Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
95 commits
Select commit Hold shift + click to select a range
a76659f
Integrate stochastic histories with sequential density weights
yichao-liang Sep 13, 2026
e4335d3
Record failed density diagnostics and continued inference runs
yichao-liang Sep 13, 2026
34a914b
Add corrected defensive proposals for conditional velocity directions
yichao-liang Sep 13, 2026
79e6f68
Record guided integration and Domino prediction stability results
yichao-liang Sep 13, 2026
6981d8c
Add ordered process-safe target batches for offline inference
yichao-liang Sep 13, 2026
a2e0c3b
Mix global block refreshes into offline posterior exploration
yichao-liang Sep 13, 2026
f2fac5c
Validate point-start ablation and complete-checkpoint forecasts
yichao-liang Sep 13, 2026
3d2fd1e
Document causal Fan prefix support validation
yichao-liang Sep 13, 2026
f609cce
Record validated parallel Fan prefix inference launch
yichao-liang Sep 13, 2026
c7fa8cb
Validate Fan checkpoint forecasts and queue matched comparisons
yichao-liang Sep 13, 2026
16b5720
Add paired sequential carried-center comparison protocol
yichao-liang Sep 13, 2026
a5f41e2
Validate combined prediction assessment and record Fan carry control
yichao-liang Sep 13, 2026
8f0c003
Record completed mixed-proposal fits and unresolved concentration
yichao-liang Sep 13, 2026
021c542
Record mixed-proposal predictions and investigate local target scales
yichao-liang Sep 13, 2026
edd4328
Record initial-state comparisons and launch inference budget checks
yichao-liang Sep 13, 2026
0bf3cc5
Add earlier-program test of repeated-data prior carrying
yichao-liang Sep 13, 2026
c8d663a
Validate complete replica comparisons across inference budgets
yichao-liang Sep 13, 2026
e088302
Isolate prior carrying from mutable simulator reference defaults
yichao-liang Sep 13, 2026
b66ba12
Validate paired Fan forecasts from complete weighted histories
yichao-liang Sep 13, 2026
65de682
Record Fan prefix outcomes and queue population-size comparison
yichao-liang Sep 13, 2026
aa86877
Validate Fan cross-budget reports and queue reserved forecasts
yichao-liang Sep 13, 2026
b05f3f5
Record matched proposal and active prior-carry outcomes
yichao-liang Sep 13, 2026
64f9ad9
Launch repeated-dataset posterior coverage reference
yichao-liang Sep 13, 2026
9d9935c
Verify coverage reporting preserves planned dataset counts
yichao-liang Sep 13, 2026
1a88814
Prepare verified full-recording Fan population comparison
yichao-liang Sep 13, 2026
f4169d5
Record repeated-dataset coverage and first full Fan fit
yichao-liang Sep 13, 2026
0c52835
Derive constant-output likelihoods with correlated error
yichao-liang Sep 13, 2026
df2d0ff
Record full Fan population disagreement and total costs
yichao-liang Sep 13, 2026
51ce711
Validate fixture proposal correction and launch matched Fan fits
yichao-liang Sep 13, 2026
9838557
Prepare canonical-scene forecast validation for guided fits
yichao-liang Sep 13, 2026
ec079f9
Validate matched Fan proposal comparison and gate forecasts
yichao-liang Sep 13, 2026
34ca72d
Validate alternative tempering and recover guided forecast runtime
yichao-liang Sep 13, 2026
3045226
Validate tempered forecasts and queue six-population comparison
yichao-liang Sep 13, 2026
c857fa3
Identify numerical dispatch behind coordinate verification mismatch
yichao-liang Sep 13, 2026
9aeac7b
Record completed Domino budget comparison and unresolved instability
yichao-liang Sep 13, 2026
14dd185
Diagnose Domino proposal movement across scales and jitter
yichao-liang Sep 13, 2026
bf8ca2a
Record first completed guided Fan forecast
yichao-liang Sep 13, 2026
2922f14
Record completed Fan budget and proposal comparisons
yichao-liang Sep 13, 2026
56d98e3
Attribute Domino score instability to robot transients
