Skip to content

Latest commit

 

History

History
108 lines (82 loc) · 7.36 KB

ROADMAP.md

File metadata and controls

108 lines (82 loc) · 7.36 KB

Roadmap

SentinelOps is delivered in five milestones that follow the weekly schedule in our project proposal (submitted November 6, 2026). Each task below is delivered as a single reviewed commit whose message ends with Refs: <task id> and closes the task's GitHub issue. Tasks are also tracked as issues grouped into milestones. Task IDs are prefixed by owner: M = Mohamed Azab, Y = Yassin Koubeisy.

Milestone Window Status
M1 — Simulator and baseline Nov 9 – Nov 15 🟢 Complete
M2 — Learned models Nov 16 – Nov 22 🟡 In progress (6/8)
M3 — Decision-making agent Nov 23 – Nov 29 ⚪ Planned
M4 — Evaluation and demo features Nov 30 – Dec 4 ⚪ Planned
M5 — Release Dec 5 – Dec 11 ⚪ Planned

Key dates: feature freeze Dec 4 · live demonstration Dec 7 – 11 · code and report submission Dec 11.


M1 — Simulator and baseline

Exit criterion: a seeded crash scenario can be run end-to-end against the rule-based baseline from the command line, reporting the share of failed legitimate queries, with identical results for identical seeds.

  • M1.1 Project scaffold, tooling and continuous integration (#1) — Mohamed · Nov 9
  • Y1.1 Architecture overview and interface contracts (#2) — Yassin · Nov 10
  • M1.2 Topology, routing preferences and traffic model (#3) — Mohamed · Nov 12
  • Y1.2 Fault model for servers, network paths and traffic (#4) — Yassin · Nov 12
  • M1.3 Simulation engine with bounded queues and failure accounting (#5) — Mohamed · Nov 15
  • Y1.3 Seeded scenario families (#6) — Yassin · Nov 14
  • M1.4 Run harness and legitimate-query failure metric (#7) — Mohamed · Nov 16
  • Y1.4 Threshold-rule baseline agent with runbooks (#8) — Yassin · Nov 16

M2 — Learned models

Exit criterion: detector precision/recall and diagnosis accuracy are reported on the tuning split; the console globe reflects live site health.

  • M2.1 Diagnostic probes and action durations (#9) — Mohamed · Nov 18
  • Y2.1 Telemetry window features (#10) — Yassin · Nov 17
  • M2.2 Labelled run generation and train/tune/test splits (#11) — Mohamed · Nov 19
  • Y2.2 Isolation Forest anomaly detector (#12) — Yassin · Nov 19
  • M2.3 Bayesian network structure and parameter fitting (#13) — Mohamed · Nov 20
  • Y2.3 Alert threshold tuning and detection metrics (#14) — Yassin · Nov 21
  • M2.4 Bayesian inference, console API and live globe (#15) — Mohamed
  • Y2.4 Policy rule engine (#16) — Yassin

M3 — Decision-making agent

Exit criterion: a service crash is repaired and verified fully autonomously; a hardware fault is escalated, approved, repaired and returned to service gradually after verification.

  • M3.1 Probe selection by expected entropy reduction per second (#17) — Mohamed
  • Y3.1 Autonomy presets, approvals and maintenance holds (#18) — Yassin
  • M3.2 Recovery forecasting over candidate actions and fault hypotheses (#19) — Mohamed
  • Y3.2 Recovery verification checklist (#20) — Yassin
  • M3.3 Action ranking and incident prioritisation (#21) — Mohamed
  • Y3.3 Gradual traffic restoration with rollback (#22) — Yassin
  • M3.4 Agent orchestration and incident/approval console panels (#23) — Mohamed
  • Y3.4 Engineer handoff reports (#24) — Yassin

M4 — Evaluation and demo features

Exit criterion: a single command regenerates every table and figure in the report from held-out seeds.

  • M4.1 Parallel experiment runner and paired bootstrap confidence intervals (#25) — Mohamed
  • Y4.1 Stress scenarios: noise, missing readings and overlapping faults (#26) — Yassin
  • M4.2 Ablations: fixed probe order and site-local forecasting (#27) — Mohamed
  • Y4.2 Autonomy-preset comparison experiment (#28) — Yassin
  • M4.3 Report figures (#29) — Mohamed
  • Y4.3 Incident report data (before/after series) (#30) — Yassin
  • M4.4 Chaos injection, replay and evaluation views in the console (#31) — Mohamed
  • Y4.4 Report draft: background and related work (#32) — Yassin

M5 — Release

Exit criterion: a fresh clone runs the demo and experiments with one command each; the report is submitted.

  • M5.1 Documentation and quickstart (#33) — Mohamed
  • Y5.1 Fresh-clone reproducibility check (#34) — Yassin
  • M5.2 Report methods and results; demo preparation (#35) — Mohamed
  • Y5.2 Report introduction, limitations and contributions; demo preparation (#36) — Yassin

Scope changes

Changes to the plan above are recorded here with the reason.

Date Change Reason
Nov 16 M1 closes on Nov 16 instead of Nov 15; M1.4 and Y1.4 move to Nov 16 M1.3 was split into three commits (types, site queue, engine) to keep each change reviewable; Y1.4 needs the engine from the last of them

Backlog

Stretch goals, taken on only if the milestones above are on schedule.

  • Spare-capacity action (temporarily add a server at a site)
  • First-in-first-out versus priority incident scheduling experiment
  • Belief timeline in the incident replay view