Docs
Living research docs, meeting + mentor updates, activity logs, and ideas. Sources: docs/*.md, docs/mentor_updates/, logs/daily + logs/weekly, docs/ideas/.
Meetings & mentor updates (13)
- Async mentor note — June 1, 2026Dan → research-log channel · Explore Persona Space. Captured June 2.mu__2026-06-02· updated 2026-09-03
- Meeting notes — Christina, June 11, 2026Raw notes, captured same day. Not yet triaged into tasks — see the todo tomu__2026-06-11-christina· updated 2026-09-03
- Mentor Meeting — 2026-05-22Mentor: Dan Mossingmu__2026-05-22· updated 2026-09-03
- Mentor Notes — 2026-05-26Mentor: Dan Mossingmu__2026-05-26· updated 2026-09-03
- Mentor Notes — 2026-05-29Mentor: Dan Mossingmu__2026-05-29· updated 2026-09-03
- Mentor Update — 2026-05-11A conditional behavior is a behavior the model produces whenever a specific signal — call it the trigger — appears in the input (or in the context preceding it). The trigger can take many forms, and the same structura…mu__2026-05-11· updated 2026-09-03
- research-log-exploring-persona-space — full channel historyThomas Jiralerspong has joined the channelmu__2026-05-27-research-log-channel-snapshot· updated 2026-09-03
- Result: "User:/Assistant:" is the chat map in different coordinates; narrative framing is what collapses it The last update left a tension: the assistant context→answer map holds up in plain "User:/Assistant:" transcripts with no chat template, but collapses for generic stories ([#825](https://eps.superkaiba.com/tasks/825)…mu__2026-07-21_framing_ablations_results· updated 2026-09-03
- Result: One context→answer operator — framing bends its coordinates, speaker identity barely moves itCombined writeup (2026-07-23): merges the 2026-07-21 framing-ablations update with the identity half (#1310, #1639, #1335) and the assistant direct test that landed tonight. Everything here is committed and verified e…mu__2026-07-23_one_operator_framing_vs_identity· updated 2026-09-03
- Result: The context→answer mapping is inherited from pretraining An earlier experiment found a linear mapping from a single context vector to the mean answer vector in the instruct model (R² ~0.7, [#779](https://eps.superkaiba.com/tasks/779))mu__2026-07-14_context_answer_map_results· updated 2026-09-03
- Weekly mentor meeting — June 11, 2026Dan 1:1 · Explore Persona Space. Source: "Research Log" Google Slides deckmu__2026-06-11· updated 2026-09-03
- Weekly mentor meeting — June 4, 2026Dan 1:1 (+ Christina) · Explore Persona Space. Source: "Research Log" Googlemu__2026-06-04· updated 2026-09-03
- Weekly update — May 28, 2026Thomas to Dan · Explore Persona Spacemu__2026-05-28· updated 2026-09-03
Activity (70)
- Daily — 2026-05-27 (0 promoted;- #398 (step-75 marker cliff is a sampling-threshold artifact) — completed the full /issue loop: upload-verification round 1 FAIL → round 2 PASS, pod terminated, analyzer drafted clean-result, interp-critic round 1 RE…daily__2026-05-27· updated 2026-09-03
- Daily — 2026-05-28 (#390 +- #390 promoted → completed (MODERATE). Clean-result: refusal-style negatives install a persona gate that generalises across 9 of 11 OOD eval framings as a verbatim one-line refusal substitution. Two user-directed bod…daily__2026-05-28· updated 2026-09-03
- Daily — 2026-05-29 (#404 +- Two clean-result drafts landed at awaitingpromotion (pending your promotion): #404 — base-model narrow↔broad cosine predicts post-SFT broad-misalignment rate (ρ=+0.75, MODERATE); #385 — [ZLT] marker leakage emerges …daily__2026-05-29· updated 2026-09-03
- Daily — 2026-05-31 (on-policy marker re-run lands rank- Clean-result spec migrated to v2. #454 moved the body spec to the 2-content-section nested design (Human TL;DR / TL;DR with ### Motivation + ### What I ran + ### Findings→#### per result / Reproducibility; confiden…daily__2026-05-31· updated 2026-09-03
- Daily — 2026-06-01 — #448 on-policy re-run lands NULL; idle-pod burn recurred 3× (~$1k); 0 promoted- #448 dominated the day (70 commits). The user halted the off-policy version mid-flight ("STOP — do not promote #448, its DV is off-policy") and forced a full re-run with the on-policy design. The 12-cell contrastive…daily__2026-06-01· updated 2026-09-03
- Daily — 2026-06-02 (6 experiments parked at awaiting_promotion; 9 workflow fixes auto-landed)- Six experiments reached a clean-result and parked at awaitingpromotion today: #444 (four fact-teaching leakage shapes + a 5-way re-judge), #464 (chat-template role-header marker localization), #466 (hypothesis-led "…daily__2026-06-02· updated 2026-09-03
