stabilizing.
Invalidated ifA decisive close above $22 on volume, or a break below $14 that holds, would invalidate the drift lower thesis.
| Checkpointtime after the call | Datewhen it gets graded | Predictedthe claim: closing price | Current priceactual close — fills in over time | Predicted returnpredicted vs $18.22 at call | Conv.brain's confidence, 1–10 | Driverwhy the brain put the point here |
|---|---|---|---|---|---|---|
| Prediction made | Sep 4, 2026 | — | $18.22 at call | — | — | The anchor — every point below is measured from this price and date. |
| 2 weeks | Sep 18, 2026 | $16.19–$20.25 typical range · internal point $18.80 | — | ±11.1% | 6/10 | Risk-on regime and Kraken narrative lift shares near-term |
| What actually happened: closed $16.73 on Sep 17, 2026 = -8.2% vs the call (predicted +3.2%) · direction MISS (called flat, was down) · off by 11.4 pp · accuracy 4/10 · typical range ±11.1%: inside the band · S&P -1.1% over the same window — lagged it | ||||||
| 1 month | Oct 4, 2026 | $15.28–$21.16 typical range · internal point $19.10 | — | ±16.1% | 5/10 | Scotia initiation momentum and crypto tailwind persist |
| What actually happened: closed $15.77 on Oct 2, 2026 = -13.5% vs the call (predicted +4.8%) · direction MISS (called flat, was down) · off by 18.3 pp · accuracy 4/10 · typical range ±16.1%: inside the band · S&P +0.1% over the same window — lagged it | ||||||
| 2 months | Nov 4, 2026 | $17.60 | — | -3.4% | 5/10 | Narrative fades, bond selloff pressure reasserts on high-beta |
| 3 months | Dec 4, 2026 | $16.20 | — | -11.1% | 5/10 | Year-end tax-loss selling on YTD laggard, weak value lens |
| 4 months | Jan 4, 2027 | $15.40 | — | -15.5% | 4/10 | January reset, quality concerns weigh without catalyst |
| 5 months | Feb 4, 2027 | $16.00 | — | -12.2% | 3/10 | Approach attractive zone, earnings anticipation stabilizes |
| 6 months | Mar 4, 2027 | $16.80 | — | -7.8% | 3/10 | Post-print reaction on 5-of-5 beat streak, modest recovery |
Why ranges on the short rungs?
A 102-prediction trial (2026-07-21) showed 2-week point calls carry no information here —
three prompt formulations all lost to simply predicting zero — while the stock's own
volatility scale (σ×√t) described those windows well. So the short rungs claim what's
actually knowable: the typical travel, with earnings timing flagged where the variance
will come from. The internal point calls keep being generated and graded to map where
real skill begins. Full evidence: stocks/predictions/V0.5-DESIGN.md §8.
Deterministic control (v0.3 value line) targets $0.00 (+0.0%) — both lines get graded at every checkpoint, so the brain has to beat the formula, not just replace it. Brain: claude-opus-4-7.
Weaknesses
0 blocking · 1 degraded · 5 notes — follow the chain top-down (cause → effect).-
Degraded
No usable fair-value anchor — brain-only (anchorless) mode: no v0.3 control line, prediction reasons from lenses/sentiment/regime/price behaviour.
-
Note
EPV de-weighted — growth IS the value proposition for this type of company
-
Note
Valuation anchor absent: composite_fair_value.
-
Note
Valuation anchor absent: dcf.
-
Note
Valuation anchor absent: anchored_pe.
-
Note
Valuation anchor absent: signal_adjusted.
