The Moat Test

The Moat Test

Lab: meeting extraction

Pick one of six synthetic transcripts, or paste your own, and read what the deterministic baseline actually pulls out of it. Then compare that against a hand-authored output of the shape the extraction contract asks for, without being told which is which until you have decided.

Synthetic: this transcript was authored for the benchmark and depicts no real meetingMeasured: Measured — archived experiment runIllustrative: Illustrative — not a measured result

Nothing here is a benchmark result. The scored runs are in section 5 of the investigation; this page shows one transcript at a time and scores nothing.

Choose a transcript

Synthetic: this transcript was authored for the benchmark and depicts no real meetingAll six were written for this benchmark. None is a real meeting.

Or use your own

Your transcript stays in this browser. It is not uploaded, and no model is called: the deterministic baseline runs locally on the text you paste.

One utterance per line, in the form [L001] Priya: Let's start with the renewal.

0 of 30,000 characters · 0.0 of 100 KB · 0 parsed lines

No model provider is configured for this deployment. The live trial and the AI extraction method are unavailable, and no measured result exists for them.

Onboarding email sequence, send owner named aloud

2 of 32 lines carry an ACTION: or DECISION: tag. Line ids are shown so any citation below can be checked against the source.

  1. L001PriyaRight, we have twenty minutes on the onboarding email sequence, so let's keep this tight.
  2. L002MayaI shared the draft copy yesterday evening. Three emails, spread over the first week.
  3. L003TomasI read them on the train. The second one is doing too much work at once.
  4. L004MayaAgreed. I split the billing explanation out into its own paragraph this morning.
  5. L005PriyaDoes legal need to see the refund wording before we send anything to customers?
  6. L006TomasThey reviewed the refund clause in July and nothing about it has changed since.
  7. L007PriyaGood. Then the blocker is the sending domain, not the copy.
  8. L008TomasThe domain warm-up finishes on the fourteenth. Anything before that risks the spam folder.
  9. L009MayaSo we schedule the first send after the warm-up completes and not a day earlier.
  10. L010PriyaDECISION: the onboarding sequence goes out after the domain warm-up completes.
  11. L011TomasWho is actually pressing the button on the first send?
  12. L012MayaMaya will send it. I built the sequence, so I should be the one to trigger it.
  13. L013PriyaFine. Maya sends the first onboarding email once the warm-up is confirmed.
  14. L014PriyaACTION: Maya will send the first onboarding email after warm-up completes.
  15. L015TomasI want the unsubscribe link tested against the mobile client before anything goes out.
  16. L016MayaThat is a five minute check. Tomas, can you take that one?
  17. L017TomasYes, I will test the unsubscribe link on the mobile client and report back here.
  18. L018PriyaNoted. Tomas owns the unsubscribe test.
  19. L019MayaThere is also the question of what happens to trial users who convert on day two.
  20. L020PriyaThey should drop out of the sequence. Otherwise we are selling to someone who already bought.
  21. L021TomasThe automation platform supports that, but nobody has configured the exit condition.
  22. L022PriyaThen that is a real task and it needs an owner today.
  23. L023MayaI would rather not take a third item. I am already on the copy and the send.
  24. L024TomasI can configure the exit condition. It is the same platform as the unsubscribe test.
  25. L025PriyaTomas, configure the sequence exit condition so converted trial users leave the sequence.
  26. L026TomasUnderstood. I will have that done before the warm-up finishes.
  27. L027PriyaAnything else that would stop us sending in the middle of the month?
  28. L028MayaOnly the analytics tagging, and that is already live in staging.
  29. L029TomasI checked it on Monday. The tags fire correctly on all three emails.
  30. L030PriyaThen we are clear. Let's not reopen the copy discussion in the channel afterwards.
  31. L031MayaUnderstood. I will treat the copy as frozen from now on.
  32. L032PriyaGood. That is everything for today, thank you both.

Tagged transcript baseline

Measured: Measured — archived experiment run

The blind comparison needs both outputs and a stored session seed. One of them is not available yet.

Summary

Extracted 2 tagged lines from 32 transcript lines: 1 tagged action and 1 tagged decision. This output covers tagged lines only. Untagged discussion is not summarised, and owners, dates and decision status are not inferred.

Decisions 1

  • the onboarding sequence goes out after the domain warm-up completes.

    StatusUnresolvedCited transcript lines: L010 — Priya: DECISION: the onboarding sequence goes out after the domain warm-up completes.

Actions 1

  • Maya will send the first onboarding email after warm-up completes.

    OwnerNot named in the transcriptDueNo dateStatusOpenCited transcript lines: L014 — Priya: ACTION: Maya will send the first onboarding email after warm-up completes.

Open questions 0

None recorded.

Uncertainties 1

  • This method does not determine whether a decision is current, superseded, or unresolved. Every tagged decision above is reported as unresolved.

    Cited transcript lines: L010 — Priya: DECISION: the onboarding sequence goes out after the domain warm-up completes.

Export

The file records what was compared, which versions produced it, and what it cannot be used to claim. Those limitations are part of the export itself, so they travel with it.

Your local data

Reading local storage… They hold the sample id, which output went in which panel, your choice, and your reason. There is no account, no server copy, and no analytics call.