Long-running work accumulates material: the original goal, later constraints, attempted routes, failed experiments and current progress. Sending everything again on each continuation repeats more and more content; keeping only a summary can make the original basis of a decision difficult to recover.
Long-running work accumulates material: the original goal, later constraints, attempted routes, failed experiments and current progress. Sending everything again on each continuation repeats more and more content; keeping only a summary can make the original basis of a decision difficult to recover.
LoopX's context delivery separates these jobs: preserve complete sources, read the facts needed for the current decision, expand supporting material around questions, and reuse unchanged structures exactly within a valid compaction cycle.
This article describes the native Codex path in released LoopX 2.2.0. Supported POSIX Codex App/SSH sessions and new benchmark heartbeat trials default to context delivery; saved off remains effective and explicit off is available. When enabled, complete content is saved first and the Agent receives a reading entrypoint for current facts. See the 2.2.0 release notes for publication and validation scope. The synthetic measurements below retain their original source and experimental conditions.
Think of a capture as a complete book. An index locates its chapters; a JSON Pointer addresses a place inside it. Progressive reading determines how deeply to open the material now. Differential delivery determines what can be reused on the next continuation.
@@ -174,7 +174,7 @@
On the next continuation, deliver what changed
When enabled, the default output is a reading entrypoint
Connect these operations back to the currently supported native Codex quota path:
Recover missing context while preserving the original work
What the measurements establish
The example separates three mechanisms: avoid repeatedly moving history not currently needed, combine reads that belong together, and avoid retransmitting unchanged verified structures. Complete sources, requirements and recovery connect them.
Reader figures include actual CLI page metadata; local path length affects a small amount of framing. Projection figures compare capture bodies; the cycle table compares codec output. The reproduction script checks these distinct measurements and exact reconstruction.
-
Real long-running tasks must also count necessary later reads, recovery, cache hits, model input/output, tool calls, waiting and human attention. Effectiveness depends on constraints understood, decisions made and accepted outcomes. This synthetic example measures the mechanical path; real model benefit still needs comparable-task evidence.
+
Real long-running tasks must also count necessary later reads, recovery, cache hits, model input/output, tool calls, waiting and human attention. Effectiveness depends on constraints understood, decisions made and accepted outcomes. The synthetic voice example above measures the mechanical path; below we examine multi-task and layered ablation observations on 18-hour trajectories.
+
Multi-task and layered ablation: benefit chain and boundary counterexamples
+
In EdgeBench ANN vector search optimization (gpt-6.1-sol / xhigh, 18h, subject to Recall@10 ≥ 0.95) and cross-task ablations, we separate Context Delivery (progressive selection and cycle deltas) from Stride (evidence-conditioned experiment sizing), auditing selected control-packet bytes, main-session model input tokens, and local/early-window counterexamples side by side (reconciled from archived trajectories and receipts by @shangzh0):
+
Multi-task ablation and layered accounting
Top (A): ANN 18h six-arm peak valid throughput QPS (Recall@10 ≥ 0.95). Bottom (B): Layered control-packet byte and main-session input token savings vs Context off (left, negative) alongside disclosed local/early-window overhead and cross-task counterexamples (right, positive). Open image for full size.
+
+
+
+
Dimension or experimental arm
+
Observed figure
+
Scope and interpretation
+
+
+
+
+
ANN 18h: Official / Single / Native-Goal
+
5,753 / 5,706 / 4,372 QPS
+
Official peaked at 1.05h and remained on a plateau afterward
+
+
+
ANN 18h: Explore · Context off
+
5,766 QPS (+0.2%)
+
Peaked at 11.37h; repeated full-context delivery consumed main-session budget
+
+
+
ANN 18h: Explore · Context on
+
6,020 QPS (+4.6%)
+
Peaked at 14.43h; progressive reads and cycle deltas preserved late-stage headroom
+
+
+
ANN 18h: Explore · Context on + Stride
+
7,159 QPS (+24.4%)
+
Breakthrough at 9.19h (QT_8bit_direct HNSW-SQ8, Recall@10 = 0.9585)
Initial baseline establishment and staged discovery add upfront overhead that only amortizes over longer horizons
+
+
+
Counterexample 3: Historical Grid (lower is better)
+
Official 1967 < Off 2054 < On+Stride 2088 < On 2242
+
Cross-task counterexample under differing durations and resources; retained without causal overclaim
+
+
+
+
This dual-track audit shows why control-packet byte reduction (−74.25% / −38.54%) must not be equated with whole-episode model token savings (−12.07% uncached, −7.46% total), nor used to hide early-window indexing overhead (4.05× in the first 2 hours) or cross-task counterexamples. Context Delivery’s practical value lies in slowing context bloat across 10–18 hour runs so that, together with Stride, preserved attention converts into effective late-stage experiments (see EdgeBench late-stage breakthroughs and five-arm attribution).