- Daily — 2026-06-03 (10 experiments parked at awaiting_promotion, 6 workflow fixes auto-shipped, 0 promoted)- High experiment throughput. Ten experiment tasks reached awaitingpromotion today: #405 (marker leakage vs K source personas), #444 (invented-fact localization, 4th fact regime + 5-way re-judge), #464 (persona-in-rol…daily__2026-06-03· updated 2026-09-03
- Daily — 2026-06-04 (#490 → awaiting_promotion; #489 blocked at SP-identity gate; 3 workflow fixes shipped)- #490 (dose-matched midpoint coupling) went proposed → awaitingpromotion in a single day: planned, approved, ran a full 7-pair × 3-seed sweep on a 4×H100 pod, then through the analyzer / interpretation-critic / clean…daily__2026-06-04· updated 2026-09-03
- Daily — 2026-06-05 (heavy autonomous run; Claude-access + RunPod-balance walls; 3 results reached awaiting_promotion, 0 promoted)- Big autonomous-experiment day. ~9 /issue sessions ran in parallel (#489, #496, #498, #501, #502, #503, #504, #505, #506). Three reached awaitingpromotion (clean-result drafted + critic PASS): #498 (custom chat-templ…daily__2026-06-05· updated 2026-09-03
- Daily — 2026-06-06 (5 autonomous experiments in flight; RunPod cap + smoke-coverage gaps dominate; 0 promoted)- Heavy autonomous-/issue day, mostly machine-driven. Five experiments running at day's end (#501, #503, #506, #507, #508), three parked blocked awaiting a Thomas decision (#504, #505, #509). Three tasks reached await…daily__2026-06-06· updated 2026-09-03
- Daily — 2026-06-07 (heavy autonomous fleet day; 4 fleet-enforcement fixes landed,- Heavy autonomous fleet day. 7 experiments running concurrently overnight (#488, #501, #503, #504, #505, #507, #511), plus #509 advancing to interpreting and #506 ending blocked. ~522 git commits in the PT window. Pe…daily__2026-06-07· updated 2026-09-03
- Daily — 2026-06-08 (compute-infra expansion + ~40 in-session workflow fixes; 0 promotions)- Compute-infrastructure expansion (the day's headline). Brought three new compute backends into reach: Mila cluster SSH access (public key added, IT ticket filed), Compute Canada / DRAC (alliancecan, Duo MFA + robot …daily__2026-06-08· updated 2026-09-03
- Daily — 2026-06-09 (96 results promoted; 25 workflow fixes auto-applied)- Promotion-backlog clear-out: 96 clean-results promoted to completed (batches of 12 + 53 + 8 + singles; #407 flipped to not-useful after the batch — its body starts "BUGGED experiment" — and #519/#520 judged not-usef…daily__2026-06-09· updated 2026-09-03
- Daily — 2026-06-10 (40 tasks filed, ~20 parked for promotion; 16 workflow fixes auto-applied; 3 promoted)- Highest-throughput day yet: 40 new tasks created (#551–#590), 4,850 commits on main. ~20 tasks reached awaitingpromotion during the day (the column now holds 25); 10 are running, 2 interpreting, 2 in same-issue foll…daily__2026-06-10· updated 2026-09-03
- Daily — 2026-06-11 (19 tasks filed, 15 results parked, 18 follow-up rounds; 2 workflow fixes auto-applied)- Highest-throughput day on record for the pipeline. 19 tasks filed (#595–#613), 15 experiment results reached awaitingpromotion for the first time (#491, #556, #561, #570, #571, #585, #591, #594, #600, #601, #602, #6…daily__2026-06-11· updated 2026-09-03
- Daily — 2026-06-12 (heavy autonomous pipeline day; ~30 workflow fixes auto-applied in-flight, 0 promotions)- Fully autonomous, very high-throughput day — 2384 commits across the fleet. Multiple per-issue sessions ran the whole /issue pipeline concurrently: #613, #621, #622, #627, #628, #633, #483, #545, #522. #633 reached …daily__2026-06-12· updated 2026-09-03
- Daily — 2026-06-13 (3 pipelines + heavy interactive day; 7 auto-fixes applied)- Three autonomous /issue pipelines ran the full clean-result stack: #632 and #634 reached completed (merged to main); #595 (prefix-carrier binding strength, determinate null) reached awaitingpromotion. #640 filed as …daily__2026-06-13· updated 2026-09-03
- Daily — 2026-06-15 (21 results promoted; orphan task folder + SKILL.md merge-guard fixed)- Big evening promotion sweep — 21 clean-results promoted, all classified useful. Thomas worked through the awaitingpromotion queue thread by thread: the role-header / marker-localization line (#464, #533, #546, #611)…daily__2026-06-15· updated 2026-09-03
- Daily — 2026-06-16 (5 results promoted; 4 workflow fixes auto-applied)- Heavy autonomous day across ~11 active issues (542, 597, 612, 640, 641, 642, 647, 649, 651, 653, 654). 5 clean-results promoted useful: #528, #556, #608, #613, #614.daily__2026-06-16· updated 2026-09-03
