Round r68b
r68b
Source-reported narrativeRound README and results text are source material. Numeric values that are not tied to a claim ledger record remain source-reported.
Round source README · rounds/r68b/README.md
Round date: HISTORICAL (before 2026-10-09) Publication badge: DESCRIPTIVE; KEPT FOR AUDIT Why not VALID: The historical inventory retains DESCRIPTIVE because of COMPARATOR, NO_PREREG, UNMATCHED_EFFORT. See the round evidence and limitations below. Recomputation status: OBSERVED SOURCE ONLY
Recomputation update: The mined raw records now provide per-cell outcome, manifest token-counter, and wall-time recomputation for the observed cohort. This does not reconstruct missing planned cells, a full intention-to-treat denominator, or model execution.
Status
DESCRIPTIVE
- NO_PREREG — The README records no pre-registration or decision rule.
- UNMATCHED_EFFORT — Requested max was clamped to xhigh, while historical controls used different effort settings.
- COMPARATOR — Controls are historical rather than contemporaneous matched conditions.
Required reproduction metadata
- Kogen commit: not recorded in the cited round source.
- Harness commit: not recorded in the cited round source as a Git commit; any adapter or runtime fingerprints are not Git commits.
- Model and effort: a complete per-cell mapping is not recorded in the cited round source; values present in the source summary remain as stated above.
- Historical execution command: not recorded in the cited round source.
- Raw records: the cited sources do not provide a complete public round-specific request and grade bundle; public delivery and outcome exports are linked above where applicable.
Terminology: public round glossary.
STATUS: DESCRIPTIVE
Data completeness: 46.6% (24 deliveries; PASS WITH DECLARED GAPS). Measurement contract; declared gaps.
Pre-registered: no
Lifecycle: Historical capture. Snapshot 2026-10-05; earlier reports may use earlier cutoffs.
Question: How did the Kogen plan-shell arm perform on the held-out Elixir/Phoenix task panel?
Venue: Fixed eight-task HELDOUT kogen-bench-us/kogen-bench-eu placement.
Design: 8 tasks × 3 reps, historical same-task direct comparisons.
Decision rule: none predeclared located in supplied sources.
Arms: Current Kogen plan-shell. Historical direct controls are public r53 cells filtered to round=r53 and arm=r53:lmax (Luna max) or arm=r53:sol-high (Sol high); see the cells export.
Planned tasks in the surviving round note:
- elx-02-ingest-supervision
- elx-04-queue-backpressure
- syn-01-live-ticket-filters
- syn-06-migration-ticket-numbers
- syn-15-validation-ticket-form
- syn-20-email-invite-flow
- syn-24-csv-import
- syn-31-inbound-email-webhook
Verdict: Recomputed 6 October 2026 from the public cell export: current r68b Kogen plan-shell passed 21/24. The r53 direct Luna-max arm (arm=r53:lmax) passed 21/24 and ran at gpt-6-luna / max; the r53 direct Sol-high arm (arm=r53:sol-high) passed 22/24 and ran at gpt-6.1-sol / high. Kogen requested max was clamped to xhigh; these small, effort-confounded cohorts do not establish superiority.
Public delivery and outcome reconciliation
This captured-delivery summary uses the public run-record export, grouped by audit round, task ID, public host ID, and the exact exported arm label. Captured counts rows; graded counts rows marked graded=true; pass, fail, and other are the run-record outcome fields. Official outcome states are in cells.jsonl, whose classifications can differ; see the register source crosswalk. Planned n is shown only when source-reported; unknown allocations are not inferred.
| Task ID | Public host ID | Arm label in export | Planned n | Captured | Graded | Pass | Fail | Other graded outcome |
|---|---|---|---|---|---|---|---|---|
| elx-02-ingest-supervision | kogen-bench-us | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| elx-04-queue-backpressure | kogen-bench-eu | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-01-live-ticket-filters | kogen-bench-us | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-06-migration-ticket-numbers | kogen-bench-eu | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-15-validation-ticket-form | kogen-bench-eu | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-20-email-invite-flow | kogen-bench-us | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-24-csv-import | kogen-bench-eu | plan-shell | 3 | 3 | 3 | 3 | 0 | 0 |
| syn-31-inbound-email-webhook | kogen-bench-us | plan-shell | 3 | 3 | 3 | 0 | 3 | 0 |
Recomputed from raw records
`recomputed.json`, mined cell records, and the token audit back observed official outcomes, per-arm pass counts/rates, manifest token counters, and wall medians. The source contains 24 rows (fail 3, pass 21); overall pass rate is 87.5% (21/24) among pass/fail grades. Per-arm detail and coverage counts are in the JSON.
Published-export outcome cross-check: 24/24 shared cell IDs match; 0/24 public-export rows have no mined source record. Exact IDs and results for unpaired rows are in recomputed.json.
No per-cell outcome mismatches were found in the shared IDs.
Token reconciliation under METHOD.md: 24/24 published metadata totals equal the manifest input + cached input + output sum; 0 differ (no mismatches). Reasoning remains a separate reported component and is not added to output.
The recompute keeps both numbers visible. 0 unexplained single-response numeric errors were identified; multi-response rows retain the published full-cell total and the manifest counter separately because per-response usage records are excluded.
Captured run records
Standard-schema records assigned to this round in the indexed snapshot.
Reproduction inventory
Published analysis replay and full experiment replay are reported separately.
- Replay type
unavailable- Script
- No per-round analysis script is published.
- Inputs
- No identified public input bundle.
- Full execution replay
- Unavailable
- Scope
- No public per-round analysis script with a complete, identified input bundle is published. See the round README, MEASURED.md, and MISSING.md for recorded scope and gaps; historical execution inputs/toolchains/grading are not bundled.
View the complete per-round reproduction inventory · Repository reproduction guide