2.2.0 ships complete sources, file entrypoints, batch selection, index depth, applicable acceptance layering, exact sharing, cycle deltas and recovery. Broader decision-stage candidate, preference and Explore delivery and consistency across all entrypoints remain unfinished. Disable in Settings → This machine → Capabilities → Differential context → Codex; saved off survives upgrade. A command or new benchmark trial can use --context-delivery off. Trials do not inherit the machine switch, and existing frozen trials are not rewritten. Under the default policy, ordinary CLI, unsupported hosts and non-heartbeat benchmarks retain their full paths; an explicit unsupported on request is rejected. Unavailable hooks use full delivery. Saving configuration, actual installation and model adoption remain separate checks.
Read the same material yourself
example.py generates this complete synthetic voice task and invokes the existing reader, progressive projection and delta owners. It checks selected values, directory addresses, all ten requirements, shared references and complete recovery. It registers no Goal, admits or settles no Turn, and calls no model or speech service.
A research run can start ahead, then barely improve for ten hours. It can keep changing code and lose its best result. These failures made us revisit goals, feedback, experiment size and recovery.
The work began with trace diagnosis in early October: repair evaluation and continuation, follow evidence into the next experiment, then charge the control plane for its own cost. This account follows that progression, including results that challenged the approach.
This article asks how mechanisms changed real research and where they did not help. For what actually reaches the next decision, begin with progressive and differential context delivery, then return to the outcomes and complete costs below.
+
Ruiteng Huang · October 29, 2026 · Updated November 3
What does an agent achieve with ten more hours?
How EdgeBench reshaped the LoopX control plane
A research run can start ahead, then barely improve for ten hours. It can keep changing code and lose its best result. These failures made us revisit goals, feedback, experiment size and recovery.
The work began with trace diagnosis in early October: repair evaluation and continuation, follow evidence into the next experiment, then charge the control plane for its own cost. This account follows that progression, including results that challenged the approach.
This article asks how mechanisms changed real research and where they did not help. For what actually reaches the next decision, begin with progressive and differential context delivery, then return to the outcomes and complete costs below.
Here, kernel means LoopX’s shared goal, admission, transaction and recovery rules. LoopX keeps the original goal, acceptance conditions, work ownership and evidence durable across sessions and hosts. Its control plane constrains state transitions and supplies feedback for the model’s next choice. It does not certify that the choice is insightful. A recorded finding, a read receipt and a replan entry are intermediate evidence; the relevant chain ends in actual work and a validated outcome.
The TypeScript refactor article explains ownership and model-assisted semantic repair. Below, we follow how these rules affect actual research work.
GoalAcceptance and scope
EvidenceResults and limitations
DecisionContinue or change course
WorkExecute and validate
RetentionKeep the best valid result
We separate three observations: an initial advantage, an improvement obtained later, and retention of the best validated artifact. A higher final score can come almost entirely from a stronger starting point. A better intermediate candidate can disappear from the terminal workspace. More runtime, Todos or graph nodes establish none of these outcomes.
-
1. The research progression
Period
Focus
Change
Implementation
Early October
Feedback and starting conditions
Repair blind-run prompts, ledger and recovery defects; establish original-task, scoring and continuation boundaries.
Cases are selected for mechanism changes and intelligible failures. PRs anchor phases rather than form an exhaustive daily release log. Dates describe research phases; linked records retain creation, merge and historical qualification details.
2. First make the feedback trustworthy
+
1. The research progression
Period
Focus
Change
Implementation
Early October
Feedback and starting conditions
Repair blind-run prompts, ledger and recovery defects; establish original-task, scoring and continuation boundaries.
Late-stage plateau breakthrough and five-arm attribution
Break the four-arm 91.94 plateau on Astra max 18h (+9 repaired declarations, 0 regressions); separate progressive deltas, Stride sizing and micro-tuning waste across ANN 18h and five-arm studies.
Cases are selected for mechanism changes and intelligible failures. PRs anchor phases rather than form an exhaustive daily release log. Dates describe research phases; linked records retain creation, merge and historical qualification details.
2. First make the feedback trustworthy
The first Portfolio experiments exposed measurement and execution defects before they could tell us much about strategy. A blind wrapper had removed iterative optimization guidance along with the judge entrypoint. A ledger check attributed earlier market returns to holdings bought only at the closing price. Both confounded the intended comparison; repairing them required new attempts while preserving the old results.