- Daily — 2026-06-17 (4 follow-up loops re-parked,- Heavy autonomous fleet day, mostly self-healing. Four same-issue follow-up loops completed and re-parked at awaitingpromotion: #642 (onpolicy-matchedlr-rank-isolation — interp + clean-result ensembles both PASS), #5…daily__2026-06-17· updated 2026-09-03
- Daily — 2026-06-18 (5 follow-up rounds parked at awaiting_promotion; analyzer self-check tightened)- Fully autonomous day — no live user chat. Five same-issue follow-up rounds ran end-to-end and parked at awaitingpromotion for user promotion: #641 (matched-recipe EM-resistant-context install, 8-seed re-fold), #657 …daily__2026-06-18· updated 2026-09-03
- Daily — 2026-06-19 (idle day, no session activity)- Idle day. No Claude Code session activity, no commits, and no task events occurred during the PT 06-19 window (07:00Z 06-19 → 06:27Z 06-20). The only transcript stamped in-window is this /daily run itself.daily__2026-06-19· updated 2026-09-03
- Daily — 2026-06-20 (quiet research/synthesis day; no experiments committed)- Quiet day on the project: no commits, no task-state changes in the repo since 2026-06-18 evening. All of today's activity was research-thinking + result-synthesis chat (5 sessions, all in the 22:38–23:27 PT window).daily__2026-06-20· updated 2026-09-03
- Daily — 2026-06-21 (PM dispatched 5 autonomous runs; all parked at awaiting_promotion)- PM session triage + dispatch. Archived #655, restarted the stalled #651 session by hand, and spawned autonomous --auto sessions for #657, #559, #642, #641 — the strongest follow-ups off the 52-deep awaiting-promotio…daily__2026-06-21· updated 2026-09-03
- Daily — 2026-06-22 (quiet day; no project activity, 1 skill fix applied)- Quiet day on the project: no commits, no task events.jsonl mutations, and no genuine interactive Claude Code sessions on 2026-06-22 (PT). Last real activity was 2026-06-18 (#657).daily__2026-06-22· updated 2026-09-03
- Daily — 2026-06-23 (marker loss-mask default change +- Theory-driven experiment-design session (d419f9fd): read the leakage-theory doc, fanned out ~16 verification subagents over its assumptions, produced docs/leakagetheorytestsuite.md (33KB) covering marker / sycophanc…daily__2026-06-23· updated 2026-09-03
- Daily — 2026-06-24 (clean-result v4 redesign shipped;- Clean-result spec migrated to v4 (four-flat-H2: Takeaways / Goal / Methodology / Results + Repro/Context footer; methodology doc = mechanical copy of the body's ## Methodology). Shipped via workflow-fix merges + a v…daily__2026-06-24· updated 2026-09-03
- Daily — 2026-06-25 (autonomous theory-program day;- The autonomous overnight leakage-predictor assumption-testing program drove the whole day (sequenced /issue --auto rollout, Thomas asleep then research-Q&A by day). Four tasks ran in parallel: #664 (Phase 2 fine-tun…daily__2026-06-25· updated 2026-09-03
- Daily — 2026-06-26 (clean-result paper-mode redesign; GCP capacity + VM-disk firefighting; 3 fixes auto-applied)- Clean-result format → LaTeX paper-mode (#657): Thomas drove a multi-step redesign — drop the standalone methodology doc, fold it into Goal/Methodology/Results, require low-level labeled-scatter data plots, demand se…daily__2026-06-26· updated 2026-09-03
- Daily — 2026-06-27 (#664 vLLM-crash saga;- #664 vLLM EngineCore-crash saga dominated the day: ~8 GPU launches / 4 strategy pivots against a reproducible hang inside vLLM 0.11.0's first llm.generate() on the elicitsecurecode path (across pods, batch sizes, ea…daily__2026-06-27· updated 2026-09-03
- Daily — 2026-06-28 (workflow-automation push lands;- Workflow-automation push landed answering today's repeated "I barely run /daily, /weekly — automate this": #711 (deterministic consolidatelessons.py cron + daily healthcheck), #713 (fold /weekly's load-bearing check…daily__2026-06-28· updated 2026-09-03
- Daily — 2026-06-29 (huge autonomous infra/workflow day; 38 infra tasks completed, 9 route-2 fixes filed)- Massive autonomous infra/workflow push: ~38 kind: infra tasks reached completed, including the structural disk fix #681 (dedicated GCP data disk + ext4 per-task quotas), #711 (deterministic nightly lesson-consolidat…daily__2026-06-29· updated 2026-09-03
- Daily — 2026-06-30 (35 tasks created, ~33 merged; big autonomous workflow-fix day)- Huge autonomous workflow-fix day. ~35 tasks created and ~33 merged to main, almost all kind: infra workflow-fixes (#741–#777) running the full planner → critic ensemble → implementer → Claude+Codex code-review → Ste…daily__2026-06-30· updated 2026-09-03
