How Save-Game Context Compaction Fixes Spec Drift in Multi-Step Coding Agents (-91.9% Input Tokens)

A developer's benchmark experiment (E12) found that compacting an AI coding agent's context into structured JSON checkpoints at each milestone—dubbed the "Save-Game" pattern—cut cumulative input tokens by 91.9% (35,869 to 2,922) and API cost by 72.2% ($0.0340 to $0.0095) compared with unbounded history accumulation. In a three-milestone power-grid telemetry pipeline run in an isolated Proxmox VM, the accumulated-history agent scored 18/20 on a hidden grading suite, failing two end-to-end tests after losing track of a spec detail mid-context, while the compaction agent passed 20/20.

💡 AI Labo Benchmark Series — Experiment E12 Ranked 6 out of 41 Patterns This post is part of our empirical benchmark series testing AI coding agent harness patterns inside network-isolated Proxmox VMs against a hidden 20-case grading suite. E12 Backlog Candidate C008 e12-20261028-vm216-01 VM216 ai-e12-216 , 0 NICs, virtio-serial QGA relay, systemd UID separation gemini-3.8-flash thinkingLevel: LOW /compact at Milestones — When an AI coding agent tackles a multi-step engineering task across several modules, keeping every prior file dump, telemetry log, and intermediate code output inside a single growing messages array Unbounded Context Accumulation causes two compounding problems: 6,556 ➔ 11,769 ➔ 17,544 input tokens . To test the solution— Milestone Context Compaction the "Save-Game" Pattern —we benchmarked a 3-milestone power-grid telemetry and carbon-credit clearing pipeline Milestone 1: collector.py ➔ Milestone 2: clearing.py ➔ Milestone 3: dispatch.py inside isolated Proxmox VM VM216 : AccumulatedHistoryRunner : VERBOSE TELEMETRY AUDIT DUMP into one continuous chat thread 0 compactions . MilestoneCompactionRunner : 120-token structured JSON checkpoint SAVE GAME COMPACTION V1 Mode A's accumulated history 35,869 cumulative input tokens suffered a classic Lost-in-the-Middle failure on Milestone 2 clearing.py , forgetting to apply .strip .upper to contract tier and failing 2 out of 20 hidden E2E tests 18 / 20 , 90.0% . Mode B's Save-Game Milestone Compaction kept every turn under 1,400 input tokens 2,922 cumulative input tokens, -91.9% reduction , cut API cost by -72.2% $0.0340 ➔ $0.0095 , and achieved a flawless 20 / 20 100% pass . | Metric | Mode A: Accumulated History AccumulatedHistory | Mode B: Milestone State Compaction MilestoneCompaction | Delta / Improvement | |---|---|---|---| | Hidden Grading Suite Score 20 Cases | 18 / 20 90.0% — failed 2 E2E cases | 20 / 20 100% Perfect Pass | +10.0 pt Zero spec drift | | Context Compactions compactions | 0 compactions | 2 compactions After M1 & M2 | Reset history at each milestone gate | | Turn 1 Input Tokens M1: collector.py | 6,556 tokens $0.00725 | 650 tokens $0.00290 | -90.1% reduction | | Turn 2 Input Tokens M2: clearing.py | 11,769 tokens $0.01247 | 1,388 tokens $0.00461 | -88.2% reduction | | Turn 3 Input Tokens M3: dispatch.py | 17,544 tokens $0.01429 | 884 tokens $0.00195 | -95.0% reduction on final turn | | Cumulative Input Tokens promptTokenCount | 35,869 tokens | 2,922 tokens | -91.9% 12.28x token efficiency | | Cumulative Output Tokens candidatesTokenCount | 1,894 tokens | 1,937 tokens | Equivalent code generation volume | | LLM Calls model calls | 3 calls | 3 calls 1 per milestone | Identical call count | | Elapsed Execution Time elapsed seconds | 24.81 sec | 22.52 sec | -9.2% 1.10x faster | | Estimated API Cost estimated micro usd | $0.0340 34,005 micro-USD | $0.0095 9,457 micro-USD | -72.2% 3.60x cheaper | +-------------------------------------------------------------------------+ | Proxmox Isolated VM216 0 NICs, virtio-serial QGA, systemd UID split | | Target Task: 3-Milestone Power Grid & Carbon Clearing Pipeline | | M1: collector.py -- M2: clearing.py -- M3: dispatch.py | | | | +-------------------------------------------------------------------+ | | | Mode A: AccumulatedHistoryRunner No Compaction | | | | - Turn 1 M1 : 6,556 in -- collector.py PASS | | | | - Turn 2 M2 : 11,769 in -- Audit logs