Equation Inputs
Everything feeding the v0.6.0 model — value anchors, the stock's own price behaviour, and the tunable knobs.Value anchors (from deep analysis)
| Price at call | $18.22 |
| Composite fair value | — |
| Signal-adjusted fair value | — |
| DCF fair value | — |
| Anchored-PE fair value | — |
| Buy-below (value lens) | — |
| Value net score | -74 |
| Value confidence | 7 / 10 |
| Quality net score | -25 |
| Memo confidence | — / 10 |
| Deserved value (base) | — |
| Quality tilt | — |
| Deserved value (used) | — |
| Coherence (value × quality) | — |
Price behaviour (trailing — shapes the realism line)
| Realized volatility (annualized) | 1m 50.0% 3m 55.9% 6m 55.9% |
| Daily σ (realism noise) | 3.5% |
| Beta vs S&P 500 | 1m 4.77 3m 2.94 5m 2.81 |
| Trailing 6-month return | -3.8% |
Knobs (equation params)
| ladder | ["w2","m1","m2","m3","m4","m5","m6"] |
| llm_model | claude-opus-4-7 |
| max_tokens | 3000 |
| control_knobs | {"w_dcf":0.5,"k_qual":0.200000000000000011102230246251565404236316680908203125,"reach_6m":0.450000000000000011102230246251565404236316680908203125,"coh_floor":0.5,"conf_gain":0.059999999999999997779553950749686919152736663818359375,"noise_phi":0.84999999999999997779553950749686919152736663818359375,"conf_floor":0.40000000000000002220446049250313080847263336181640625,"w_composite":0.5,"noise_window":"3m","flat_band_pct":5,"horizon_months":6,"noise_sigma_fallback":0.01499999999999999944488848768742172978818416595458984375} |
| Neutral band (%) | 5 |
| earnings_brief | v1 |
| prompt_version | p1 |
| control_version | v0.3.0 |
Version & Tech Chain
What produced this prediction — stamped on the record so versions can be compared by success rate later.
Model
v0.6.0
#8eb3df85c300 Tech composition (what this version is built from)
| # | Process | Kind | Group |
|---|---|---|---|
| 1 | Synthesis valuation-synthesis |
deterministic | final |
| 2 | Scenario Analysis scenario-valuation |
deterministic | valuation |
| 3 | Valuation / Mispricing ext-lens-value |
llm | extended |
| 4 | Company Quality ext-lens-quality |
llm | extended |
| 5 | Forensic Memo (combiner) ext-forensic-memo |
llm | extended |
| 6 | Classification company-classification |
llm | foundation |
| 7 | Live Quote fmp-quote |
data | market-data |
Inputs used (the runs that fed this prediction)
| Upstream process | Ran | Lag | Run id |
|---|---|---|---|
company-classification |
Sep 4, 2026 | freshest | SOFI-20260904-013001-6316 |
ext-forensic-memo |
Sep 4, 2026 | freshest | ext-SOFI-20260904-020235-ac1d |
ext-lens-quality |
Sep 4, 2026 | freshest | ext-SOFI-20260904-020235-ac1d |
ext-lens-sentiment |
Sep 4, 2026 | freshest | ext-SOFI-20260904-020235-ac1d |
ext-lens-value |
Sep 4, 2026 | freshest | ext-SOFI-20260904-020235-ac1d |
scenario-valuation |
Sep 4, 2026 | freshest | SOFI-20260904-013001-6316 |
valuation-synthesis |
Sep 4, 2026 | freshest | SOFI-20260904-013001-6316 |
Inputs flagged 7d+ behind the freshest one may be stale relative to the rest of the analysis.
Saved History
1 frozen snapshot — the record we score once predictions mature.| Made | Model | Call | At call | Target | Exp. | Eval due | |
|---|---|---|---|---|---|---|---|
| Sep 4, 2026 | v0.6.0 | Bear | $18.22 | $16.80 | -7.8% | Mar 2027 | viewing |
Not Yet Modeled
What this v0.6.0 forecast deliberately ignores — the work ahead, in plain sight.- Scenario probabilities — one path, no bull/bear split (base_prob = 1.0).
- Intraday / sub-2-week behaviour — the first claim is at 14 days by design (EOD product).
- Position sizing / portfolio context — the path is a price claim only.