#1199 stage 16: the long-form spikes ship dark #1217

Activity

1. dkackman

   added

   owner:testerTester's turn to actTester's turn to act

   status:fixed-pending-verifyImplementer fixed, awaiting tester verificationImplementer fixed, awaiting tester verification

   stageA child issue: one stage of a featureA child issue: one stage of a feature

   target:lem

   on Nov 7, 2026
2. dkackman

   marked this as blocking on Nov 7, 2026

   • Long-form media: one playbook, measured levers, quality-first defaults for pieces longer than one clip #1199
3. dkackman commented on Nov 7, 2026

   

   dkackman

   on Nov 7, 2026

   OwnerAuthor

   More actions

   Verified by the tester, model claude-opus-5-5 via provider anthropic, against lem develop @ 85bcd335, main server (no DW_EXPERIMENTAL_REFINE_WINDOWS), workspace bench.

   No suite case carries pending: #1217 (searched every regression-suite-*.md), so this verify rests on the stage's own acceptance, checked over MCP:

   Each key is refused and the refusal names the flag

   • experimental_latent_carry: validate_workflow(name="s2-arc", arguments={"latent_carry": true, "segments": 2}) gives valid: false at steps[0].pipeline.chain.experimental_latent_carry: "...a scratch spike, off unless DW_EXPERIMENTAL_REFINE_WINDOWS=1 is set where the pipeline loads".
   • experimental_tst: s2-arc with tst={"tau":0.5,"per_head":false} and attention_backend=null gives valid: false at steps[0].pipeline.arguments.experimental_tst, naming the flag.
   • experimental_selflift: s2-arc with selflift={"lift_step":2,"start_scale":0.5} at 1024x576 gives valid: false at ...arguments.experimental_selflift, naming the flag.
   • keystone_slot sigma-gated form: s1-static with keystone="segment_1" and keystone_slot={"frame":60,"enter_below_sigma":0.5} gives valid: false: "keystone_slot enter_below_sigma is a scratch spike (the windowed denoiser runs the gate), off unless DW_EXPERIMENTAL_REFINE_WINDOWS=1...". Control: the plain {"frame":60} form without the gate is not refused as experimental (it gets the ordinary multiple-of-17 error).
   • experimental_capstone: validate passes it, but run_workflow(s2-arc, {"capstone": true, "segments": 2}) (job ff57912f8f48) failed after 7 s in phase loading with denoise_step: null: "Chain experimental_capstone is a scratch spike, off unless DW_EXPERIMENTAL_REFINE_WINDOWS=1 is set where the pipeline loads".
   • experimental_refine_windows: as a chain key, validate rejects it as an unknown chain property. As an H3 call argument (inline workflow), validate passes it, but the run (job 023e5d2726ce) failed after 3 s in phase loading with denoise_step: null: "experimental_refine_windows is experimental and off in this process - start the server with DW_EXPERIMENTAL_REFINE_WINDOWS=1".

   All six are refused before any denoise, and every refusal names the flag. Both failed run directories were deleted.

   No catalog template sets any of them. templates/minimax/chained-segments (the H3 chain template, the one that also takes keystone_slot) has no experimental_* key in its chain or arguments, and its keystone_slot presets use only last and null. The keys appear only in the bench workspace's own s1-static/s2-arc probe workflows (origin workspace, not catalog), where every one defaults to null. I didn't open every catalog template; this rests on the chained-shot listing plus that template's full definition. I also searched the loaded dw plugin skills for experimental_capstone, experimental_refine_windows and keystone_slot and found nothing.

   Matches the acceptance. Closing as verified. No new regression case: a security-style gate case for these spikes would be better written from the plan than from this one pass.
4. dkackman

   added

   status:verifiedTester confirmed the fix via a real MCP callTester confirmed the fix via a real MCP call

   and removed

   status:fixed-pending-verifyImplementer fixed, awaiting tester verificationImplementer fixed, awaiting tester verification

   on Nov 7, 2026
5. dkackman

   closed this as completedon Nov 7, 2026