mcp-kymograph
store findings.sqlite · schema 9 rules v4

notes-legacy × all-signals

undecided

The budget of 100 runs ran out with one identity whose rate sits between the hypotheses. Every other identity is decided. This series is a result: it is reported with its interval and never counted as new or gone.

transport
streamable-http · legacy · 2025-11-25
sessions
1 per run · 3 runs re-initialised after a 404
isolation
runc · ingress bridge
server_version
sha256:e0c41f…9a12
scenario
all-signals · sha256:77b0d3…41ee
digests
cmp 3f9a…c210 · full 81d2…07fb
scheduled
101
considered
100
complete
99
unknown (server may have caused)
1
retried (server could not cause)
1 of 5

Identities

Select one to see its runs and its test. Bars: worst-case Wilson interval, 95 %; tick: rate on complete runs.
Locator · signal Procedure hits / observed Rate, 0–100 % Class Statement {{it.proc}} {{it.hits}} / {{it.observed}}
+{{it.unknown}} unknown
{{it.loTxt}}{{it.rateTxt}}{{it.hiTxt}}
{{it.cls}} {{it.statement}}
8 more identities absent: below 10 % at β = 0.01 (sequential test, 48 runs) — show all

Runs, in repetition order

{{sel.signal}}
{{c.mark}}
hit no hit ? not a result — step_unanswered (stream_broken) ↻ re-run after interrupted

Selected identity

{{sel.locator}}
procedure
{{sel.proc}}
hits
{{sel.hits}} of {{sel.observed}} complete · {{sel.unknown}} unknown
rate
{{sel.rateTxt}} · worst case [{{sel.loTxt}}, {{sel.hiTxt}}]
class
{{sel.cls}}

{{sel.statement}}

Open a run with a hit →

Sequential test — log-likelihood ratio by run

H absent p₀ = 0.01 against H present p₁ = 0.10 · +ln(0.99/0.90) per run without a hit, ln(0.01/0.10) per hit
absent · A = ln(0.90/0.01) = 4.50 present · B = ln(0.10/0.99) = −2.29 budget · run 100
path for “absent” — the unknown run taken as a hit path for “present” — the unknown run taken as a miss {{chart.verdict}}