- Daily — 2026-07-01 (fleet-scale workflow-fix day; 0 promotions)- Heaviest pipeline day yet: 1,988 workflow markers across 61 tasks (#658–#824), ~2,350 commits on main, 0 clean-results promoted (5 tasks newly parked at awaitingpromotion: #778, #779, and the #810 family in flight).daily__2026-07-01· updated 2026-09-03
- Daily — 2026-07-02 (backfill — 3 clean-results parked, 79 tasks filed, 0 promoted)- Heaviest fleet day on record: 143 sessions active, 3,881 workflow markers across 100 tasks, 79 tasks created (#825–#903) — 40 of them workflow-fix filings from 12 parent tasks, most implemented and merged to main th…daily__2026-07-02· updated 2026-09-03
- Daily — 2026-07-03 (48 completions incl. the workflow-fix wave; API-key outage; 56-item problem sweep)- Workflow-fix wave landed: 48 tasks reached completed, dominated by the #930–#959 infra band (plus #955/#956/#957/#959/#960/#961 lifecycle work into the evening). No clean-results were promoted today.daily__2026-07-03· updated 2026-09-03
- Daily — 2026-07-04 (10 experiments to awaiting_promotion, 66 infra fixes landed, 0 promoted)- Huge autonomous throughput day: 4,143 commits across 113 tasks. 10 experiment lines reached awaitingpromotion (#742, #810, #825, #841, #922, #923, #928, #931, #952, #958) and ~66 tasks reached completed — almost all…daily__2026-07-04· updated 2026-09-03
- Daily — 2026-07-05 (0 promoted; 2 fixes applied, 10 filed for review)- Three context→answer-map results parked at awaitingpromotion: #952 (own-answer advantage is position-uniform) and #811 (taught-fact chain follows answer content tokens) finished their follow-up rounds, and #923 (att…daily__2026-07-05· updated 2026-09-03
- Daily — 2026-07-06 (infra wave — ~15 fixes landed;- Infra/workflow-fix wave landed. Nearly all of yesterday's 10 daily-filed fixes completed end-to-end through the full pipeline today: #1075 (verifyplan GPU-basis), #1076 (test regressions on main), #1077 (step9c dirt…daily__2026-07-06· updated 2026-09-03
- Daily — 2026-07-07 (27 workflow fixes merged; Codex quota outage; 2 refusal wedges; 4 science rounds landed)- Science rounds landed on 4 issues. #928 folded the matched-length answer-span control (headline retitled: conditioning on realized CoT predicts the thinking model's answer state no better than a matched-length slice…daily__2026-07-07· updated 2026-09-03
- Daily — 2026-07-08 (Step C backlog cleared — 71 filed; 21 infra fixes merged; 3 science rounds; 0 promoted)- #1092 (realistic sparse-crossed corpus) ran its full pipeline end-to-end. Launches 5→9 on pod-1092 (8×H100) through crash-fix rounds 8.5–8.9 (vLLM H100 illegal-memory-access mitigation, bf16-null gate recalibration,…daily__2026-07-08· updated 2026-09-03
- Daily — 2026-07-09 (56-task workflow-fix wave executed; 33 more filed; watcher respawn-loop found)- Yesterday's 56-task workflow-fix wave executed end-to-end. The route-2 review tasks filed by the 07-08 /daily (#1148–#1214) ran as autonomous /issue --auto sessions through the full pipeline (plan → critic ensemble …daily__2026-07-09· updated 2026-09-03
- Daily — 2026-07-10 (40 infra fixes landed; /issue boot-killer found + hotfixed; 0 promotions)- #1092 landed its clean-result and parked at awaitingpromotion (20:25 UTC): "Answer-state transport on real conversations runs almost entirely through the query-bearing context state." Full pipeline in one day — 24 G…daily__2026-07-10· updated 2026-09-03
- Daily — 2026-07-11 (24 infra fixes merged; 11 new review filings)- 24-task infra drain completed end-to-end. #1251–#1274 (the 07-10 nightly's daily-fix: filings plus workflow-fix tasks) all ran through autonomous /issue --auto sessions and merged to main — ~680 commits on 07-11 UTC…daily__2026-07-11· updated 2026-09-03
- Daily — 2026-07-12 (11-task infra drain done; 12 new filings; one 12.5h refusal stall)- The 07-11 nightly's full 11-task filing batch drained end-to-end. #1275–#1285 all ran through autonomous /issue --auto sessions and merged to main (~297 commits on 07-12 UTC, PRs #1031–#1040). Landed: file-only Code…daily__2026-07-12· updated 2026-09-03
- Daily — 2026-07-13 (14-task infra drain merged clean; 1 applied + 9 filed)- The 07-12 nightly's filing batch drained end-to-end, again. 14 kind: infra tasks completed and merged to main today (PRs #1041–#1054): the 12 route-2 filings #1286–#1297 plus #1299/#1300, which were filed mid-drain …daily__2026-07-13· updated 2026-09-03
