[spending-forecast] Daily spending forecast - 2026-08-29

[github-actions]

Overview

Daily spending forecast for github/gh-aw, generated from gh aw forecast on 2026-08-29 (as_of 2026-08-29T09:44:33Z), using a 30-day history window (history_days: 30) and Monte Carlo projection (10,000 iterations per workflow).

Of the 50 workflows reported, 41 had at least one sampled run and 26 produced non-zero AIC (操作/Infra Cost) observations; 24 workflows show zero AIC because they consume no metered agentic-run cost (standard CI/CodeQL/Dependabot/build jobs) or had zero-cost/failed sampled runs.

Executive summary

• Total observed AIC across all sampled runs (30-day window): $45,241.64 across 988 sampled runs (41 active workflows).
• Aggregate weekly forecast (sum across active workflows): P10 (10th percentile — optimistic) $3,351.36, P50 (50th percentile — median) $8,374.37, P90 (90th percentile — conservative) $15,208.83.
• Aggregate monthly forecast (sum across active workflows): P10 (10th percentile — optimistic) $23,547.09, P50 (50th percentile — median) $37,068.05, P90 (90th percentile — conservative) $53,686.77.
• Top 3 cost drivers by observed 30-day AIC: Go Logger Enhancement ($9,485), Agentic Workflow Audit Agent ($7,993), Semantic Function Refactoring ($6,376) — together ~53% of total observed spend.

Charts

Key metrics — active workflows (sampled_runs > 0, sorted by observed 30-day AIC)

Workflow	Sampled runs	Observed AIC (30d)	P50/run AIC	P95/run AIC	Weekly P50 (median)	Monthly P50 (median)	Success rate	Monthly P10–P90 range
Go Logger Enhancement	31	9,485.35	323.33	466.63	1,641.7	7,327.2	77%	4,877–10,223
Agentic Workflow Audit Agent	30	7,992.92	273.94	408.03	1,519.5	6,658.7	83%	4,433–9,195
Semantic Function Refactoring	31	6,376.34	219.12	315.77	1,256.9	5,553.9	87%	3,702–7,700
CLI Version Checker	31	4,336.37	128.08	300.61	928.5	4,052.5	94%	2,761–5,621
Tidy	32	2,320.86	67.10	150.55	424.6	1,880.4	81%	1,248–2,610
Copilot Agent PR Analysis	29	1,904.71	73.67	94.49	359.7	1,565.6	83%	1,029–2,225
Lockfile Statistics Analysis Agent	30	1,843.25	62.67	88.05	406.4	1,784.6	97%	1,218–2,432
Dev	31	1,585.24	57.42	118.42	275.1	1,225.5	77%	761–1,765
Smoke Copilot	42	1,397.65	43.20	60.75	126.3	557.7	40%	336–818
Duplicate Code Detector	30	1,372.61	17.66	141.45	183.7	858.1	63%	471–1,337
Smoke Claude	17	1,247.44	76.93	110.74	231.9	1,028.7	82%	581–1,568
Daily 新建s	23	962.67	39.93	83.54	207.5	922.9	96%	592–1,329
Terminal Stylist	32	841.70	26.05	36.27	167.6	737.8	88%	505–1,008
GitHub MCP Remote Server Tools Report Generator	4	835.27	238.39	248.69	238.4	835.3	100%	244–1,811
Scout	10	692.95	74.39	131.36	109.3	473.1	70%	174–881
Weekly Workflow Analysis	4	607.72	137.68	214.73	137.7	603.7	100%	168–1,313
Documentation Unbloat	32	440.75	14.40	23.76	88.5	386.2	88%	259–532
Daily Documentation Updater	31	303.07	0.00	33.84	61.9	280.1	94%	166–421
Weekly Issue Summary	5	249.16	67.23	113.46	0.0	68.5	40%	0–250
Smoke Codex	19	186.81	2.53	56.04	9.3	91.4	53%	21–190
Artifacts Usage Report	5	120.29	29.32	46.55	0.0	73.7	60%	0–167
Smoke OpenCode	2	85.00	33.60	51.40	0.0	85.0	100%	0–237
仓库 Tree Map Generator	5	53.51	17.36	18.76	0.0	17.4	40%	0–54

Remaining zero-cost / no-sample workflows (24)

These reported sampled_runs > 0 but avg_aic = 0 (standard CI/build/scan jobs with no metered agentic cost), or sampled_runs = 0 (no runs in the 30-day window):

