The short version.
We ran SYNC's research for ourselves, with iSyncSO as the seller. Over eight weeks it researched companies 594 times, scored 569 of those, and placed 239 in the two tiers that SYNC works.
It also made mistakes we could see. Its own audit flagged 35 companies as wrong matches, most of them too large for what we sell. And of the 73 dossiers it marked ready, 23 have a date we can read on the reason. The other 50 do not, although the rule on this site asked for a dated reason at the time.
This is a company judging its own product on its own data. Read it as a description of what the software did, and as nothing more.
What the runs produced.
We started 16 research runs between 20 July and 15 September 2026, for six offers that we described. Ten runs completed, five failed, and one was set up and never started.[1]
The scoring has four tiers. A and B are worked. C and D are both left alone. Elsewhere this site names tiers A, B and C. D is the lowest tier and is treated like C. The sums 239 and 330 are ours.
After scoring, a second step in the software reviews the list and flags companies that scored well without being real buyers. It flagged 35. Of those, 16 had been scored A or B.
| Why it was flagged | Flagged | Of which scored A or B |
|---|---|---|
| Too large for the profile we described | 24 | 13 |
| Sells the same kind of service, or is a competitor | 8 | 3 |
| More funding than the profile allows | 1 | 0 |
| Outside the geography we described | 1 | 0 |
| Wrong entity, or not an operating company | 1 | 0 |
No error rate follows from this. The flag is the software auditing itself, and no person checked it. A company without a flag may never have been audited at all, so “not flagged” does not mean “correct”.
What the dossiers concluded.
SYNC wrote 244 dossiers. For two in three it concluded that we should wait.
When we measured, the site stated our rule as “No dossier, no message. An account without a dated reason gets nothing from SYNC.” The software did not hold to the second half. We looked at the reason each ready dossier rests on and counted whether it carries a date that can be read as a date. 23 of the 73 do. The other 50 do not.
In those 244 dossiers it mapped 1,600 people into roles. It marked 1,317 of those roles as its own inference and 260 as stated by the source.
Drafts went to a person for approval. That person released 65 and rejected 4. Another 41 are still waiting for a decision. We do not publish what happened after sending, because our reply and open counts are not yet reliable.
What we changed in the product.
Two changes bear on the wrong matches. Both were made in August, while our runs were going on. We do not claim that our own runs prompted them.
- 01A flag now has a consequence. Before, a flagged company kept its tier and its place in the list, and the flag was only a sentence in the report. Now a flagged company drops one tier, and the list is ordered by tier first.
- 02The companies a seller has excluded are now taken into account when an account is scored.
Two questions were open when we wrote this piece. We went back to the record for both.
- 01The size limits. Every run keeps a copy of the profile it used. The first run asked for companies of 50 to 500 employees. The runs from 26 July to 8 August asked for 10 to 200. From 19 August, five of the six runs asked for 11 to 50, and one records no size range. All 24 companies flagged as too large come from four of those five runs. So we narrowed the limit before the first flag, and we did not change it after. The record does not say why we narrowed it. No run before 19 August carries a flag, and the record does not say whether the audit ran on them.
- 02The five failed runs. Two stopped at the last step, the writing of the report, after three attempts, because our credit with the provider of the language model had run out. Their research was complete. Two were test runs that our own test tooling created as failed on purpose, and they researched nothing. One we stopped ourselves and replaced with a new run, to check a correction to the profile. The record does not say what was changed because of the two real failures.
Today the check asks for a cited signal and its source, and does not require a date. We decided to keep this: a cited signal without a date can still make a dossier ready. The reviewer sees that the date is missing, and we are making that more visible in the review screen. The wording of the rule on this site now says a reason we can cite, not a dated reason.
What speaks against this.
A company judging its own product on its own data is the weakest form of evidence. We chose the profile, ran the software, read the result and wrote this piece. Nobody outside iSyncSO has checked any of it.
Our workspace is also where we test. The runs used six offers with changing criteria, and five of sixteen failed. Three of those five were our own tests or our own decision to stop. The size limit in the profile changed twice during the period. So this is not one clean experiment.
The wrong-match count cannot be turned into a rate. It counts what the software's own audit caught. What the audit missed is unknown. The reasons were sorted into categories by matching words in a free text, and a reason that names both size and geography counts as size.
Nothing here shows that the companies in tier A would buy from us. A tier is a score against criteria we wrote ourselves. Most dossiers say wait, and 41 drafts have not been decided on. The figures describe one workspace and do not carry over to any customer.
We counted what SYNC recorded in our own workspace: research runs and their outcome, researched companies with their tier and their wrong-match flag, dossiers with their conclusion, people with their role, and drafts with the decision of the reviewer. The runs date from 20 July to 15 September 2026. Nothing was left out. A company researched in more than one run is counted once per run, so 594 is a count of research results and not of different companies.
The tiers are given as they stood on the day of measurement. We give no reply, open or meeting figures. The two product changes were read in the software and its record of changes. They were not measured. For the failed runs and the size limits we read, on 29 September, what each run recorded: its final state, the step it stopped at, the error, and the size range in the copy of the profile kept with the run.
- [1]iSyncSO. Everything SYNC recorded in our own workspace between 20 July and 15 September 2026, counted on 29 September 2026.our own measurement