- Daily — 2026-07-14 (0 promotions; 4 fixes applied, 15 filed for review, 1 held)- Very heavy fleet day: 590 task-state commits across ~33 live sessions. ~10 workflow-fix / daily-fix infra tasks completed end-to-end and merged to main (#1301–#1309, #1311 pod-side sentinel read-back tolerance, #131…daily__2026-07-14· updated 2026-09-03
- Daily — 2026-07-15 (heaviest infra day on record — 35 fixes filed, ~15 shipped same-day; 0 promotions)- Workflow-fix pipeline ran at record throughput: ~15 kind: infra tasks completed same-day (#1311–#1313, #1316–#1319, #1322, #1326–#1327, #1340, #1348, #1355–#1356, #1360, #1362–#1365 …), every code-review round singl…daily__2026-07-15· updated 2026-09-03
- Daily — 2026-07-16 (1 result promoted; 5 fixes applied, 34 filed)- #1073 promoted useful (04:25Z): a single greedy answer is an adequate stand-in for 10-rollout-averaged answers — trims the #952/#1092 line's eval cost.daily__2026-07-16· updated 2026-09-03
- Daily — 2026-07-17 (0 promoted; 12 review filings; Codex ensemble outage day 1)- Heavy autonomous day: 3,278 task-state commits across ~45 tasks. Most active: #1434 (258 commits), #1345 (164), #1090 (154), #1335 (153), #1415 (109) — experiment pipelines with crash-fix rounds, review rounds, and …daily__2026-07-17· updated 2026-09-03
- Daily — 2026-07-18 (workflow-fix fleet day; 0 results promoted)- Workflow-fix fleet day: 33 kind: infra tasks completed (#1487–#1524 range), landing yesterday's 12 /daily route-2 filings (#1493–#1504) plus the watcher-dispatched siblings through the full planner → critic ensemble…daily__2026-07-18· updated 2026-09-03
- Daily — 2026-07-19 (28 infra merges; 3 experiments re-parked; 18 filed + 1 applied)- Workflow-fix wave landed: ~28 kind: infra tasks completed + squash-merged today (#1523–#1536, #1538–#1548, #1550–#1553 band — the 07-18 /daily's 21 filings plus in-flight parks), including the ruff full-ruleset pin …daily__2026-07-19· updated 2026-09-03
- Daily — 2026-07-20 (15 workflow-fix merges;- A fleet of ~15 autonomous workflow-fix/daily-fix sessions (#1554–#1573, filed by yesterday's /daily + the workflow-fix protocol) all ran their full pipelines and merged to main — 615 commits on main today. Highlight…daily__2026-07-20· updated 2026-09-03
- Daily — 2026-07-21 (12 overnight fixes merged;- All 12 of last night's daily-filed fix tasks (#1574–#1585) ran their autonomous /issue pipelines end-to-end and merged to main (PRs #1349–#1360): the GCP/SLURM log-excerpt structural digest (#1574), the expensive-ba…daily__2026-07-21· updated 2026-09-03
- Daily — 2026-07-22 (compute-starved day;- #779 landed its follow-up round and a user-authored sibling. The l26-kernel-gate-recovery same-issue round completed → awaitingpromotion (PR #1389 merged, methodology doc exported); the persona-vector pre-image sani…daily__2026-07-22· updated 2026-09-03
- Daily — 2026-07-23 (915 commits; 2 clean-results parked; ~26 wf-fixes landed; 21 filed tonight)- 915 commits landed on main. #1586 (four fiction-character leakage-lattice run) dominated with 176 commits: 4×H100 RunPod run after a ~5.5h GCP 8×A100 queue-starve, ≥7 overnight crash-fix rounds (disjointness assert,…daily__2026-07-23· updated 2026-09-03
- Daily — 2026-07-24 (26 infra fixes merged;- The 07-23 wf-fix wave landed end-to-end: ~26 kind: infra tasks (#1640–#1666) ran the full autonomous pipeline and ALL completed + merged same-day — among them the daily-pin-test registration (#1645), the bare-commit…daily__2026-07-24· updated 2026-09-03
- Daily — 2026-07-25 (19 workflow-fix tasks merged; ~117 problems swept; 23 filed)- 19 workflow-fix tasks merged to main in one day — #1667–#1683, #1687, #1688 — thedaily__2026-07-25· updated 2026-09-03
- Daily — 2026-07-26 (19 wf-fix tasks merged; main red 15.5h; 153 problems swept)- Last night's 23-task filing wave fully drained. All 19 route-2 workflow-fix tasksdaily__2026-07-26· updated 2026-09-03
- Daily — 2026-07-27 (22 wf-fix tasks merged-1; NEW main red ~17h; a lost urgent park; 90 problems swept)- The wf-fix wave kept draining: 22 tasks completed today (#1713–#1735), 21 merged.daily__2026-07-27· updated 2026-09-03
- Daily — 2026-07-28 (context→answer fleet running;- #1689 reached awaitingpromotion (surface-framing vs speaker-identity costs on the context→answer map) after a user-directed bootstrap descope (200→0 draws; the fit ladder had been mis-sized ~11×, ETA 38h→5-6h). #109…daily__2026-07-28· updated 2026-09-03