Workflow	Sampled runs	Success rate	Note
Doc Build - Deploy	85	68%	zero AIC (build job, no agent cost)
Copilot cloud agent	98	99%	zero AIC (billed elsewhere)
CodeQL	100	100%	zero AIC (scanning job)
Dependabot Updates	77	91%	zero AIC (dependency bot)
CI	74	53%	zero AIC (standard CI)
Go Pattern Detector	22	100%	zero AIC
Copilot Setup Steps	11	100%	zero AIC (setup job)
MCP Inspector Agent	5	0%	zero AIC, all runs failed
Mergefest	1	0%	single failed run
Notion Issue Summary	1	0%	single failed run
Plan Command	1	0%	single failed run
Poem Bot - A Creative Agentic Workflow	1	0%	single failed run
Rebuild the documentation after making changes	1	0%	single failed run
Resource Summarizer Agent	1	0%	single failed run
Commit Changes Analyzer	1	0%	single failed run
Basic Research Agent	1	0%	single failed run
Video Analysis Agent	1	0%	single failed run
Dev Hawk	1	0%	single failed run
Sentry Issue Analyzer	0	—	no runs in window
Q	0	—	no runs in window
.github/workflows/test-proxy	0	—	no runs in window
CI Failure Doctor	0	—	no runs in window
Test	0	—	no runs in window
Test Claude	0	—	no runs in window
Test Copilot CLI Engine	0	—	no runs in window
Test Copilot GitHub Integration	0	—	no runs in window
Format, Lint, Build and Commit	0	—	no runs in window

Data quality & accuracy notes

• All non-zero-AIC workflows share a consistent 30-day history window (history_days: 30), so cross-workflow comparisons in the table above are apples-to-apples. No inconsistent date windows were found.
• Sparse samples (n < 5) inflate confidence-interval width and Monte Carlo is explicitly flagged is_reliable: false for these: GitHub MCP Remote Server Tools Report Generator (n=4), Weekly Workflow Analysis (n=4), Smoke OpenCode (n=2), Weekly Issue Summary (n=5), Artifacts Usage Report (n=5), 仓库 Tree Map Generator (n=5), and the 11 workflows with a single sampled run (Mergefest, Notion Issue Summary, Plan Command, Poem Bot, Rebuild the documentation, Resource Summarizer Agent, Commit Changes Analyzer, Basic Research Agent, Video Analysis Agent, Dev Hawk, MCP Inspector Agent). Impact: their P10–P90 ranges (e.g. 仓库 Tree Map Generator: $0–$54, a 3x P50 spread) should be treated as directional, not budget-grade, until more runs accumulate. These are excluded from the top-driver ranking above but included in the appendix for completeness.
• Zero AIC is expected, not a data gap, for standard CI/build/scan workflows (CodeQL, CI, Dependabot Updates, Doc Build - Deploy, Copilot cloud agent, Copilot Setup Steps, Go Pattern Detector) — these do not run the metered agentic engine, so avg_aic = 0 correctly reflects no cost, and the forecaster reasonably omits weekly_monte_carlo/monthly_monte_carlo sections for the 9 workflows with sampled_runs = 0.
• No implausible run frequencies detected. observed_runs_per_period values (e.g. 31 for daily-triggered workflows over a 30-day window) are consistent with expected trigger cadence; no workflow showed run counts inconsistent with its schedule.
• No workflows were missing from the report relative to the repository's active .github/workflows/*.md set as far as could be cross-checked from forecast.json alone; a full reconciliation against the live workflow list was not performed since gh aw forecast already enumerates all discovered workflows including zero-run ones.
• The prepared forecast.json output was internally consistent (field types, percentile ordering P10 ≤ P50 ≤ P90 held for all active workflows) — no rerun of gh aw forecast was needed.

Assumptions

• Costs are expressed in AIC (操作/Infra Cost) units as reported by gh aw forecast; no currency conversion applied.
• Weekly/monthly aggregates are simple sums of each workflow's independent Monte Carlo percentile, not a joint simulation — true portfolio-level P10/P90 may be narrower than the naive sum implies.
• Workflows with sampled_runs = 0 or all-zero AIC are excluded from the executive-summary totals but retained in the appendix for transparency.

Forecast date: 2026-08-29 · History window: 30 days · Run: §33245968459

Generated by 📈 Daily Spending Forecast · copilot · auto · 52.9 AIC · ⌖ 5.13 AIC · ⊞ 11.3K · ◷

• expires on Sep 5, 2026, 1:51 AM UTC-08:00