PortentDevRequest access
Knowledge bank

Not a prompt. A version-controlled corpus

Every lesson Portent applies is retrievable, actionable, and traced to a real event. It's categorized on four axes so knowledge learned in one place never leaks into a place it doesn't belong, and it grows only through a human-gated, ratcheted pipeline.

342
canonical lessons
promoted & attributed
~10,700
candidates in review
categorized, awaiting gate
109
detect → fix patterns
compiled, deterministic
994
tests green
coverage ratchet
Four-axis categorization

Every lesson is tagged: layer × substrate × framework × subject

The tagging is what keeps an EFA lesson from being applied on InfiniBand, or a 70B-FSDP tuning from bleeding into an inference job. The axes and categories below are the real taxonomy; the bar magnitudes are illustrative until we publish the breakdown.

by layerillustrative
  • L6 fabric
  • L7 node/driver
  • L3 comms
  • L2 framework
  • L0/L1 kernel
  • L8 harness
by substrateillustrative
  • EFA
  • InfiniBand
  • RoCE
  • TCP
  • NVLink
  • single-host
by frameworkillustrative
  • FSDP
  • Megatron
  • vLLM
  • SGLang
  • TRT-LLM
by subjectillustrative
  • misconfig
  • perf tuning
  • correctness
  • bring-up order
  • provenance
Sample lessons

What a lesson looks like: a rule and its negative prime

Each canonical lesson pairs an actionable rule with the mistake it prevents. Every customer, account, and host specific has been scrubbed.

L6EFA attributed
Rule

Validate multi-rail routing before enabling RDMA on any non-primary subnet.

Don't do this

Don't assume a secondary subnet inherits the primary's RDMA configuration. It silently drops traffic.

L7substrate-agnostic attributed
Rule

Enumerate NICs on the node and verify the count before the workload starts.

Don't do this

Don't launch against 16 NICs because the instance type 'should' have 16. This one has 8.

L3InfiniBand attributed
Rule

Re-tune the NCCL algorithm and buffer sizes for the actual cluster size.

Don't do this

Don't reuse an 8-node ring configuration at 64 nodes; the topology assumptions no longer hold.

L4TCP attributed
Rule

Verify the transport plugin's ABI matches the running NCCL build.

Don't do this

Don't accept a silent fallback to the socket transport as a 'slow' run; surface it.

How it grows

Mined from real runs, promoted only by a human

Nothing enters the canonical corpus automatically. A candidate is mined, categorized, scrubbed, and verified; then a human approves the promote. The coverage ratchet guarantees every promoted lesson stays retrievable, actionable, and attributed, and the bar never lowers.

  1. 01
    Mine
    every run, pass or fail
  2. 02
    Categorize
    4 axes
  3. 03
    Scrub
    strip sensitive data
  4. 04
    Verify
    zero-false-positive gate
  5. 05
    Human-gate
    a person approves
  6. 06
    Promote
    into the corpus

promote → recall: a promoted lesson is loaded on the next relevant run, closing the loop.