context anchors analysis

what are context anchors?

threads that explicitly reference prior work via:

Continuing work from thread T-... (spawn pattern)
Continuing from https://ampcode.com/threads/... (URL pattern)
read_thread tool usage
explicit thread links in db

sample sizes

cohort	n
anchored	1,507
unanchored	1,981

success rates

metric	anchored	unanchored	delta
resolved	40.5%	43.9%	-3.4pp
committed	9.0%	6.9%	+2.1pp
success (res+comm)	49.5%	50.8%	-1.3pp
frustrated	0.3%	0.4%	-0.1pp
handoff	34.5%	1.8%	+32.7pp

key finding: continuity vs isolated success

metric	anchored	unanchored	delta
continuity rate	84.0%	52.6%	+31.4pp

continuity = resolved + committed + handoff (thread doesn’t die without purpose)

interpretation

anchored threads are not “more successful” per-thread - they have marginally lower single-thread resolution rates (-1.3pp). this is expected: they’re fragments of larger workflows.

but anchored threads almost never die pointlessly. 84% either finish their piece or hand off cleanly vs 52.6% for unanchored threads.

the high UNKNOWN rate for unanchored (864/1981 = 43.6%) suggests many threads start, do some work, and just… stop. anchored threads have lower UNKNOWN (231/1507 = 15.3%).

anchor type breakdown

type	count
explicit_context (spawn pattern)	1,506
read_thread_tool only	1

almost all anchoring comes from the spawn pattern Continuing work from thread T-.... the read_thread tool is rarely the sole anchor (usually combined with explicit context).

implications

multi-thread orchestration works - spawned sub-agents complete or hand off 84% of the time
context passing is valuable - anchored threads have clear purpose and know when to stop
unanchored threads need better termination signals - nearly half end without clear resolution
the “continuing from” pattern should be standard - it creates accountability chains

caveats

anchored threads are often spawned for specific, scoped tasks (easier to complete)
unanchored threads include exploratory/quick sessions that inflate UNKNOWN
the +2.1pp commit rate for anchored suggests they’re more likely to ship when they do succeed

raw data

{
  "anchored": {
    "total": 1507,
    "byStatus": {
      "RESOLVED": 610,
      "COMMITTED": 136,
      "HANDOFF": 520,
      "UNKNOWN": 231,
      "EXPLORATORY": 6,
      "FRUSTRATED": 4
    },
    "rates": {
      "resolved": "40.5",
      "committed": "9.0",
      "handoff": "34.5",
      "success_combined": "49.5"
    },
    "continuity_rate": "84.0"
  },
  "unanchored": {
    "total": 1981,
    "byStatus": {
      "UNKNOWN": 864,
      "COMMITTED": 136,
      "RESOLVED": 870,
      "FRUSTRATED": 7,
      "EXPLORATORY": 61,
      "PENDING": 7,
      "HANDOFF": 36
    },
    "rates": {
      "resolved": "43.9",
      "committed": "6.9",
      "handoff": "1.8",
      "success_combined": "50.8"
    },
    "continuity_rate": "52.6"
  }
}