EXTRA diagnostic (not in the plan), 2026-10-20 ~11:25 UTC+8: why github_issues and pypi_info saved 0%.
Same messages as c1b, compress(..., config=CompressConfig(protect_recent=0)), Python logging at DEBUG. Only Headroom's routing decisions are kept below (payload text removed).

DEBUG:headroom.transforms.content_router:content_router adaptive: pressure=0.91, min_ratio=1.00, read_protect_window=4/4 msgs
content_router_output: selected_strategy=mixed; routing_log=[{"content_type": "text", "strategy": "passthrough", "original_tokens": 211093, "compressed_tokens": 211093, "confidence": 1.0, "section_index": 0, "compression_ratio": 1.0}]
INFO:headroom.transforms.content_router:content_router: 4 msgs — 1 skipped (user), 1 skipped (<250 tok msg / <500 chars block), 1 unchanged (ratio>=1.00), 1 non-string, 1 cache misses, cache[0 results, 1 skips, 2999ns avg]
INFO:headroom.transforms.content_router:[router] route_counts={'user_msg': 1, 'small': 1, 'ratio_too_high': 1, 'non_string': 1, 'cache_miss': 1} compressed=0 frozen=0 msgs=4
DEBUG:headroom.transforms.pipeline:Pipeline complete: no token savings [_deep_copy=0ms _initial_token_count=40ms content_router=185ms parallel_compress_total=133ms compressor:mixed=133ms _final_token_count=0ms pipeline_total=185ms]
content_router_output: selected_strategy=mixed; routing_log=[{"content_type": "json_array", "strategy": "passthrough", "original_tokens": 78029, "compressed_tokens": 78029, "confidence": 1.0, "section_index": 0, "compression_ratio": 1.0}]
INFO:headroom.transforms.content_router:content_router: 4 msgs — 1 skipped (user), 1 skipped (<250 tok msg / <500 chars block), 1 unchanged (ratio>=1.00), 1 non-string, 1 cache misses, cache[0 results, 2 skips, 5180ns avg]
INFO:headroom.transforms.content_router:[router] route_counts={'user_msg': 1, 'small': 1, 'ratio_too_high': 1, 'non_string': 1, 'cache_miss': 1} compressed=0 frozen=0 msgs=4
DEBUG:headroom.transforms.pipeline:Pipeline complete: no token savings [_deep_copy=0ms _initial_token_count=16ms content_router=90ms parallel_compress_total=80ms compressor:mixed=80ms _final_token_count=0ms pipeline_total=91ms]

Reading: the router classified each payload as one section, chose strategy 'passthrough' and left it unchanged (ratio 1.00). The README says 'Savings scale with how repetitive the payload is ... prose and already-dense output compress very little'; both payloads are JSON arrays whose objects are mostly unique prose (issue bodies, package summaries).