BANTAM FACTORY
11 / 11jobs completed
All matchups recordedREAL WORK. ON RECORD.
Put the factory
in the ring.
Choose a job. Pick two contenders. See every action, test, and delivered file.
11/11 jobs published71 recorded attempts
THE HARNESS MAKES A DIFFERENCE
Same model. Different results.
11 published tasks. The same local Qwen 27B weights. A completed job means accepted work and a clean finish.
DeepSeek Harness
7 / 11jobs completed
All matchups recordedHermes
5 / 11jobs completed
All matchups recordedOpenCode
7 / 11jobs completed
All matchups recordedPi
4 / 6jobs completed
6 of 11 recorded · 5 pendingCompleted jobs / recorded attempts. Unrun matchups stay pending. Each card includes its recording conditions. Open the fights ↓
THE FIGHT CARDS
Let the work talk.
Open a card to compare the work side by side. Search by job or harness.
Codex, inside the factory.
Solo and supervisedAstra, Sol and Terra in BANTAM FACTORY and their native CLI. Also watch Astra supervise a Terra or Sol worker, with both roles counted. Compare time, tokens, tests and delivered code.
REPAIR01 / 23
Astra · job planner · round 5
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS159.9s
BANTAM FACTORY · AstraPASS109.0s
Complete token accounting.
Open fight
REPAIR02 / 23
Astra · job planner · round 6
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS103.4s
BANTAM FACTORY · AstraPASS99.7s
Complete token accounting.
Open fight
EXTEND07 / 23
Snapshot checker · latest round
4 contendersThe updated factory extends a real file checker. Compare native Astra, factory Astra, and Astra supervising Terra or Sol, with every result and both roles counted.
Codex · AstraPASS139.5s
BANTAM FACTORY · AstraPASS122.8s
BANTAM FACTORY · Astra → TerraPASS199.0s
BANTAM FACTORY · Astra → SolPASS313.3s
Complete token accounting.
Open fight
BUILD08 / 23
Context packer · latest round
4 contendersAstra in native Codex, Astra inside BANTAM FACTORY, and Astra supervising Terra or Sol. Four fresh attempts with the updated factory; every result and both supervised roles included.
Codex · AstraPASS102.4s
BANTAM FACTORY · AstraPASS98.6s
BANTAM FACTORY · Astra → TerraPASS228.3s
BANTAM FACTORY · Astra → SolPASS192.6s
Complete token accounting.
Open fight
REPAIR09 / 23
Stream decoder · latest round
4 contendersAstra in native Codex, Astra inside BANTAM FACTORY, and Astra supervising Terra or Sol. Four fresh attempts with the updated factory; every result and both supervised roles included.
Codex · AstraPASS144.1s
BANTAM FACTORY · AstraPASS127.4s
BANTAM FACTORY · Astra → TerraPASS353.3s
BANTAM FACTORY · Astra → SolPASS492.6s
Complete token accounting.
Open fight
Earlier recordings 18 cards · complete records
REPAIR03 / 23
Astra · job planner · round 3
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS157.9s
BANTAM FACTORY · AstraPASS95.4s
Complete token accounting.
Open fight
REPAIR04 / 23
Astra · job planner · round 4
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS163.1s
BANTAM FACTORY · AstraPASS110.3s
Complete token accounting.
Open fight
REPAIR05 / 23
Astra · job planner · round 1
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS97.1s
BANTAM FACTORY · AstraPASS105.4s
Complete token accounting.
Open fight
REPAIR06 / 23
Astra · job planner · round 2
2 contendersThe same Astra repairs a dependency planner in native Codex and BANTAM FACTORY. Both paired repeats are published, with every action, delivered file, token counter and independent check.
Codex · AstraPASS149.4s
BANTAM FACTORY · AstraPASS107.1s
Complete token accounting.
Open fight
EXTEND10 / 23
Snapshot checker · first round
4 contendersFour complete recorded workflows from an earlier factory revision. Follow the edits, reviews and tests that led to the next improvement.