yichao-liang Sep 13, 2026
5abcae5
Validate Domino joint-transition model and launch matched inference
yichao-liang Sep 13, 2026
298523d
Validate weighted Domino transition forecasts and queue comparison
yichao-liang Sep 13, 2026
d0c7068
Audit historical Bridge and Boil dynamics as separate inference controls
yichao-liang Sep 13, 2026
0e2e614
Validate Boil joint discrepancy and prefix fixture-state controls
yichao-liang Sep 13, 2026
4854747
Add density-corrected fixture proposals for offline uncertainty infer…
yichao-liang Sep 13, 2026
fc728cb
Preserve unobserved joint uncertainty and verify Boil fixture support
yichao-liang Sep 13, 2026
f1d7fa3
Record verified Boil joint-state support and literal rule reuse
yichao-liang Sep 13, 2026
4affabe
Record Domino transition-model disagreement and budget follow-up
yichao-liang Sep 13, 2026
4a1c6f5
Document verified Boil full-scene support and parameter-independent r…
yichao-liang Sep 13, 2026
1f36f99
Check budget-specific Domino runtime identities independently
yichao-liang Sep 13, 2026
206a864
Validate canonical Boil initialization and launch prefix inference
yichao-liang Sep 13, 2026
cd29e87
Record first completed Balloons joint-inference pilot
yichao-liang Sep 13, 2026
4654d06
Document validated canonical Boil forecasts and paired assessment
yichao-liang Sep 13, 2026
2a82554
Add verified subclass Boil incumbent comparison protocol
yichao-liang Sep 13, 2026
3c4f3e9
Document verified prefix-only Balloons inference pilots
yichao-liang Sep 13, 2026
1515f09
Record verified Boil incumbent held-out predictions
yichao-liang Sep 13, 2026
3025b3b
Document verified prefix-only Balloons future forecasts
yichao-liang Sep 13, 2026
d6e1e18
Add verified Balloons population and incumbent forecast protocols
yichao-liang Sep 13, 2026
1bc7a11
Diagnose structural glue-law and face-selection errors in Bridge
yichao-liang Sep 13, 2026
5243d2e
Evaluate prefix-fitted Bridge deposition geometry hypotheses
yichao-liang Sep 13, 2026
cadc483
Validate revised Bridge glue dynamics in causal native replay
yichao-liang Sep 13, 2026
aa9e5c2
Record completed Balloons prefix fits and verified incumbent forecast
yichao-liang Sep 13, 2026
de2a322
Verify Bridge subclass parity and exact glue-rate conditioning
yichao-liang Sep 13, 2026
8acf3f5
Freeze and queue the revised Bridge incumbent comparison
yichao-liang Sep 13, 2026
9326292
Record completed larger Domino fits and active forecast checks
yichao-liang Sep 13, 2026
b519262
Document verified Bridge reset-scene prior and long replay audits
yichao-liang Sep 13, 2026
6a6db14
Record completed uncertainty forecasts and unresolved prediction regr…
yichao-liang Sep 13, 2026
f55e7fa
Add shared variance inference after isolating Boil forecast perturbat…
yichao-liang Sep 13, 2026
383e8cc
Update uncertainty validation status and completed Balloons comparison
yichao-liang Sep 13, 2026
72e6cec
Record shared-variance forecast validation and Bridge conditioning audit
yichao-liang Sep 13, 2026
d47bc3a
Add prior-preserving categorical guidance and document Bridge support…
yichao-liang Sep 14, 2026
a414f8b
Record validated Bridge orientation support and remaining inference g…
yichao-liang Sep 14, 2026
d486a95
Add budgeted joint-support initialization for offline inference
yichao-liang Sep 14, 2026
73de3f9
Record Bridge inference validation and Boil forecast comparison
yichao-liang Sep 14, 2026
b962d82
Validate Bridge joint forecast adapter and reserved assessment
yichao-liang Sep 14, 2026
7681ecf
Validate and queue weighted Bridge posterior forecasts
yichao-liang Sep 14, 2026
f355822
Reproduce Boil inference collapse and validate supported fitting pilots
yichao-liang Sep 14, 2026
75fabb2
Validate weighted forecasts for supported Boil inference