One small oracle made the accounting error clear. Start with 100,000 cash, buy 100 units at a closing price of 200, and pay 10 commission. Immediate equity is 99,990 and NAV is 0.9999; the new holding did not earn the preceding rise from 100. Independent reconciliation then matched all 243 days in each of two frozen samples, with equity residual below 4×10⁻⁹ and no share-count difference. This established ledger consistency, not investment skill. The versioned repair is documented in EdgeBench #14.
Continuation had a related failure: an outer process could wake repeatedly after a work item ended without recovering a useful next action. The governing question must return to the original goal. Completing a Todo can justify another experiment, waiting, or completion; it cannot by itself settle goal acceptance. Correct recovery and feedback make subsequent research interpretable.
@@ -47,9 +47,14 @@
When should a long, unfinished task review its direction
Historical cohort
Observed result
Interpretation
Sol/xhigh, old/new LoopX, Ad Placement · 14h
49,589,589,573 → 49,615,495,839
Newer run trailed at 2h and led at 14h; its terminal workspace matched its retained best
Astra/xhigh, Official Grid · 6h → 12h
1862 → 1810, lower is better
A stronger foundation baseline still produced useful late improvements
+
Astra max, five-arm study, New Foundations Consistency · 18h
91.9412 → 92.3419 (2065 → 2074/2246)
Four arms plateaued at ≤91.9412; Explore climbed step-wise across 9.65h–17.57h (+9 repaired declarations, 0 regressions)
Ad Placement provides a concrete instance of later progress rather than a universal plateau. A worse subsequent candidate was rejected and the best version survived. Its raw-score lead is modest—about 0.052%—and only that arm has the cited terminal-workspace verification. The old/new packages changed core and runner together, so this is not an isolated LoopX-feature effect.
The Astra result expands the foundation-model frontier in our observations. Official separated speed changes from search changes and obtained successive native improvements. That progress belongs to the foundation/Official path; it cannot be credited to LoopX. Stronger models also raise the standard for a useful harness: context and review costs must earn their place against an improving baseline.
+
New Foundations Consistency (Astra max, 18h): breaking a four-arm late-stage plateau
+
In New Foundations Consistency (Astra max, 18-hour five-arm archive by @shangzh0), the trajectories diverge sharply after hour 6: the official baseline reached 91.9412 (2065/2246) at 5.72h and produced zero gain across 11 subsequent submissions over the next 11.5 hours; heartbeat-resume also plateaued at 91.9412, while native-goal and single stopped at 91.8077 (2062) and 91.6296 (2058).
+New Foundations Consistency (Astra max, 18h) best-so-far trajectories across five arms, with an embedded 6h–18h (91.5–92.5 score) zoomed inset. heartbeat-explore made four step-wise jumps between 9.65h and 17.57h to reach 92.3419 (2074/2246, +9 repaired declarations vs Official with 0 regressions). Open image for full size.
+
As shown in the main trajectory and the lower-right zoomed inset, only heartbeat-explore—combining the evidence ledger with periodic direction reviews—broke out of the plateau after 9.65h (91.99 → 92.03 → 92.07 → 92.25) and jumped at 17.57h to 92.3419 (2074/2246), yielding a net gain of +9 repaired declarations with zero regressions over Official. This is single-run descriptive five-arm evidence showing that when late-stage reviews drive targeted verification of remaining declarations, hours 6–18 can still deliver verifiable gains; it does not guarantee breakthroughs on every task—which is why the Schema counterexample immediately follows.
Schema: an initial lead is not sustained improvement
Schema is the detailed counterexample. In the Sol old/new 14-hour window, the newer run scored 0.804620 versus 0.723684. It already led at the first native result, 0.794620 versus 0.713158, received at different elapsed times. Subsequent increments were only 0.010000 and 0.010526. The larger endpoint advantage mostly reflected the initial result; continuing code changes did not close the six original target requirements.
A separate Astra/xhigh 12-hour comparison was less favorable: Official reached 0.881813 and LoopX 0.786257, with neither improving its best between 6h and 12h. LoopX recorded 36 completed compactions versus four, and about 5.32 times the uncached input. Compaction can overlap useful work, and versions, recovery, shared load and scorer conditions differ. The trace still identifies the practical challenge: producing more changes and more control activity did not produce more accepted value.
