Stress Test the Estimates
Most of this model's inputs are one analyst's judgment - 83 of 162 sub-score cells, every pricingAccess score, and all whitespace multipliers. This screen separates opinion from fact: it classifies every cell, perturbs every judged one, re-ranks the plays under six named stress scenarios, and shows exactly which analyst-set input the flagship wedge conclusion actually hinges on.
The moat arithmetic is a glass box, but arithmetic is only as good as its inputs. 83 of 162 cells (51%) are one analyst's judgment - either a raw estimate or anchored only to a synthesis (author-thesis) paragraph. The rank-integrity test elsewhere on this site covers directional cells only. This screen stress-tests everything it does not.
⚠ Pricing access - the axis that carries the "mid-market Palantir alternative" wedge - has no third-party anchor for ANY of its 27 vendor scores. Its impact is quantified below, not hidden behind a badge.
The tested invariant guarantees no directional cell (±20) can change the default vendor top-10. Running the same ±20 perturbation over every judged cell instead: 28 of 83 judged cells change the top-10 on their own. The ranking is not invariant to the estimates - anyone claiming otherwise would be overselling. Every mover is listed; all of this is re-computed by the assertion harness on every run.
Each column re-scores the model under a named, reproducible assumption shift and re-ranks all ten plays. Under the balanced lens, the #1 play (MCP-native data governance gateway) stays #1 in 4/6 scenarios; the flagship mid-market wedge ranges #3-#4.
| Play | Nominal | Skeptic −15 | Bull +15 | Pricing = 50 | Pull = 1.0 | Adversarial |
|---|---|---|---|---|---|---|
| MCP-native data governance gateway | #1 | #1 | #1 | #1 | #5▼4 | #5▼4 |
| Agent-first context layer | #2 | #2 | #2 | #2 | #6▼4 | #6▼4 |
| Contract-native agentic data quality | #3 | #4▼1 | #3 | #3 | #1▲2 | #2▲1 |
| Mid-market 'Palantir alternative'flagship wedge | #4 | #3▲1 | #4 | #4 | #4 | #3▲1 |
| Framework-agnostic agent observability + hard cost control | #5 | #5 | #5 | #5 | #2▲3 | #1▲4 |
| Domain-specific data agents for regulated verticals | #6 | #6 | #6 | #7▼1 | #7▼1 | #7▼1 |
| Agentic data engineer that opens PRs to dbt/Airflow with full context | #7 | #7 | #7 | #6▲1 | #3▲4 | #4▲3 |
| AI-ready unstructured data pipeline | #8 | #8 | #8 | #8 | #8 | #8 |
| Cross-platform metadata unification | #9 | #9 | #9 | #9 | #9 | #9 |
| OSS agentic data pipeline orchestrator | #10 | #10 | #10 | #10 | #10 | #10 |
Nominal: the mid-market wedge scores 68.6 EIQ-fit, rank #4. Each bar varies ONE analyst-set input across a plausible range and shows the resulting fit + rank.
Measured finding: the single most powerful driver is marketPull - an unsourced constant that multiplies the final score. On its own it swings the wedge between rank #8 and #1. The pricing estimates move the score far less than the multiplier does.
A strategy input is only as good as its kill criteria. Each tripwire names a real-world event to watch, the estimated cells it would re-score, and the wedge's re-computed fit and rank - derived from the same model, not narrated. The re-scored values are illustrative analyst judgment; the deltas are computed.
Wedge assumes Palantir stays enterprise-only (pricingAccess estimated 18).
Watch: PLTR earnings calls & pricing page - any self-serve / published-price SKU below 7-figure TCV.
Wedge assumes the giants keep selling top-down (pricingAccess estimated 46 / 34).
Watch: Microsoft Ignite & Databricks Data+AI Summit - packaged 'agent + semantic layer' SKUs with published pricing.
Wedge leans on EIQ's pricingAccess = 82 - an estimate about the author's own subject.
Watch: EIQ's first ~10 mid-market wins/losses - if sales cycles & TCV look enterprise-shaped, re-score to ~60.
Measured takeaway: at these magnitudes no single tripwire demotes the wedge's rank - the conclusion is robust to one-vendor re-scores but NOT to the marketPull multiplier above. That asymmetry is the finding.
Everything on this screen is deterministic and re-derivable: the census, the perturbation count, the scenario grid, the tornado and the tripwires are computed from the same pure scoring module as the live sliders, and frozen as assertions in model.test.ts - if any number here drifts from the model, the harness fails.