yichao-liang Sep 14, 2026
1212d14
Verify Fan speed ambiguity against complete forecast populations
yichao-liang Sep 14, 2026
4dad9e8
Isolate Fan fixture uncertainty and validate static sensor inference
yichao-liang Sep 14, 2026
860d2ca
Validate and queue weighted Fan static-fixture forecasts
yichao-liang Sep 14, 2026
f0ea440
Record posterior prediction diagnostics and recover Bridge fits
yichao-liang Sep 14, 2026
f90aa56
Diagnose Balloons inference errors and test default-centered guidance
yichao-liang Sep 14, 2026
63f1604
Diagnose Balloons parameter concentration and test selective guidance
yichao-liang Sep 14, 2026
720c1b8
Validate selective Balloons posterior forecasts and queue comparisons
yichao-liang Sep 14, 2026
388c622
Integrate unobserved Boil heating parameters in weighted forecasts
yichao-liang Sep 14, 2026
4c32425
Represent independent prior factors and validate reduced Boil target
yichao-liang Sep 14, 2026
d160986
Validate reduced Boil forecasts with independent heating priors
yichao-liang Sep 14, 2026
fc0d905
Finalize noisy five-domain sweep results
yichao-liang Sep 14, 2026
137e83b
Document continual baseline integrity results
yichao-liang Sep 14, 2026
a6036bf
Archive continual comparison scorecards
yichao-liang Sep 14, 2026
8317fd1
Add continual comparison figure renderer
yichao-liang Sep 14, 2026
7c47be3
Clarify replay and simulator discrepancy migration requirements
yichao-liang Sep 14, 2026
c93fe80
Validate conditional Boil heating inference and record completed comp…
yichao-liang Sep 14, 2026
dbc1a3a
Type and lint the continual comparison figure renderer
yichao-liang Sep 14, 2026
d33fac2
Merge master into pr/11 (history reconciliation after #179 squash)
yichao-liang Sep 15, 2026
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
135 changes: 135 additions & 0 deletions docs/comparisons/bridge-balloons-integrity-results.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,135 @@
{
"generated_by": "/home/ycliang/predicators/logs/bridge_balloons_integrity_20260912/report.py",
"updated_at": "2026-09-13T09:07:21.054343+00:00",
"source_commit": "cad1000f92c5e5e94714ae613426b89761823ee0",
"rows": [
{
"domain": "Bridge",
"approach": "agent_continual",
"arm": "MB + uncertainty",
"seed": 0,
"job": "22646173_0",
"job_state": "COMPLETED",
"status": "Finished",
"scorecard": "/home/ycliang/predicators/logs/agent_continual/bridge-agent_continual_span_transfer_r1/seed0/run_20260912_165804/scorecard.json",
"finished": true,
"success": true,
"wins": 2,
"levels": 2,
"steps": 2939,
"resets": 0,
"end_reason": "all_levels_won",
"quota_message": null
},
{
"domain": "Bridge",
"approach": "agent_continual_model_free",
"arm": "MF direct coding agent",
"seed": 0,
"job": "22646174_0",
"job_state": "COMPLETED",
"status": "Finished",
"scorecard": "/home/ycliang/predicators/logs/agent_continual_model_free/bridge-agent_continual_model_free_span_transfer_r1/seed0/run_20260912_165804/scorecard.json",
"finished": true,
"success": true,
"wins": 2,
"levels": 2,
"steps": 2634,
"resets": 0,
"end_reason": "all_levels_won",
"quota_message": null
},
{
"domain": "Balloons",
"approach": "agent_continual_model_free",
"arm": "MF direct coding agent",
"seed": 0,
"job": "22649185_0",
"job_state": "COMPLETED",
"status": "Finished",
"scorecard": "/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed0/run_20260912_180112/scorecard.json",
"finished": true,
"success": true,
"wins": 3,
"levels": 3,
"steps": 771,
"resets": 1,
"end_reason": "all_levels_won",
"quota_message": null
},
{
"domain": "Balloons",
"approach": "agent_continual_model_free",
"arm": "MF direct coding agent",
"seed": 1,
"job": "22649185_1",
"job_state": "COMPLETED",
"status": "Finished",
"scorecard": "/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed1/run_20260912_180112/scorecard.json",
