mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-29 07:34:31 +00:00
zeekayandHanzo Dev
07694f8e88
Describes /v1/graph at hanzo-inc/cloud 1fffdf408: one append-only table of assertions, a server-derived knowable instant, reads at a valid and a transaction instant, versions that correct each other only when learned later, declared cardinality and schema as assertions, diff, path, erasure with a receipt, Louvain communities and a citation-checked answer. CronQuestions from hanzoai/benchmarks: 99.88% Hits@1 for the control, semantica 0.7.0 and the store, question for question; 0.063 ms against 5,357 ms per call. The first run's 70.03% is reported with its two causes and the two defects the harness found in the fix. LongMemEval-S is included as a measurement of the retrieval engine, not of the store. Co-authored-by: Hanzo Dev <[email protected]>
56 lines
2.9 KiB
TeX
56 lines
2.9 KiB
TeX
\section{Conclusion}
|
|
\label{sec:conclusion}
|
|
|
|
The store keeps one table of assertions and derives every answer from it. Each
|
|
statement carries its own valid interval; the transaction instant is derived on
|
|
the server as the later of the filer's claim and the write; versions correct
|
|
each other only when learned later; a relation holds one value or several as
|
|
the organization declares; and every read names a valid and a transaction
|
|
instant. From those definitions, an answer as known at an instant is unchanged
|
|
by later writes other than erasure, a filer cannot place a row earlier in
|
|
transaction time than its write, and, for a many-valued pair without
|
|
retractions whose versions were learned together, a valid-time answer does not
|
|
depend on when the facts were filed.
|
|
|
|
On CronQuestions, the store, a dictionary control and semantica~0.7.0 give the
|
|
same answer to each of \res{cron/n} covered questions, \res{cron/kg/all}\%
|
|
Hits@1, whether the history is filed at one instant or as it happened; the
|
|
store answers a call in \res{cron/wire/p50}\,ms at the median against
|
|
semantica's \res{cron/sem/p50}\,ms. The previous version of the store scored
|
|
\res{cron/first/wire/all}\% on the same lane, and the reasons are specific: its
|
|
one read instant was a transaction time, and a retraction ended every value of
|
|
its pair. The same harness, run against the correction, found two more defects
|
|
before the change was merged.
|
|
|
|
What remains is mostly retention: an operation that sets a litigation hold, one
|
|
that disposes of rows under an organization's policy, and erasure that reaches
|
|
the other stores holding what an organization says about a subject.
|
|
|
|
\section*{Reproducing}
|
|
|
|
\begin{lstlisting}[language=bash]
|
|
# the store: github.com/hanzo-inc/cloud at 1fffdf408
|
|
go test ./apps/graph/ ./claim/
|
|
|
|
# CronQuestions: hanzoai/benchmarks at 57a55cf
|
|
bash cronquestions/run.sh # the test split
|
|
bash cronquestions/run.sh valid # the validation split
|
|
# the first run, as committed
|
|
git show 2e6d394:cronquestions/runs/hanzo-wire-test/metrics.json
|
|
|
|
# LongMemEval-S: the same repository, from brain/
|
|
EMBED=all-minilm node longmemeval/retrieve.mjs # cosine
|
|
node longmemeval/engine.mjs --embed=minilm # engine
|
|
\end{lstlisting}
|
|
|
|
\noindent \texttt{run.sh} fetches CronQuestions \texttt{data\_v2.zip} (SHA-256
|
|
\texttt{0a2f7714\ldots fd4734e}) if it is absent, builds the fact file and the
|
|
plan, runs the control, compiles the store driver from the cloud source with
|
|
\texttt{go test -overlay}, runs semantica from its own checkout in a separate
|
|
environment, and scores every row. Each run directory records the subject's
|
|
commit, the host and the load average it ran at. The values in this paper are
|
|
read from those files into \texttt{results-data.tex}. The derived values there
|
|
--- the repeated-statement counts, the ratio of medians and the sign tests ---
|
|
are computed from the fact file the lane builds and from the reported counts,
|
|
and are marked as such.
|