Open Research Dataset

Why it is useful

Who	What the data lets you do
Finance and accounting researchers	Test whether firms that miss announced milestones show it in returns, guidance credibility, cost of capital or analyst revisions. Every entry has a permanent ID and, where known, the date its evidence became public; use only records with a known evidence date to limit look-ahead bias.
Management and strategy scholars	Study execution: which kinds of projects slip, how far, in which sectors, and whether restating a date early predicts a miss.
Law schools and governance researchers	Examine forward-looking statements, disclosure quality and the gap between what is said and what is done.
Computer science and NLP	A labelled corpus of corporate commitments: language strength (firm, expected, aspirational), restatements, and outcomes tied to evidence passages. Useful for training and evaluating models that read disclosures.
Students	A real, clean dataset for theses, coursework and competitions, with a question everyone understands.
Journalists and fact-checkers	Check a company's track record before repeating its next promise.

Research questions waiting for an answer

• When a company moves a deadline, how often does it then meet the new one?
• Do firm words (“will”) predict delivery better than soft ones (“aims”)?
• Are some sectors or project types systematically optimistic? By how many days?
• Does a history of missed milestones change how markets react to the next announcement?
• Do companies that lose key executives miss more of the commitments made under them?

What is in it, and what is not

The dataset holds every settled commitment: deadlines that have passed and been checked. Open deadlines (the forward Commitment Calendar), Miss Risk ratings and market-reaction data are licensed products and are not included. Entries withdrawn as outside the methodology are not included; their history is in the changes log. Outcomes that are not yet confirmed are included and marked, so you can choose to use confirmed outcomes only.

How each entry is made, checked and evidenced is set out in the methodology.

How to cite

Cite the dataset and the date you downloaded it, and cite individual commitments by their Kept ID, which never changes:

KEPT INDEX LLC (2026). KeptIndex Open Research Dataset. Retrieved 27 Oct 2026, from https://keptindex.org/research/

Frozen quarterly editions will be archived so that a published result can always be reproduced from the exact file used. If you publish with the data, tell us at contact@keptindex.org and we will list your work here.

For universities

• Full academic access. Faculty and doctoral researchers can request the complete record, including open deadlines and the version history of every restatement, free for non-commercial research.
• Student Verifier Program. Students can join as paid, part-time reviewers on the Verification Desk, checking outcomes against SEC filings. It is real research training, and the work is credited.
• Teaching pack. A ready-made class exercise: pick a company, read its filings, record its promises, compare with the record.

Data dictionary

Field	Meaning
ki_id	Permanent KeptIndex identifier; cite it, it never changes
company	Company name as filed
ticker	Ticker at the time of the build
cik	SEC Central Index Key
sector	GICS sector (S&P 500) or a broad sector from the SEC industry code
announced_date	Date the commitment was first announced
commitment	What the company said it would deliver, in plain words
company_words	The company's own sentence, matched word for word against the filing
deadline_as_stated	The time frame in the company's words
deadline	Last day of that time frame (ISO date)
event_type	Kind of milestone (product launch, facility, deal closing, regulatory filing and so on)
outcome	Met, Late, Not met, Partly met, Withdrawn, Superseded, Condition not met, No record found
delivered_date	When it was delivered, where the evidence states it
evidence_grade	A regulator/SEC, B company or exchange, C news, D other
evidence_url	The main source for the outcome
source_url	The filing the commitment came from
restated	Whether the company restated the commitment before its deadline
deadline_drift_days	Days between the original and latest deadline the company gave
wording_weakened	Whether the wording softened (firm → expected → aspirational)
checks_agree	Whether a second, independent AI check reached the same outcome blind (blank: not yet checked)
confirmed	Whether the outcome is confirmed (two agreeing checks on primary evidence, a tie-break on a positive outcome, or a person)
permalink	Permanent page for the entry