Codex · AstraPASS134.2s
BANTAM FACTORY · AstraPASS139.9s
BANTAM FACTORY · Astra → TerraPASS460.0s
BANTAM FACTORY · Astra → SolPASS478.6s
Partial accounting disclosed in the replay.
Open fight
EXTEND11 / 23
Snapshot checker · batch edit round
4 contendersFour complete recorded workflows from an earlier factory revision. Follow the edits, reviews and tests that led to the next improvement.
Codex · AstraPASS135.6s
BANTAM FACTORY · AstraPASS152.9s
BANTAM FACTORY · Astra → TerraPASS462.3s
BANTAM FACTORY · Astra → SolPASS447.6s
Complete token accounting.
Open fight
EXTEND12 / 23
Snapshot checker · root review round
4 contendersFour complete recorded workflows from an earlier factory revision. Follow the edits, reviews and tests that led to the next improvement.
Codex · AstraPASS143.1s
BANTAM FACTORY · AstraPASS138.6s
BANTAM FACTORY · Astra → TerraPASS470.0s
BANTAM FACTORY · Astra → SolPASS486.5s
Complete token accounting.
Open fight
BUILD13 / 23
Context packet
6 contendersFit complete, attributable context sections into a strict UTF-8 byte budget.
Codex · AstraPASS106.3s
Codex · SolPASS329.4s
Codex · TerraPASS123.4s
BANTAM FACTORY · AstraPASS106.6s
BANTAM FACTORY · SolOUTPUT_ONLY221.8s
BANTAM FACTORY · TerraPASS52.8s
Partial accounting disclosed in the replay.
Open fight
BUILD14 / 23
Astra · efficiency round 1
2 contendersThe same context-packet work order, in native Codex and BANTAM FACTORY. A fresh paired run after the context improvements.
Codex · AstraPASS107.5s
BANTAM FACTORY · AstraPASS99.2s
Complete token accounting.
Open fight
BUILD15 / 23
Astra · efficiency round 2
2 contendersThe same context-packet work order, in native Codex and BANTAM FACTORY. A fresh paired run after the context improvements.
Codex · AstraPASS98.0s
BANTAM FACTORY · AstraPASS104.9s
Complete token accounting.
Open fight
BUILD16 / 23
Sol · efficiency round 1
2 contendersThe same context-packet work order, in native Codex and BANTAM FACTORY. A fresh paired run after the context improvements.
Codex · SolPASS225.3s
BANTAM FACTORY · SolPASS138.8s
Complete token accounting.
Open fight
BUILD17 / 23
Sol · efficiency round 2
2 contendersThe same context-packet work order, in native Codex and BANTAM FACTORY. A fresh paired run after the context improvements.
Codex · SolPASS219.7s
BANTAM FACTORY · SolPASS106.8s
Complete token accounting.
Open fight
BUILD18 / 23
Astra · context packer · round 1
2 contendersAstra with a leaner factory context, against native Codex. The same task and checks, with both paired rounds published.
Codex · AstraPASS112.8s
BANTAM FACTORY · AstraPASS86.9s
Complete token accounting.
Open fight
BUILD19 / 23
Astra · context packer · round 2
2 contendersAstra with a leaner factory context, against native Codex. The same task and checks, with both paired rounds published.
Codex · AstraPASS103.0s
BANTAM FACTORY · AstraPASS89.3s
Complete token accounting.
Open fight
EXTEND20 / 23
Astra · patch transaction · round 1
2 contendersAstra with a leaner factory context, against native Codex. The same task and checks, with both paired rounds published.
Codex · AstraPASS98.8s
BANTAM FACTORY · AstraPASS86.4s
Complete token accounting.
Open fight
EXTEND21 / 23
Astra · patch transaction · round 2
2 contendersAstra with a leaner factory context, against native Codex. The same task and checks, with both paired rounds published.