@@ -111,6 +116,12 @@
Measure structural input separately
The context study recorded on October 28 starts Grid, Schema and Ad Placement from their original tasks with Sol/max and an 18-hour upper limit per case. Historical Official or older LoopX runs provide descriptive references where usable. Their model/effort, source, scorer, opportunity and clock differences remain visible; they are not strict matched controls. A new baseline is warranted where the old material cannot answer the relevant question. Attempts with different source versions, continuity bases or transports are analyzed separately, rather than pooled as one treatment.
First verify that actual delivery changes: the initial full read, a subsequent delta with its continuity basis stated, complete fallback after known baseline loss, and a full review-to-work interaction. Then compare useful progress, retained artifacts and the complete usage vector. Requested flags, receipts and smaller local packets cannot substitute for that evidence. If the feature fails its intended behavior, stop the affected attempt, preserve its costs and history, repair it and use a new attempt identity.
The first Grid and Ad interactions on the new native Hook path have completed full paging, one delta and subsequent task work. Within each same capture, the delta payload was 19.53% and 19.48% smaller than its corresponding full packet. These are single-read payload comparisons: they exclude page envelopes, subsequent model input, hooks, recovery and review costs. They do not establish lower total cost. A complete review-to-native-outcome cycle, sustained benefit and failure recovery remain to be verified.
Our effectiveness target is useful progress after many hours and reliable retention of it. Our efficiency target is comparable accepted value with less complete cost and human attention. The next result must address both. Current runs retain the original deadline, and failed or replaced attempts retain their costs and history.
-
This remains the October 28 research checkpoint. The 2.2.0 release and default-on policy do not automatically pass its unresolved checks. The mechanism article’s synthetic byte measurements, release-path installation/recovery checks, and this section’s real model trajectories answer questions about mechanical delivery, product usability and research effectiveness respectively. Their versions, windows and metrics differ; they are not a combined savings result.
+
This remains the October 28 research checkpoint. The 2.2.0 release and default-on policy do not automatically pass its unresolved checks. The mechanism article’s synthetic byte measurements, release-path installation/recovery checks, and this section’s real model trajectories answer questions about mechanical delivery, product usability and research effectiveness respectively. Their versions, windows and metrics differ; they are not a combined savings result.
+
Early-November update: ANN 18h layered ablation and cross-task five-arm trajectory attribution
+
Subsequent early-November studies added two pieces of empirical evidence:
+
+
Context Delivery × Stride multi-task and layered ablation (ANN 18h, gpt-6.1-sol / xhigh, Recall@10 ≥ 0.95): Across 18-hour vector-search runs, Explore · Context on + Stride reached 7,159 QPS (+24.4% over Official at 5,753 QPS, and ahead of Explore · Context on at 6,020 QPS and Context off at 5,766 QPS). Paired control-packet selected bytes fell 38.54% (3.09 MB → 1.90 MB, with 24 delta events reduced by 74.25%), while full-18h main-session uncached input tokens fell 12.07% (8.57M → 7.53M) and total input tokens fell 7.46% (272.03M → 251.74M). Alongside these gains, we retain three explicit boundary counterexamples: project+read process stdout was 1.57×–1.83× raw selected file bytes, first-2h early-window uncached input tokens rose 4.05× (0.25M → 1.00M) during initial baseline establishment, and the cross-task Historical Grid counterexample ordered Official 1967 < Off 2054 < On+Stride 2088 < On 2242. Full charts and dual-track accounting appear in the context-delivery ablation section.
+
Cross-task five-arm trajectory attribution (Portfolio, Grid, Graph, Ad Placement): In these reproduced five-arm attribution report (#7960) and cross-task inflection & replan overhead study (#7959), the five-arm taxonomy labels 86 classified replan events, while the three managed-arm totals are 44 + 36 + 24 = 104; valid-jump counts likewise differ (27 taxonomy vs 5 + 11 + 15 = 31). We therefore present 5/44 = 11.4%, 11/36 = 30.6% and 15/24 = 62.5% as arm-local ratios, without claiming one reconciled common denominator or cross-arm causal total. These are descriptive, arm-local observations rather than a matched repeated C2 comparison and do not justify a production default change. The reports also describe native-feedback retention on Portfolio (Official Native 53.87, Explore 31.19, Stride 35.82) and an observed Graph increase from 0.7194 at 2.1h to 0.7240 at 9.6h, followed by 7 post-peak attempts consuming 159.87M input tokens. The report’s fixed-classification formula evaluates to 0.74329, not its recorded 0.73903; it is a conditional estimate for that fixed component vector, not an upper bound on every linear model or proof that later work was necessarily wasted. The same reports record protocol tax from 23.9% to 12.4%–14.1% descriptively.