- Daily — 2026-07-29 (main un-redded; 6 applied, 29 filed)- Huge infra day: 62 tasks created (#1785–#1846, mostly the 07-28 nightly's filings) and ~30 daily-fix/workflow-fix pipelines ran end-to-end and merged to main via Step 10d (#1785–#1813 chain + #1829/#1833/#1834/#1838…daily__2026-07-29· updated 2026-09-03
- Daily — 2026-07-30 (8 follow-up rounds landed; 18 fixes filed; 0 promotions)- #1739 (behavior-prediction map horse race) ran end-to-end: all 3 GCP lanes terminal (1,906 cells), upload-verification PASS, interpretation + clean-result critics PASS, merged (7ac4098f45), parked awaitingpromotion.…daily__2026-07-30· updated 2026-09-03
- Daily — 2026-07-31 (31 tasks completed; 17 fixes filed for review; 0 promoted)- Heavy fleet day: 31 tasks completed (dominated by the workflow-fix wave #1851–#1872 plus #1887, #1889, #1903, #1911, #1912, #1937, #1941, #1953–#1956), 6 tasks reached awaitingpromotion (#1768, #1895, #1900, #1901, …daily__2026-07-31· updated 2026-09-03
- Daily — 2026-08-01 (34 tasks completed; 2 route-1 fixes; 25 filed for review; 0 promoted)- Heavy pipeline day: ~1,764 task-state commits. 28 tasks created — the 17 daily-fix filings from the prior nightly (#1962–#1978), 10 workflow-fix filings from in-flight sessions (#1980–#1989), and 1 new experiment (#…daily__2026-08-01· updated 2026-09-03
- Daily — 2026-08-02 (0 promoted; ~20 infra fixes merged; 5 applied + 18 filed by the sweep)- Huge autonomous infra day: 1,144 task-state commits; ~20 workflow-fix/daily-fix infra tasks completed AND merged to main in one day (#1938–#1944, #1948–#1950, #1955, #1957, #1958, #1960, #2017, #2020, #2021, #2023, …daily__2026-08-02· updated 2026-09-03
- Daily — 2026-08-03 (22 infra fixes landed, GCP backend removed, 0 results promoted)- 22 tasks reached epm:done — almost the whole 08-02 daily-fix batch landing at once (#1962–#1978 daily-fix, #1980/#1982 workflow-fix), plus #2032, #2050 (pgrep -f under-match), #2051 (pod bootstrap full-clones 10.5 G…daily__2026-08-03· updated 2026-09-03
- Daily — 2026-08-04 (autocompact-thrash storm day; root cause found + fixed in-window; recovered by the 08-05 nightly)- RECOVERY NOTE: the 08-04 nightly /daily died at 2026-08-05T07:03Z on repeated API 529 Overloaded errors ~36 min in (stub + miner brief only). This file was recovered via the SKILL.md Edit-in-place husk branch by the…daily__2026-08-04· updated 2026-09-03
- Daily — 2026-08-05 (heavy autonomous merge day; 08-04 nightly died to 529s — recovered tonight)- #1491 scale ladder completed compute: all 6 rungs (0.5B → 32B) captured + fit on pod-1491 (8×H200), every rung upload-verified (PASS markers 07:17Z–09:51Z), pod terminated after verified-done teardown; interpretatio…daily__2026-08-05· updated 2026-09-03
- Weekly — 2026-W22 (2 results promoted;- 2 clean-results promoted to completed: #390 (refusal-style negatives install a persona gate that generalises across 9 of 11 OOD eval framings, MODERATE) and #391 (could not selectively implant sycophancy on Qwen2.5-…weekly__2026-W22· updated 2026-09-03
Ideas (6)
- Subliminal learning: does it need multiple epochs? (design draft, 2026-10-26)Status: draft. Nothing launched. Not yet a task.idea__2026-10-26-subliminal-epoch-test-design· updated 2026-10-26
- Analyses of the context→answer linear map W — methods survey (2026-07-06)Chat-session lit survey (4 parallel agents: dynamical systems/control, Markov/transfer-operator/idea__2026-07-06-context-answer-map-analyses· updated 2026-09-03
- Ideation braindump — 2026-06-05Raw "next steps for this week" braindump from Thomas, stored verbatim, thenidea__2026-06-05· updated 2026-09-03
- Theoretical analysis plan — the four context→answer maps at the large-n grain2026-07-28. Scope: the fitted maps from the #779 (up to 1M rows) and #1092 (21,193-row crossed real-conversation corpus) lines. Grounding: the methods survey (docs/ideas/2026-07-06-context-answer-map-analyses.md), the…idea__2026-07-28-four-arm-map-theoretical-analysis-plan· updated 2026-09-03
- Why do more contrastive negatives cause MORE source implantation? (#472)Literature + internal-evidence synthesis, 2026-06-11. Companion to task #472idea__2026-06-11-why-negatives-amplify-implantation· updated 2026-09-03
- Context-mapped forecasting of reward hacking and misaligned actionsTest whether an agent's activation at a decision boundary, before it generates any new reasoning or action, predicts the probability of a subsequent reward hack or misaligned action. Test whether the project's already…idea__context_mapped_agent_risk_experiment· updated 2026-08-31
