Free
- 3 runs a month
- 2 documents per comparison
- Full Scorecard output
- Excel export
Spec comparison for procurement teams
Feed it datasheets.
Get one Scorecard back.
xpecfy reads competing vendor PDFs — crisp exports, dense spec tables, even scanned pages — and returns a single normalized comparison table. Units converted, footnotes surfaced, gaps marked as gaps.
PRIVATE BETA · FREE PLAN AT LAUNCH: 3 RUNS / MONTH
PDF IN → SCORECARD OUT1 RUN = 1 SCORECARDMAX 50 MB / DOCUMENT
Document manifest01
- samsung-pm9a3-datasheet.pdf2.4 MBUploaded
- micron-7450-pro-product-brief.pdf1.1 MBUploaded
- kioxia-cd8-series.pdf8.7 MBUploading64%
Normalization02
"Max 4K Read IOPS""IOPS (4K Rnd Read)""Random Read Performance"
random_read_iops
Scorecard03
| SKU | Random read | Power active |
|---|---|---|
| Kioxia CD8-R 1.92 TB | 1,150K IOPS | 14.0 W |
| Samsung PM9A3 3.84 TB | 1,100K IOPS | 8.9 W |
| Micron 7450 PRO 3.84 TB | 1,000K IOPS | 5.0 WUncertain |
The pipeline
Three steps. One explicit run.
- 01
Drop the PDFs as-is
Crisp digital exports, table-heavy product briefs, scanned pages. No renaming, no cleanup, no pre-formatting — the pipeline does the normalizing.
- 02
Run the comparison
Runs are explicit and metered — uploading never triggers one. You decide when the meter ticks, and one run produces one Scorecard.
- 03
Read one table, not ten PDFs
Every product variant is a row, every spec a canonical column. Units arrive converted to SI; what a vendor didn't publish stays visibly missing.
The output
The Scorecard: facts, aligned.
Objective extracted facts only — no black-box scores. When the system isn't sure, it says so in the cell itself, not in a footnote you'll never find.
Enterprise NVMe SSDs — Q3 buy · Run 04 · 3 documents · 6 SKUs
| AIdentity | BPerformance | CPower | DReliability | ECommercial | |||
|---|---|---|---|---|---|---|---|
| SKU | Capacity | Seq. read | Random readSorted · desc | Power (active) | Power (idle) | Endurance | Warranty |
| Kioxia CD8-R 1.92 TBkioxia-cd8-series.pdf | 1.92 TB | 7,200 MB/s | 1,150K IOPS | 14.0 W | 5.5 W | 1.0 DWPD | —NewFound instead: "TCG Opal 2.01" — new attribute proposed, pending approval. |
| Samsung PM9A3 3.84 TBsamsung-pm9a3-datasheet.pdf | 3.84 TB | 6,950 MB/s | 1,100K IOPS | 8.9 W | 4.0 W | 1.3 DWPD | 5 yr |
| Samsung PM9A3 1.92 TBsamsung-pm9a3-datasheet.pdf | 1.92 TB | 6,900 MB/s | 1,000K IOPS | 8.4 W | 4.0 W | 1.3 DWPD | 5 yr |
| Micron 7450 PRO 3.84 TBmicron-7450-pro-product-brief.pdf | 3.84 TB | 6,800 MB/s | 1,000K IOPS | 12.0 W | 5.0 WUncertainLow-confidence mapping — queued for review. | 1.0 DWPD | 5 yr |
| Micron 7450 PRO 1.92 TBmicron-7450-pro-product-brief.pdf | 1.92 TB | 6,800 MB/s | 800K IOPS | 11.5 W | 5.0 W | 1.0 DWPD | 5 yr |
| Solidigm D5-P5336 7.68 TBsolidigm-d5-p5336-datasheet.pdf | 7.68 TB | 7,000 MB/s | 750K IOPS | 12.5 W | 4.5 W | 0.3 DWPD | 5 yr |
Uncertain — low-confidence mapping, kept in its column and marked for reviewNew — raw value in a newly minted attribute, not yet normalized
Why it holds up
Honest by design.
Units converted, not assumed
Every dimensioned value is normalized to SI. MB/s versus MiB/s stops being your problem.
Gaps marked as gaps
A spec a vendor didn't publish shows up as missing — never silently filled in.
Doubt is flagged
Low-confidence mappings stay in the table, marked uncertain, so you review a cell instead of re-reading a PDF.
Every SKU kept
A datasheet with forty variants doesn't collapse to one "representative" row. Long lists fold; nothing is lost.
A vocabulary that compounds
"Max 4K Read IOPS" and "Random Read Performance" land in the same column. Every reviewed run makes the shared vocabulary richer.
Explicit, metered runs
Nothing runs on upload. A run is an action you take — and a run that fails to produce a Scorecard costs nothing.
Plans
Start free. Upgrade when the meter says so.
Pro
- 25 runs a month
- 10 documents per comparison
- Everything in Free
Business
- 100 runs a month
- 10 documents per comparison
- Up to 5 seats, shared history
- Shared custom vocabulary
A run only counts when it produces a Scorecard · Prices announced at launch
Questions
Asked by engineers, answered plainly.
Do I have to tell it what product category I'm comparing?
No. Drop the PDFs and run — there's no category picker and no template to choose. The canonical vocabulary is shared across domains, so temperature is temperature whether the datasheet describes an SSD or a motor.
What does an "uncertain" flag in my table mean?
The extraction wasn't fully confident where that value belongs. The value stays in its most likely column, visibly flagged, so you can verify one cell instead of re-reading the source PDF. Nothing uncertain is ever silently accepted into the shared vocabulary.
Does a failed run count against my quota?
No. A run is only metered when it successfully produces a Scorecard. Failed runs cost nothing, and re-runs count like any other run.
Can I get the Scorecard out of xpecfy?
Yes — export it to Excel and take it into whatever your team already uses for the decision.
Your next vendor comparison is one drop away.
One email at launch — nothing else