Skip to content

Correct the findings count, the cut table, and DELPHI's place in the portfolio #15

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jadoon200 merged 1 commit into from
Aug 9, 2026
Merged

Correct the findings count, the cut table, and DELPHI's place in the portfolio #15

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,13 +8,22 @@ series and get a verdict. On a free tier, so the first request after an idle per
about a minute to wake the container; the snapshot is baked at image build and the badge
says `demo` because it is.

> **The infrastructure lane of the portfolio.** Sibling to
> [SENTINEL](https://github.com/jadoon200/sentinel) (cyber threat intelligence),
> [ARGUS](https://github.com/jadoon200/argus) (all-source fusion),
> [PHAROS](https://github.com/jadoon200/pharos) (maritime) and
> [HORUS](https://github.com/jadoon200/horus) (air). Those four are detection systems:
> signal → anomaly → incident → human review. DELPHI is a decision system: forecast →
> uncertainty → cost-weighted choice → plan → measured consequence. The evaluation problem
> differs in kind, because you have to reason about an action you did not take.

DELPHI set out to show that a calibrated forecast beats conventional autoscaling. Measured
against the recommender Kubernetes actually ships, it mostly does not. The useful result is
the boundary: forecasting pays when capacity is committed for hours at a time *and* demand
has real daily structure, and a single number you can compute in seconds tells you which
side of that boundary you are on.

## The three findings
## The four findings

1. **In the autoscaling regime, forecasting loses to a trailing percentile.** Demand is
~0.98 autocorrelated at one minute on every workload measured here, so recent load
Expand Down
1 change: 0 additions & 1 deletion docs/ROADMAP.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,6 @@ marketing document.
| Milestone | Decision | Reason |
|---|---|---|
| M9–M13 — specialist agents | **Cut** | The pre-registered cut order chose the GPU lane over the agent layer when the two competed for the same days. The agent layer's own pre-registered expectation (Q5) was that it would *not* improve decision quality, only auditability — so cutting it removes a likely-null result rather than a likely finding. The decision ledger it would have written is not claimed anywhere in the product. |

| M18 — Alibaba spot-GPU | **Cut** | No licence file at the repository root; the zero-cost audit will not redistribute or depend on unclearly licensed data. |
| M19 — carbon-aware deferral | **Cut** | Elegant, not load-bearing. First on the pre-registered cut list. |
| M24 — live Wikimedia lane | **Cut** | Replay carries the argument; a live arrival process is presentation, not evidence. |
Expand Down
Loading