Codex · AstraPASS85.5s
BANTAM FACTORY · AstraPASS89.8s
Complete token accounting.
Open fight
BUILD22 / 23
Context packer · solo and supervised
8 contendersNative Codex, BANTAM workers, and Astra supervising Terra or Sol. The same assignment and independent checks; both roles included in supervised totals.
Codex · AstraPASS105.4s
Codex · SolPASS241.1s
Codex · TerraPASS99.1s
BANTAM FACTORY · AstraPASS97.2s
BANTAM FACTORY · SolPASS165.1s
BANTAM FACTORY · TerraPASS83.3s
BANTAM FACTORY · Astra → TerraPASS190.4s
BANTAM FACTORY · Astra → SolPASS203.1s
Complete token accounting.
Open fight
REPAIR23 / 23
Stream decoder · solo and supervised
4 contendersNative Codex, BANTAM workers, and Astra supervising Terra or Sol. The same assignment and independent checks; both roles included in supervised totals.
Codex · AstraPASS121.9s
BANTAM FACTORY · AstraPASS141.4s
BANTAM FACTORY · Astra → TerraPASS375.8s
BANTAM FACTORY · Astra → SolPASS505.0s
Complete token accounting.
Open fight
Launch series
6 work ordersThe first six work orders: build, extend and repair, against local harnesses and frontier CLIs.
BUILD01 / 06
Context packet
6 contendersFit complete, attributable context sections into a strict UTF-8 byte budget.
BANTAM FACTORY · localPASS58.1s
DeepSeek HarnessOUTPUT_ONLY600.0s
OpenCodeOUTPUT_ONLY600.0s
HermesPASS512.2s
PiPASS350.7s
Codex · AstraPASS110.6s
NVIDIA RTX 4090 · 24 GB105.2 tok/s generation
Partial accounting disclosed in the replay.
Open fight Try the tool it built ↗
EXTEND02 / 06
Patch transaction
6 contendersCheck every preimage, then apply a set of compatible edits as one transaction.
BANTAM FACTORY · localPASS81.8s
DeepSeek HarnessOUTPUT_ONLY600.0s
OpenCodePASS559.0s
HermesTIMEOUT600.0s
PiPASS457.2s
Codex · AstraPASS90.3s
NVIDIA RTX 4090 · 24 GB82.6 tok/s generation
Partial accounting disclosed in the replay.
Open fight
REPAIR03 / 06
Stream framer
6 contendersRecover complete events across arbitrary byte chunks, with explicit stream termination.
BANTAM FACTORY · localPASS432.7s
DeepSeek HarnessFAIL552.0s
OpenCodeTIMEOUT600.0s
HermesTIMEOUT600.0s
PiTIMEOUT600.0s
Codex · AstraPASS160.7s
NVIDIA RTX 4090 · 24 GB84.8 tok/s generation
Partial accounting disclosed in the replay.
Open fight
BUILD04 / 06
Receipt reducer
8 contendersTurn a noisy stream of events into a trustworthy account of what happened.
BANTAM FACTORY · localPASS112.5s
DeepSeek HarnessPASS442.7s
OpenCodePASS480.1s
HermesPASS577.6s
PiPASS308.1s
Codex · AstraPASS114.1s
Codex · SolPASS166.7s
Codex · TerraPASS107.2s
NVIDIA RTX 4090 · 24 GB102.6 tok/s generation
Partial accounting disclosed in the replay.
Open fight
EXTEND05 / 06
Snapshot drift
8 contendersBind a snapshot to real files. Detect what changed, and validate what did not.
BANTAM FACTORY · localPASS124.3s
DeepSeek HarnessPASS465.8s
OpenCodePASS403.2s
HermesPASS497.0s
PiOUTPUT_ONLY600.0s
Codex · AstraPASS146.2s
Codex · SolFAIL513.4s
Codex · TerraPASS106.2s
NVIDIA RTX 4090 · 24 GB88.4 tok/s generation
Partial accounting disclosed in the replay.