"finished": true,
"success": true,
"wins": 3,
"levels": 3,
"steps": 554,
"resets": 1,
"end_reason": "all_levels_won",
"quota_message": null
},
{
"domain": "Balloons",
"approach": "agent_continual_model_free",
"arm": "MF direct coding agent",
"seed": 2,
"job": "22649185_2",
"job_state": "COMPLETED",
"status": "Finished",
"scorecard": "/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed2/run_20260912_180113/scorecard.json",
"finished": true,
"success": true,
"wins": 3,
"levels": 3,
"steps": 829,
"resets": 2,
"end_reason": "all_levels_won",
"quota_message": null
}
],
"aggregate": [
{
"domain": "Bridge",
"arm": "MB + uncertainty",
"finished": 1,
"expected": 1,
"successes": 1,
"solve_rate": 1.0,
"level_solve_rate": 1.0,
"steps_mean_success": 2939,
"steps_n": 1,
"resets_mean_finished": 0
},
{
"domain": "Bridge",
"arm": "MF direct coding agent",
"finished": 1,
"expected": 1,
"successes": 1,
"solve_rate": 1.0,
"level_solve_rate": 1.0,
"steps_mean_success": 2634,
"steps_n": 1,
"resets_mean_finished": 0
},
{
"domain": "Balloons",
"arm": "MF direct coding agent",
"finished": 3,
"expected": 3,
"successes": 3,
"solve_rate": 1.0,
"level_solve_rate": 1.0,
"steps_mean_success": 718,
"steps_n": 3,
"resets_mean_finished": 1.3333333333333333
}
]
}
25 changes: 25 additions & 0 deletions docs/comparisons/bridge-balloons-integrity-results.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
<!-- Generated by logs/bridge_balloons_integrity_20260912/report.py; do not edit. -->
# Bridge transfer and sustained-hover Balloons pilots

Updated 2026-09-13T09:07:21.054343+00:00.

Bridge: one seed per MB/MF arm, three-block train span and four-block test span. Balloons: three MF seeds, original non-hatch task distribution, 25-step sustained hovering. Historical cohorts and other comparisons remain separate.
Whole-run success means winning every training and test level.
Solve rates and mean resets use finished agent seeds only.
Mean steps uses only whole-run successes, with the qualifying count shown.
Unfinished, setup, and infrastructure outcomes are excluded from averages.
These tables are refreshed from scorecards and Slurm accounting.

| Domain | Arm | Finished | Whole-run solve | Level solve | Mean successful steps (n) | Mean resets |
|---|---|---:|---:|---:|---:|---:|
| Bridge | MB + uncertainty | 1/1 | 100.0% (1/1) | 100.0% | 2,939 (n=1) | 0 |
| Bridge | MF direct coding agent | 1/1 | 100.0% (1/1) | 100.0% | 2,634 (n=1) | 0 |
| Balloons | MF direct coding agent | 3/3 | 100.0% (3/3) | 100.0% | 718 (n=3) | 1.3 |

| Domain | Arm | Seed | Status | Wins | Steps | Resets | Job | Scorecard |
|---|---|---:|---|---:|---:|---:|---|---|
| Bridge | MB + uncertainty | 0 | Finished | 2/2 | 2,939 | 0 | 22646173_0 | [scorecard](/home/ycliang/predicators/logs/agent_continual/bridge-agent_continual_span_transfer_r1/seed0/run_20260912_165804/scorecard.json) |
| Bridge | MF direct coding agent | 0 | Finished | 2/2 | 2,634 | 0 | 22646174_0 | [scorecard](/home/ycliang/predicators/logs/agent_continual_model_free/bridge-agent_continual_model_free_span_transfer_r1/seed0/run_20260912_165804/scorecard.json) |
| Balloons | MF direct coding agent | 0 | Finished | 3/3 | 771 | 1 | 22649185_0 | [scorecard](/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed0/run_20260912_180112/scorecard.json) |
| Balloons | MF direct coding agent | 1 | Finished | 3/3 | 554 | 1 | 22649185_1 | [scorecard](/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed1/run_20260912_180112/scorecard.json) |
| Balloons | MF direct coding agent | 2 | Finished | 3/3 | 829 | 2 | 22649185_2 | [scorecard](/home/ycliang/predicators/logs/agent_continual_model_free/balloons-agent_continual_model_free_dwell_r1/seed2/run_20260912_180113/scorecard.json) |
Loading
Loading