Skip to main content

Goal Timeline

What the Goal Timeline shows​

Every goal produces a detailed timeline — every action the agent took, what it passed in, what came back, and how long it took. It’s kept with the run and opened from the Goal Progress panel.

The timeline answers: "What exactly did the agent do to produce this result?"

Accessing the timeline​

From the Goal Progress panel:

While a goal is running (or after it completes), click View Timeline at the bottom of the goal panel.

From the chat:

After a goal completes, a "View full execution log" link appears in the completion message.

Reading the timeline​

Goal: "Research NVIDIA Q1 2026 earnings and create a slide deck"
Status: Completed  •  8 iterations  •  47.2 seconds

──────────────────────────────────────────────────────────
[Iteration 1]  09:15:32  ──────────────────────────  0.8s
  Action:   web_search
  Args:     { "query": "NVIDIA Q1 2026 earnings results" }
  Result:   8 results returned
  Tokens:   412 prompt / 89 completion

[Iteration 2]  09:15:33  ──────────────────────────  1.4s
  Action:   web_fetch
  Args:     { "url": "https://investor.nvidia.com/..." }
  Result:   Revenue: $26.0B (+262% YoY), EPS: $5.98 (beat by $0.12)
  Tokens:   1,847 prompt / 156 completion

[Iteration 3]  09:15:35  ──────────────────────────  1.1s
  Action:   web_search
  Args:     { "query": "NVIDIA Q1 2026 analyst reactions Wall Street" }
  Result:   6 results returned

[Iteration 4]  09:15:36  ──────────────────────────  0.9s
  Action:   web_fetch
  Args:     { "url": "https://seekingalpha.com/..." }
  Result:   Analyst consensus: Buy. Price target raised to $1,100.

[Iteration 5]  09:15:38  ──────────────────────────  3.2s
  Action:   execute_python
  Args:     { "script": "# Structure earnings data as JSON\n..." }
  Result:   { "revenue": 26.0, "eps": 5.98, "yoy_growth": 262, ... }

[Iteration 6]  09:15:42  ──────────────────────────  6.1s
  Action:   generate_powerpoint
  Args:     { "title": "NVIDIA Q1 2026 Earnings", "slides": [...] }
  Result:   File created: nvidia_q1_2026.pptx (8 slides, 2.4 MB)

[Iteration 7]  09:15:48  ──────────────────────────  0.4s
  Action:   validate_file
  Args:     { "path": "nvidia_q1_2026.pptx" }
  Result:   File exists, 8 slides, non-empty ✅

[Iteration 8]  09:15:49  ──────────────────────────  0.3s
  Assessment: All success criteria met
  → Goal COMPLETED
──────────────────────────────────────────────────────────
Total: 8 iterations  47.2s  2,847 prompt tokens  512 completion tokens  ~$0.03

Self-correction in the timeline​

When a script fails and self-corrector fires, it appears as extra iterations:

[Iteration 5]  ──────────  2.1s  ❌ FAILED
  Action:   execute_python
  Result:   AttributeError: 'NoneType' object has no attribute 'text'

[Iteration 5a] ──────────  1.8s  (self-correction)
  SelfCorrector analyzing error...
  Diagnosis: price element selector changed — updated XPath

[Iteration 5b] ──────────  2.4s  ✅ RETRY
  Action:   execute_python  (corrected script v2)
  Result:   Price: ₹2,847.50

Self-corrections are labeled with letter suffixes (5a, 5b, 5c) and include the diagnosis the SelfCorrector produced.

Timeline filtering​

The timeline panel has filters:

  • Show all — every iteration including successful ones
  • Errors only — iterations that failed or triggered self-correction
  • File operations — iterations that created, read, or wrote files
  • External calls — iterations that called external APIs (web, integrations)

Using the timeline for debugging​

If a goal produced an unexpected result, the timeline shows exactly where it went wrong:

  1. Find the first iteration marked with ❌
  2. Expand the full result — often the actual error message is there
  3. Look at the args — did the agent use the right query/URL/parameters?
  4. Check the self-correction attempt — did it diagnose the error correctly?

Common patterns:

  • Web scraping failures: the page structure changed, or the site returned a CAPTCHA — look for "selector not found" in iteration results
  • Wrong data: the agent found results for the wrong ticker/company — the search query may need more specificity in your goal
  • Timeout at iteration 40: the goal hit the max iteration limit — try breaking it into smaller sub-goals

Comparing timelines across runs​

Run the same goal twice and compare timelines to see:

  • Whether the skill learned from the first run made the second faster
  • Whether a self-correction produced a better script version
  • Token usage trends across iterations (useful for cost optimisation)

Past goals and their timelines stay available — open any completed run from the goal list and read its timeline exactly as you would a live one.