Properties of a "research announcement"
This is an extended comment on history of research announcements. The principal mathematical journal which did this was Comptes Rendus published by the French Academy. It published (and still publishes) very short notes (1-2 pages), usually without complete proofs or with very short sketches of proofs. The papers were either by the members of the Academy, or approved by the members, and the time of publication was very short. Usually (but not always) several publications on the same topic were followed by a long paper, with complete proofs, covering the same material.
Soviet Doklady was analogous. Later, Bull Amer. Math. Soc. followed (it published surveys and research announcements). The Soviet analog, Uspekhi Mat. Nauk (translated as the Russian Math Surveys) also published research announcements, but they were not translated into English. Another similar publication was Abstracts of the talks presented to the AMS. It published very short (few lines) announcements based on the talks actually made in the AMS meetings.
The purpose of these publications was manifold. First, to secure priority. Second, to inform the experts, what results you obtained. (For a real expert in the area, a short announcement may be enough to understand what is going on). After you secure priority and inform people on what you have done, you may relax and spend several years of time on writing a complete definitive version.
Such publications, (at least in theory) were very quick: the papers of the members of the Academy were not refereed, while other papers were only endorsed by a member. (I remember how once I explained my result to Paul Malliavin, on a conference, and gave him a note. Next morning he told me that it is accepted:-)
If you look at the publication lists of such people as Picard, Poincare, Krein or Kolmogorov, you see that about 70% of their publications are CR/Doklady notes.
Of the very famous results published only as "research announcements" by their authors, let me mention KAM theory by Kolmogorov, and Krein's "theory of string". As far as I know these authors never published complete papers leaving this to others. Of more modern examples, the works of Sullivan and Douady -Hubbard which made the revolution in holomorphic dynamics were published as research announcements in Comptes Rendus in 1982. Sullivan published a complete version in 1985, when several alternative expositions based on his announcement were already available, while the work of Douady Hubbard is still available only as a preprint. On my opinion it was very important that interested people could learn about this breakthrough in 1982 three years before Sullivan published a complete paper. It was possible for an expert to understand his proof from the short announcement.
The papers of Yoccoz and Smirnov (for which Fields medals were awarded) were also published in Comptes rendus, without complete details.
EDIT. Nowadays, when we have the arXiv where we can publish anything, with no delay, the role of research announcements declines. On my opinion the arXiv performs both main functions of research announcements: quickly spreads the information and secures priority.
I think research announcements were more common in the past. Before the Internet. You send your paper to the journal, but since it may be more than a year before it appears in print, you also send an announcement to another journal. In mathematics, the announcement states the main results, without proof. But, of course, nowadays, when you send your paper to the journal you can also place it at arXiv.org so that the announcement is not needed.
Anecdotally: back in those days, some Soviet journals published research announcements, but corresponding papers never appeared subsequently, leading some in the West to suspect that: in fact there were no proofs known when the announcement appeared, but (as a sort of gamesmanship) that Soviet author wanted to discourage anyone else from working on the problem.
The question in the OP is one of best practices. This has been articulated by Jaffe and Quinn in this AMS bulletin:
Principle of best practice:
Research announcements should not be published, except as summaries of full versions that have been accepted for publication. Citations of unpublished work should clearly distinguish between announcements and complete preprints.
I notice that one journal that publishes announcements, Applied Mathematics Letters, follows this principle (albeit in a somewhat weakened form: "submitted" instead of "accepted") by requiring that:
The Research Announcements should be 3-4-page summaries of important results in a longer paper recently submitted to a leading journal by a well-established researcher.
Since priority issues can be settled by the arXiv submission, the purpose of announcements has shifted from achieving faster publication to reaching a broader audience. In physics the journals Physical Review Letters (PRL) and Physical Review aim for such a dual role, as described in this editorial:
The submission of an expanded version of an [announcement in] PRL to a topical Physical Review journal is an established practice that provides readers with easier access to important additional information. If authors simultaneously submit a Letter to PRL and a regular paper to one of the Physical Review journals, the two manuscripts are then reviewed coherently, typically by the same reviewers. If both papers receive favorable reviews, we aim to publish them at the same time, unless this leads to undue delay for one of them. We also ensure that the paper and the Letter cite each other.