Reference (41)
- Living-docs ⇄ /issue workflow integration (plan)proposalHow experiment completion keeps open_questions.md, papers.md, and claims.yaml consistently updated — hook points, components, consistency guarantees, phased rollout.living-docs-workflow-integration-plan· updated 2026-09-03
- Research Ideas2026-09-17 — Persona geometry update: [Task 2673](https://eps.superkaiba.com/tasks/2673) measured the Story Imprinting persona panel directly on Qwen3.8-27B: 10 personas × 240 questions, all 64 blocks. Sarcasm/list pr…research_ideas· updated 2026-09-18
- #1776 smoke log (accumulates across rounds A-D; round D reads this for the report)Env prefix on every command: OMPNUMTHREADS=8 MKLNUMTHREADS=8 OPENBLASNUMTHREADS=8 NUMEXPRNUMTHREADS=8 MALLOCARENAMAX=2; all run from the worktree root, timeout-bound; smoke outputs diverted to /tmp/issue-1776-smoke/ (…issue_1776_smoke_log· updated 2026-09-03
- #2094 patching dashboard — slot × layer effectiveness, coherence, and qualitative examplesUser-chat inline round (2026-08-07, monitoring session). All values are replace-dose (full-state patch), steered arm, coherent draws only (one greedy draw per pair per cell); cell text = ncoherent/ntotal pairs, = und…issue2094_patching_dashboard· updated 2026-09-03
- Agent-memory index curation — task #2093Bodies deleted: 0 across all three agents (curation is index-row-levelagent_memory_curation_2093· updated 2026-09-03
- Anthropic API throughput guidelinesPhase 3 of docs/apithroughputplan.md. Decision table for completingapi_throughput_guidelines· updated 2026-09-03
- Are conditional LoRA edits represented linearly in the residual stream? — verified lit synthesisCompiled 2026-06-09 from a deep-research sweep (21 primary sources fetched, 104 claims extracted,lora-edit-linearity-lit-review· updated 2026-09-03
- Behavior-Generalization Testbed (v1 design)Status: design draft, 2026-06-09 (rev 3: Group C delta-rule/associative-memory predictors (P1–P6) derived from Dan's framing + mechanistic asymmetry; content-stripping dropped (content predictors first-class); all six…behavior_generalization_testbed· updated 2026-09-03
- Claim-4 controls — numbers note (task #1739, claim4-controls round)Numbers + coverage only; claims stay with the writeup author.map_behavior_prediction_claim4_controls_note· updated 2026-09-03
- Claims registryAuto-generated from docs/claims.yaml by scripts/renderclaims.py.claims· updated 2026-09-03
- Comprehensive plan: testing the leakage-predictor assumptionsSource of truth: the Overleaf theory paper "Predicting fine-tuning–inducedtheory_assumption_test_plan· updated 2026-09-03
- Conditional Behavior / Persona-Space Interventions — Related WorkCompiled 2026-05-28 from an 8-cluster deep literature sweep (personas/steering, function/task/in-context vectors, emergent misalignment, sleeper agents/backdoors, inoculation/removal, SDF/character-transfer, data attr…conditional-behavior-related-work· updated 2026-09-03
- Context bank — prefixes and queries tried in the context-vector patching lineTwo banks feed this line: the core crossed bank of [#2094](https://eps.superkaiba.com/tasks/2094) (all headline patching results) and the 21-type extension sweep of [#2162](https://eps.superkaiba.com/tasks/2162) (whic…issue2094_context_bank· updated 2026-09-03
- Context-Generalization Testbed (v1 design)Status: design draft, 2026-06-09 (rev 2: larger cross-family negative panel, EM row contrastive, behavior-instruction "tricky" contexts, WildChat length bins). Not yet planned or run.context_generalization_testbed· updated 2026-09-03
- Emergent-Misalignment Narrow-Finetuning Dataset CatalogCatalog of narrow-behavior SFT datasets from the emergent-misalignment (EM) literature,em_dataset_catalog· updated 2026-09-03
- Experiment suite — testing the leakage-theory assumptionsStatus: design doc (review before any task is created or run).leakage_theory_test_suite· updated 2026-09-03
- External Repo ReferenceThree reference codebases in external/. Read their configs when you need training recipes.external_repos· updated 2026-09-03
- GCP quota requests — paste-ready justifications (2026-06-12)File from the console (IAM & Admin → Quotas, project eps-persona-gpu-jun2026) withgcp_quota_requests· updated 2026-09-03
- Glossary — prefix / query / context / answer (the context->answer mapping line)Canonical vocabulary for the context->answer-map experiments (#779, #810/#658 line, and successors).glossary_context_answer_map· updated 2026-09-03
