cs.AIOct 19, 2026
SaveFrom Trace to Fix: Residual Task State in Long-Horizon Agents
Organizations: University of Central Florida
Abstract
Recovering from failures in long-horizon, stateful language agents is challenging because the action introducing a defect, the symptom exposing it, and the persistent state requiring repair frequently diverge. Existing methods largely focus on retrospective failure attribution, leaving sequential diagnosis and state repair poorly integrated under constrained budgets. We introduce TRACE-FIX, a unified recovery framework centered on versioned residual task state. Trace transforms execution logs into a structured dependency-linked ledger, prioritizing unresolved task requirements rather than salient surface errors. Fix formulates recovery as budgeted planning over admissible actions, incorporating outcome forecasting and explicit repair-reserve constraints to prevent premature budget exhaustion. To rigorously decouple localization accuracy from downstream repair capability, we present LongH-Debug, evaluating root-cause attribution, verified-fault recovery, and end-to-end closed-loop execution independently. Our empirical analysis demonstrates that structured residual tracking substantially improves causal localization over raw-trace baselines, while budget-aware sequential planning converts verified fault signals into more reliable task recovery without destructive trial-and-error. Code is available at https://github.com/toffeecat/tracedp.
Figures & tables
| A Delayed state mutation db-wal-recovery live run, 97 tool calls 0 ls -la /app: base 8192 B log 16512 B 2 sqlite3 main.db … PRAGMA wal_checkpoint: log deleted 9 cp main.db-wal /tmp/…bak (already too late) 38 the same read deletes the rebuilt log again 95 the actor states the mechanism itself end timeout, reward 0.0 A successful read removes the only carrier of the six missing records. Nothing errors, so the run repeatedly rebuilds and destroys the log. | B Residual obligation openssl-selfsigned-cert live run, 7 checks 1 script-runs: /app/check_cert.py exits 1 2 verification-file: subject and fingerprint 3 key-attributes: 2048-bit RSA, mode 600 4 script-runs re-run: the same predicate passes 5 combined-pem parses as a certificate 6 ssl-directory-layout holds the deliverables 7 certificate-attributes: subject, 365-day validity end reward 1.0 The actor rewrites the script after the first failure; the same predicate then passes and the task reaches reward . |
| C Blame is not a current fault bn-fit-modify live run, 8 checks 0 learned_dag.csv exists in to,from format 1 intervened DAG exists in to,from format 2 intervened sample keeps the column names 3 the DAG has six total edges 4 intervened edges are a subset of the learned DAG 5–7 repeated checks of the same conditions end 8/8 supported, reward 1.0, 0 repairs The history contains a failure, but the live state already satisfies every check. Historical blame is therefore not a current repair target. | D Budgeted recovery cancel-async-tasks live run, 8 checks, 2 repairs 1–2 import-form and signature supported 3 concurrent-execution refuted 4–6 max-concurrency, cancel-below, cancel-at supported 7 cancel-above-limit refuted 8 import-form re-run: supported end 2 repairs tried, reward 0.0 Eight checks and two repairs exhaust the twelve-unit budget before either repaired predicate is re-verified. |
Figure 1: Four long-horizon failure modes from retained Terminal-Bench trajectories. (A) A successful operation silently destroys state whose loss appears only later. (B) A failed predicate is subsequently repaired, leaving only its residual consequences relevant. (C) A historical failure remains in the trajectory although the current state is already healthy. (D) Checks and repairs share a finite budget, which can be exhausted before repair verification. Rows quote retained commands or registered checks; and denote supported and refuted outcomes.
Figure 2: Runtime path from public history to recovery. The task contract, public execution record, resource versions, and registered checks are compiled into a residual state. TRACE uses the resulting dependency-linked ledger to rank historical events relevant to unresolved requirements. FIX receives the live action slate and a bounded public-state forecast, while runtime code checks action identity, cost, reachability, and time before computing the depth-two decision. The selected action is then executed; its receipt updates resource versions and evidence, the residual state is recompiled, and planning repeats before the same actor continues. Hypothetical planning outcomes are kept separate from observed execution receipts throughout.