drown out M2 spec | | | | Forgets contract tier.strip .upper | | | | - Turn 3 M3 : 17,544 in -- Cumulative 35,869 input tokens | | | | -- Final Score: 18 / 20 Crashes on lowercase/padded tiers | | | +-------------------------------------------------------------------+ | | | | +-------------------------------------------------------------------+ | | | Mode B: MilestoneCompactionRunner Save-Game JSON Compaction | | | | - Turn 1 M1 : 650 in -- collector.py PASS | | | | - Compaction 1 : Discards raw logs; writes 120-tok JSON state | | | | - Turn 2 M2 : 1,388 in -- Fresh context nails all M2 rules | | | | - Compaction 2 : Updates verified contracts in JSON checkpoint | | | | - Turn 3 M3 : 884 in -- Cumulative 2,922 input tokens -91.9% | | | | -- Final Score: 20 / 20 100% Perfect Pass | | | +-------------------------------------------------------------------+ | +-------------------------------------------------------------------------+ AccumulatedHistory : The Missing .strip .upper on Turn 2 Inspecting accum-bundle.json Mode A's final submitted code reveals a textbook example of Attention Dilution : collector.py station id and grid zone . clearing.py VERBOSE TELEMETRY AUDIT DUMP M1 plus the full Turn 1 code had bloated the prompt to 11,769 tokens contract tier via ❌ Mode A accum-bundle.json — clearing.py generated at 11,769 input tokens def settle grid readings, contract tier="BASE", credit offset=0 : if not isinstance readings, list or len readings == 0: raise ValueError "readings must be a non-empty list" BUG: Forgot isinstance contract tier, str and contract tier.strip .upper if contract tier not in ALLOWED TIERS: raise ValueError f"contract tier must be one of {ALLOWED TIERS}" Because of that single omitted .strip .upper , Mode A crashed with a ValueError on two hidden E2E integration tests that passed padded/lowercase tier strings contract tier=" green " in e2e 01 and contract tier="hydro" in e2e 02 . MilestoneCompaction : The Exact Save-Game JSON Checkpoint In Mode B, as soon as Milestone 1 collector.py passed verification, the harness wiped the raw conversation history and started Turn 2 with this compact SAVE GAME COMPACTION V1 JSON checkpoint : { "checkpoint type": "SAVE GAME COMPACTION V1", "completed modules": "collector.py" , "history compacted": true, "milestone completed": 1, "next target module": "clearing.py", "verified contracts": { "collector.collect readings": "PASS returns list of dicts with station id, grid zone in NORTH/SOUTH/WEST, mwh units 0, rate cents =0, 0<=carbon bps<=1500, non-mutating " } } Because Turn 2 ran in a fresh 1,388-token context containing only the verified upstream contract and Milestone 2's specification, Mode B had zero attention dilution—properly implementing clean tier = contract tier.strip .upper and passing all 20 / 20 test cases: | Test Case Group | Verification Target | Mode A Accumulated | Mode B Save-Game Compact | |---|---|---|---| | collector 01 – collector 06 | Milestone 1: Non-mutation, type checks, bounds 0..1500 bps | 6 / 6 PASS | 6 / 6 PASS | | clearing 01 – clearing 08 | Milestone 2: First-seen order, 400 vs 401 bps levy, 12000 tier boundary | 8 / 8 PASS | 8 / 8 PASS | | e2e 01 green below 12000 minus zero and at separator | E2E: Normalize contract tier=" green " + @ delimiter + -0 | ❌ FAIL ValueError | ✅ PASS | | e2e 02 hydro levy and credit dispatch | E2E: Normalize contract tier="hydro" + 20% discount + credit offset | ❌ FAIL ValueError | ✅ PASS | | e2e 03 – e2e 06 | E2E: Zero rate cents, line order preservation, cross-stage error propagation | 4 / 4 PASS | 4 / 4 PASS | | Total | 20 Hidden Cases | 18 / 20 90.0% | 20 / 20 100% | When building multi-step agent workflows or using /compact in interactive coding agents : Lost-in-the-Middle starts happening well above 10,000–15,000 tokens of noisy tool logs. Compact E05 Subagent Folding , natural-language summaries often drop exact type contracts. Instead, record a structured JSON object containing completed modules , verified contracts exact function signatures & invariants , and next target module . E20 : MEMORY.md E21 : E16 : E19 : E14 :