When a referral comes through on the fax line, the intake clock starts when the last page lands, not the first. On a dense document, that wait can stretch past an hour. We ran nine tests across different page counts and document types to see exactly where the time goes, and whether streaming changes the math.
The challenge: the clock starts when the last page lands
When a referral arrives by fax, someone on the intake team is stuck waiting for every last page before they can do anything useful with it. On a dense, image-heavy document, that wait can run well past an hour. Streaming changes when the first page shows up, but in most workflows that early arrival doesn't help anyone, because a person still can't decide until the whole packet is in front of them. The intake clock keeps running.
The finding: Basalt turns the streaming window into a clinical answer
Because Basalt starts running admission criteria at page four, the clinical answer is ready well before the fax has even finished coming through. Facesheets run two to three pages, so once page four lands, Basalt has enough to check the referral against your criteria and surface an answer, on a typical or dense referral, around the eight-minute mark instead of somewhere past the hour.
- Sparse documents (15–50 pages): 1.6–5.8× faster
- Typical documents (mixed text, tables): 3.3–10.3× faster
- Dense documents (images, poor compression): 3.2–13.2× faster
Where the time goes
Take one representative case: a 50-page typical referral at 4.92 MB. Without streaming, the whole packet takes 1h 23m to arrive. Streaming gets the first page on screen in about a minute and a half, but the decision still waits for the rest. Basalt fires inference at page four and has a clinical answer in eight minutes.
50-page typical referral · 4.92 MB, plotted on a 90-minute axis. Streaming exposes a window; Basalt uses it, with no change to the referring facility's workflow and no waiting for the last page.
Test results
Nine tests on a streaming-capable fax line, across page count and document complexity. Times are measured from fax send. "First inference" is Basalt running admission criteria on the first four pages.
| Test | Pages | File size | Full fax delivered | First page | First inference | Improvement |
|---|---|---|---|---|---|---|
| Sparse · mostly whitespace, long pixel runs | ||||||
| 1a | 15 | 221 KB | 3:42 | 1:06 | 2:17 | 1.6× |
| 1b | 25 | 370 KB | 8:53 | 1:22 | 2:25 | 3.7× |
| 1c | 50 | 742 KB | 16:21 | 1:37 | 2:48 | 5.8× |
| Typical · mixed text and tables, moderate transitions | ||||||
| 2a | 15 | 1.47 MB | 25:48 | 1:34 | 7:55 | 3.3× |
| 2b | 25 | 2.46 MB | 41:41 | 1:11 | 7:38 | 5.5× |
| 2c | 50 | 4.92 MB | 1:23:04 | 1:30 | 8:05 | 10.3× |
| Dense · many transitions, photos and images, poor compression | ||||||
| 3a | 15 | 1.61 MB | 28:28 | 1:54 | 8:47 | 3.2× |
| 3b | 25 | 2.68 MB | 45:42 | 1:26 | 8:29 | 5.4× |
| 3c | 50 | 5.37 MB | 1:32:38 | 3:15 | 7:02 | 13.2× |
Times as mm:ss or h:mm:ss from fax send. G4-compressed TIFF over PSTN, streaming-capable line. Preliminary results, synthetic faxes.
What the data shows
Non-streaming time scales with pages
A 50-page typical referral takes 1h 23m end to end; at 15 pages, 25 minutes. The intake clock starts when the last page lands.
Streaming holds first-page time flat
First page arrived in 1:06–3:15 across all nine tests, regardless of document length.
Basalt fires inference at page four
Facesheets run two to three pages. Once page four lands, Basalt can run admission criteria and surface a clinical answer in 7–8 minutes.
Bigger referrals, bigger gains
Small sparse referrals: 1.6× faster. Large dense referrals: 13.2×. The slowest documents see the biggest gains, roughly 75 minutes faster on a dense 50-page referral.
Streaming exposes a window. Basalt closes the gap on the receiving side and uses it, with no change to the referring facility's workflow, no change to the fax number, and no waiting for the last page.