Open fight
REPAIR06 / 06
Job planner
8 contendersOrder dependent work deterministically. Propagate failure without losing the plan.
BANTAM FACTORY · localPASS193.2s
DeepSeek HarnessPASS483.5s
OpenCodePASS324.1s
HermesTIMEOUT600.0s
PiPASS483.5s
Codex · AstraPASS118.4s
Codex · SolPASS240.9s
Codex · TerraPASS119.3s
NVIDIA RTX 4090 · 24 GB100.5 tok/s generation
Partial accounting disclosed in the replay.
Open fight
Controls
2 work ordersDeterministic edit receipts and retry arithmetic, where a wrong boundary is silent until it is not.
BUILD01 / 02
Redaction plan
7 contendersRemove selected literals deterministically, with an exact edit receipt and byte accounting.
BANTAM FACTORY · localPASS133.2s
DeepSeek HarnessPASS391.8s
OpenCodePASS341.6s
HermesTIMEOUT600.0s
Codex · AstraPASS117.0s
Codex · SolPASS241.2s
Codex · TerraPASS134.3s
NVIDIA RTX 4090 · 24 GB97.0 tok/s generation
Partial accounting disclosed in the replay.
Open fight
REPAIR02 / 02
Retry budget
7 contendersRespect backoff, server retry hints and a hard remaining-time budget without overflow.
BANTAM FACTORY · localPASS141.7s
DeepSeek HarnessPASS445.8s
OpenCodePASS375.4s
HermesPASS562.0s
Codex · AstraPASS137.3s
Codex · SolPASS231.8s
Codex · TerraPASS143.2s
NVIDIA RTX 4090 · 24 GB87.5 tok/s generation
Partial accounting disclosed in the replay.
Open fight
Selection
3 work ordersChoosing and resolving: which paths match, which stay inside the workspace, which version wins.
BUILD01 / 03
Glob select
5 contendersMatch paths against glob patterns, last pattern deciding, with a receipt per path.
BANTAM FACTORY · localPASS264.1s
DeepSeek HarnessFAIL551.4s
OpenCodeTIMEOUT600.0s
HermesTIMEOUT600.0s
Codex · AstraPASS155.8s
NVIDIA RTX 4090 · 24 GB84.8 tok/s generation
Partial accounting disclosed in the replay.
Open fight
EXTEND02 / 03
Path scope
5 contendersDecide whether a path stays inside a workspace root, without trusting a shared prefix.
BANTAM FACTORY · localPASS75.1s
DeepSeek HarnessPASS313.6s
OpenCodePASS424.1s
HermesPASS275.6s
Codex · AstraPASS107.9s
NVIDIA RTX 4090 · 24 GB93.6 tok/s generation
Partial accounting disclosed in the replay.
Open fight
REPAIR03 / 03
Semver range
5 contendersOrder versions by real precedence and admit prereleases into a range only when invited.
BANTAM FACTORY · localPASS399.9s
DeepSeek HarnessPASS536.4s
OpenCodeTIMEOUT600.0s
HermesOUTPUT_ONLY600.0s
Codex · AstraPASS136.2s
NVIDIA RTX 4090 · 24 GB84.7 tok/s generation
Partial accounting disclosed in the replay.
Open fight
No cards match this view.
How to read a fight card Clocks, checks & conditions +
PASS requires both accepted work and clean completion. OUTPUT_ONLY means the artifact passed but the run did not finish cleanly. A timeout is not time to successful completion. Missing token counters remain unknown, never zero.
The roster shows which systems ran each task. A solo card contains one recorded attempt. Completed comparisons include every planned contender. These are selected development work orders; the recording dates and conditions are attached to each card.
Time bars show elapsed run time, not progress or correctness. Local and frontier models are distinguished in the replay. Read each card's conditions for configuration and accounting details. Open a contender to read its actions, inspect its delivered files and see its test results. Cards with a demo also let you try the recorded tool. This static site loads no external assets and sends no telemetry.