- ICL vs fine-tuning on the same examples — verified lit synthesisCompiled 2026-06-10 from a deep-research sweep (5 search angles, 21 primary sources fetched,icl-vs-finetuning-lit-review· updated 2026-09-03
- Implanting a Targeted Behavior into an LLM — A Method-Family Literature ReviewThe project models a trained system as (W, C) -> B: weights W, context C (everything before the assistant turn — system prompt + Q/A history), and a behavior B that is a region of the policy piW(.|C), not a single out…behavior-implantation-lit-review· updated 2026-09-03
- Interim results — Behavior prediction through the context→answer map (#1739)- Persona vectors predicts behavior from the context vector by projecting the persona vector on the context vector and shows that it can pretty reliably predict whether the answer will exhibit a certain behaviormap_behavior_prediction_interim_results· updated 2026-09-03
- Issue #1482 — SAE-feature and context analysisWorking writeup section. Plots and dashboards are filled in; takeaways are leftissue_1482_writeup_section· updated 2026-09-03
- Issue #226 verification anchorsThis file records the verification-only items from the issue #226 planissue-226-verification-anchors· updated 2026-09-03
- Leakage paper ↔ theory-assumption mapTwo papers:leakage_paper_assumption_map· updated 2026-09-03
- Literature notesReading list with one-paragraph summaries for papers relevant to this project. Citation keys map to overleafpaper/NeurIPS/references.bib. Raw PDFs / arxiv MCP notes live in .arxiv-papers/.papers· updated 2026-09-03
- map_behavior_prediction_results_writeup<!-- Thomas's writeup, verbatim except: 'probe' / 'ridge regression' / 'direct -->map_behavior_prediction_results_writeup· updated 2026-09-03
- Mapping-transfer tiersA typology for cross-cell transfer of fitted context→answer mapsmapping_transfer_tiers· updated 2026-09-03
- Marker training recipe — deep dive (2026-06-08)Synthesis of ~30 marker-implantation experiments on the over- vs under-trainingmarker_training_recipe· updated 2026-09-03
- Metrics for mapping similarityWe refer to several different kinds of mapping similarity (between base and instruct model, from assistant to other characters in stories, across post-training stages).mapping_similarity_metrics· updated 2026-09-03
- Open Questions — Explore Persona SpaceSynthesized 2026-05-27 from: RESULTS.md, all tasks/awaitingpromotion/ + tasks/completed/ clean-results (hascleanresult=true), docs/mentorupdates/{2026-05-11,2026-05-22,2026-05-26,2026-05-27-research-log-channel-snapsh…open_questions.proposed· updated 2026-09-03
- Open-questions evidence back-fill report (proposal)Generated 2026-05-28. Audit surface for the proposed docs/openquestions.proposed.md.open_questions-backfill-report· updated 2026-09-03
- Overview & Open Questions<!-- living-docs-changelog:begin -->open_questions· updated 2026-09-03
- Plan: Anthropic API max-throughput characterization + unified dispatcherStatus: proposed (kind: infra)api_throughput_plan· updated 2026-09-03
- Predicting behavior expression through the context→answer mapStatus: draft plan for review (not yet filed as a task). Drafted 2026-07-27, revisedmap_behavior_prediction_plan· updated 2026-09-03
- Project notesPrivate working notes — things to try, reminders, not polished claims. Not shown in the public narrative.project_notes· updated 2026-09-03
- Result: Fine-tuning a behavior into a context rewrites the context→answer map — in proportion to its weights-carried dose — while the context vector barely moves(Chat-drafted writeup of task #1768, 2026-07-30. All numbers from the #1768 clean-result body; artifact record: https://eps.superkaiba.com/tasks/1768)issue_1768_finetuning_write_writeup· updated 2026-09-03
- Size-dependent generalization of finetuned behaviors — literature synthesisEmergent misalignment (EM) — narrow harmful finetuning generalizing to broad misalignment — is a robust phenomenon that occurs across model families and sizes (0.5B-671B), and the dominant evidence says it gets STRONG…size-dependent-generalization-lit-review· updated 2026-09-03
- TODO<!-- Project todos and next steps -->TODO· updated 2026-09-03
- Token-to-token activation continuity in LLMs, and how it changes with depthLiterature dive — 2026-06-29.token-to-token-activation-continuity-lit-review· updated 2026-09-03
- Writing theoretical-framework documents: a field guideA reference for the kind of document where you propose a model, state itstheory_writing_guide· updated 2026-09-03