InteCrawl
Methodology

How we source, score and correct

Every InteCrawl Opportunity Card is something you should be able to check. This page explains how — including where we've been wrong.

Sources

  • Every material claim on a card must appear in a source linked on that card. A plausible number with the wrong citation is treated as an error.
  • Regulatory, funding and permitting claims need a primary source — the agency, company filing or official release — or the card says plainly that primary confirmation is missing.
  • Each source is labeled primary or secondary. Trade press and news are secondary, however reputable.
  • We use public, lawfully accessible sources and licensed data only. We do not bypass logins or access controls, and we do not claim or sell non-public information.
  • Quoted excerpts stay short and within the source's terms.

Observation vs. inference

Each card separates what a source says happened (Observed — sourced fact) from what we think it means (Why it matters and Who benefits, labeled inference). The recommended action is our judgment. Every card also has a What might make us wrong section for counter-signals and open questions. If the evidence is thin, the card says so; it doesn't get filled in.

Priority points — not probabilities

The number on each card is a 0–100 priority score from the rubric below, set by an analyst. It ranks what deserves attention first. It is not a probability of winning business, and we don't publish win probabilities: we don't yet have enough tracked outcomes to make them meaningful. Weights are versioned and will be adjusted against real outcomes as they build up.

FactorMax ptsWhat it measures
Fit to the buyer30How closely the change matches the customer's product, buyer, geography and deal size. For public cards: fit to the typical supplier named under “Who benefits”.
Strength of evidence25Primary vs secondary, how many independent sources, and whether the status is current.
Reachable buying path20Whether there is a plausible person, process or procurement route to act through.
Timing and actionability15Whether there is a window to act now, and a concrete next step.
Value relative to effort10Likely economic value compared with the cost of pursuing it.

Hard stops — a card is rejected regardless of score if it has an inaccurate or unverified material claim; wrong geography or a prohibited entity for the customer; a clearly expired opportunity; no lawful right to use the underlying data; no plausible match to what the customer sells.

Evidence and urgency bands

Evidence and urgency are shown as bands, not percentages, so we don't imply more precision than we have.

Strong
Primary sources confirm the material claims, and nothing we found contradicts them.
Moderate
Credible secondary reporting, or primary sources with a status that is still unresolved or reported inconsistently.
Limited
A single secondary source, an expired timing claim, or a status we could not confirm. Limited cards are not used for recommendations.

Urgency is High when there's a dated deadline or a window closing within weeks, Medium when the window is months away, and Low when the change is structural with no near-term deadline. The NOW / EARLY / FORMING stamp shows where the opportunity sits in its life cycle.

Freshness, expiry and staleness

Every card shows four dates: when the change happened, when we first logged it, when we last verified it, and (where relevant) a review-by date. If we can't re-verify a card, it's marked Stale, drops to the bottom of the list, and stops generating alerts until a primary source confirms its status.

Kind of changeRe-verification cadence
Fast-moving (regulatory actions, deadlines)Re-checked every 24–72 hours while active
Event-driven (venues, sponsorships, RFPs)Weekly until the key deadline passes
Long-cycle capital projects (fabs, campuses)Weekly or monthly, depending on project phase

Conflicts and human review

When sources disagree, we use the primary documentation and say what's unresolved on the card. AI helps with retrieval, extraction and drafting. A human analyst reviews every card before it's published or delivered. We also re-audit a sample of published and rejected cards, so we catch both errors and good opportunities we wrongly threw out.

InteCrawl doesn't send outreach to anyone on a customer's behalf without that customer's explicit approval. If we recommend a service provider we have a relationship with, we disclose it.

Corrections policy and log

If we get a material fact wrong, we fix or withdraw the card, record it here with the date, and tell any customer who received it. We never quietly delete entries from this log. Last full re-verification of all public cards: 2026-10-11.

2026-10-11Withdrawnmaterial

Micron confirms $5B for its own new 300mm wafer fab in Sherman, Texas

Wrong. Micron did not announce its own Sherman fab or a $5B Sherman figure. On July 9, 2026 it committed up to $3B to its U.S. supply chain, including $500M to GlobalWafers' Sherman plant. The accurate facts now live on the GlobalWafers card.

Manufacturing Dive
2026-10-11Withdrawnmaterial

Applied Materials commits roughly $150M to expand its Austin operations

Unsupported. The cited August 6, 2025 release describes Austin as Applied's largest manufacturing and logistics site and a $200M+ Arizona facility — no $150M Austin investment or new 724,000 sq ft warehouse. A ~725,000 sq ft Austin logistics center was completed in 2022, so it is not a new signal.

Semiconductor Digest
2026-10-11Withdrawnminor

Austin-area AI venture funding reaches $1.2B in H1 2026, up 40% year over year

Below evidence standard. The only source is an unattributed article that cites no underlying dataset. Withdrawn rather than published as a precise market figure.

2026-10-11Correctedmaterial

Williamson County identifies 145 companies scouting sites to supply the Samsung fab

The '145 companies' and '$920M' figures did not appear in the cited source. $920M is the EDP's 12-company total since 2023. Card rewritten and priority lowered from 85 to 58.

2026-10-11Correctedmaterial

OpenAI and Oracle advance a 4.5GW data center campus in Abilene, Texas

4.5GW is the multi-site Oracle–OpenAI agreement, not the Abilene campus (about 1.2GW). The Oracle–OpenAI expansion at Abilene was reported dropped in March 2026. Card rewritten.

2026-10-11Correctedminor

Tesla's $16.5B AI6 deal makes Samsung's Taylor fab its first anchor client

Removed an unsourced '1.2 million chips per quarter' figure and stale tool-installation timing; added the September 2026 order-book report.

2026-10-11Correctedminor

Texas ties AI data centers to new state standards; OpenAI is first to comply

Removed an unsourced 'nearly 20 Austin-area projects' claim; the standards are a gubernatorial directive, not adopted state rules.

2026-10-11Correctedminor

Dell selected to build IGNITE, a state-funded AI platform for Texas A&M

Location corrected: IGNITE is being installed in an Equinix data center in Dallas, not Bryan/College Station.

2026-10-11Marked staleminor

Taylor's workforce housing shortage intensifies as Samsung staffing ramps

A '~1,500 employees by end of 2026' figure was not in the source (it says 1,800 hires over five years). The August 2025 project's financing outcome is unconfirmed.

2026-10-11Marked staleminor

NVIDIA commissions two Texas plants with Foxconn and Wistron to build AI supercomputers

The stated 12–15 month production window has elapsed with no confirmation found. Suspended pending primary verification.