Standing analysis
Where the project stands, in plain language.
The same standing analysis as the German snapshot at /stat, in English and continuously updated. The explanatory parts are fixed and move only with an owner decision. Every number and every dated statement is resolved at build time from the generated status bundles, so a landing that changes a value changes this tab.
Sources: the Progress bundle for measured values, the proof ladder and the result cells, the curated day paragraph for the finding, and the canonical front mirror for what is open. The Progress view stays the single source of truth for direction and evidence.
Agent session results
The log.
Start with the bold result. Each card gives a plain summary first. The claim identifier and technical record stay available for readers who want to inspect the evidence.
- Verified means the declared measurement passed. It does not by itself mean a new capability shipped.
- Honest negative means the measurement said no and names the missing piece, so the same dead end does not need to be repeated.
- No teach means the scored question was not shown with its answer immediately before the test.
- Gamma ground is the registry certificate that a relation was grounded from independent field traces.
The whole picture
What is being built, what stands, and what has to move.
This is the single status view for direction and evidence. The former Outcome tab is now the goals and result board block inside this section. It shows the big picture, what already stands, the four programs with their result cells, and the dated movement, not a count of registry rows. A verified claim means a measurement was clean, not that the system can do something new.
- Acquire: the system forms the material itself.
- Cause: removing the mechanism removes the effect.
- Use: the result remains useful after a fresh boot.
- Transfer: it works on a separate surface without retuning.
- Move: the frozen result cell itself changes.
Every number below is resolved at build time from a committed run artifact, named by claim, registry entry and a pointer into that run. There is no hand maintained freshness text on this page: a landing that changes a number rebuilds the section. Counters above zero name the arm they came from, and claim counts, config keys and liveness numbers are deliberately absent.
Canonical work fronts
The fronts.
Every active research item belongs to a named front. The cards below are mirrored from the canonical front files, then cross checked against the verification registry and the public session log. The counts describe measurement records, not a count of capabilities.
Verification tree
The research map.
The map groups every registered experiment by verification track and family. Open a row to see the question and measured outcome, or follow its front link to see how it fits the wider program.
- Verified: the fixed measurement passed.
- Honest negative: the measurement failed and stays on record.
- Planned: registered, but not yet measured.
Generated directly from the verification registry. Claim totals are a record of experiments, not a progress score.
Research host
GPU occupancy.
Three GPUs on the research host, plus four DGX Spark worker lanes. Green means no active reservation on that device; red means a worker holds the GPU. Utilization and memory come from nvidia-smi; labels come from the shared reservation ledger. The Spark lanes show unified memory read from the free command, because nvidia-smi reports the total as N/A on those GB10 hosts.