Figure 3: LongH-Debug benchmark: scale, state diversity, and executable recovery coverage. (a) Published failure-trajectory counts for prior agent-debugging benchmarks, with the LongH-Debug row shown separately. (b) The planned 1,500-trajectory corpus allocates quotas across seven environments grouped into search/retrieval, symbolic/embodied, transactional/social, and persistent artifact/system state. (c) Protocol coverage distinguishes retrospective attribution from executable recovery. Filled circles denote explicitly supported capabilities, half-filled circles partial or alternative protocols, dots unreported capabilities, and crosses explicit exclusions. LongH-Debug’s replay and labels apply to its qualified subsets rather than to every trajectory in the planned corpus.
| Method | Hit@1 | Hit@3 | MRR | Cand. recall | VerifiedRoot @ | Tok. (M) | Rel. cost |
|---|---|---|---|---|---|---|---|
| Generic one-shot | 42.0 | 74.5 | .571 | 93.0 | 31.0 | 2.40 | 1.00 |
| Dependency-only | 24.0 | 46.0 | .351 | 64.0 | 15.5 | 0.52 | 0.22 |
| Who&When (adapted) | 8.0 | 12.0 | .097 | 17.0 | 4.5 | 6.00 | 2.50 |
| TraceElephant (adapted) | 44.0 | 61.0 | .508 | 71.0 | 33.0 | 12.20 | 5.08 |
| Residual Context | 43.0 | 74.5 | .573 | 93.0 | 32.0 | 7.80 | 3.25 |
| Our Trace | 68.0 | 89.0 | .782 | 97.0 | 60.5 | 3.10 | 1.29 |
Table 1: TRACE localization on paired natural failures. Candidate recall asks whether the adjudicated event is present before ranking; VerifiedRoot @ requires an independent qualified check sequence within the verification budget. Token counts are total model input tokens over the paired population and relative cost is normalized to Generic one-shot.
| Fixer | TaskRecovery | LocalResolve | Harmful | Cost/success |
|---|---|---|---|---|
| Same-Actor Continue | 35.0 | 29.2 | 0.0 | 1.00 |
| Fresh Fixer | 42.5 | 40.8 | 5.8 | 0.92 |
| Fixed-order sweep | 47.5 | 50.0 | 4.2 | 0.86 |
| Information gain | 50.0 | 53.3 | 4.2 | 0.82 |
| Information gain + repair reserve | 61.7 | 66.7 | 3.3 | 0.69 |
| Residual-Greedy | 56.7 | 61.7 | 3.3 | 0.73 |
Table 2: Recovery under matched verified-fault input ( ). TaskRecovery is the benchmark-owned outcome after same-actor continuation; LocalResolve asks whether the targeted live incident is gone. Harmful counts unnecessary or state-damaging interventions. Cost/success is normalized to Same-Actor Continue.
Figure 4: Why TRACE should help. (a) Representation ablation. Only the endpoints, Raw trajectory ranker and Full TRACE, are measured (Table 1 ); the three intermediate rows are planned targets with no measured values yet. (b) TRACE/Generic Hit@1 by supported dependency depth and feedback delay. Each cell is a planned target with no measured value yet. Circles mark Hit@1 and squares mark VerifiedRoot@B.
Figure 5: FIX recovery, cost, and mechanism. (a) Recovery–cost frontier for the methods in Table 2 , plotting TaskRecovery against cost per success; lower cost is better. (b) Mechanism ablation under matched information and legal actions. The strongest planning control is information gain with a repair reserve, not a budget-oblivious strawman.
Figure 6: From diagnosis to task recovery. (a) The TRACE FIX crossing reports benchmark-owned TaskRecovery . Replacing only TRACE or only FIX isolates each component; the diagonal gives the end-to-end system. (b) Per-environment TaskRecovery for and on the 60 closed-loop tasks, which span three environments (26, 18 and 16 tasks).
| Controller | TaskRecovery | Harmful | Cost/succ. |
|---|---|---|---|
| Full FIX | 67.5 | 2.5 | 0.61 |
| No structured residual state | 52.5 | 5.0 | 0.78 |
| No version invalidation | 58.3 | 5.8 | 0.74 |
| No receipt-triggered recompilation | 49.2 | 6.7 | 0.82 |
| Residual-Greedy | 56.7 | 3.3 | 0.73 |
| Information gain + repair reserve | 61.7 | 3.3 | 0.69 |
Table 3: Controller ablations under matched verified-fault input. The information-gain+reserve row is also a strong alternative policy, not merely an ablation.
More in this topic
LLM Agent Reliability
Large language model (LLM) agent reliability concerns dependable, safe, and verifiable behavior across tools, multi-step tasks, and changing environments.
See all 801 papers