neiam
/apollos_crib
It's Apollo's Crib
public · Issues · Pulls 1 · Labels · Forks · Compare · Actions failure · Badges · Packages
⭐
Log in to mark this repository.
Scope the Plani
5856372 · Gabriel Morell · 2026-09-03 16:36
Message
{commit_body(@commit)}
Files changed
added
PLANI.md
+116
−0
@@ -0,0 +1,116 @@
| 1 | +# Plani | |
| 2 | + | |
| 3 | +A foci that moves. | |
| 4 | + | |
| 5 | +A [Foci](README.md) anchors a query in one place. A Plani anchors it wherever | |
| 6 | +the client reporting to it currently is, and falls back to a home foci when | |
| 7 | +nothing has reported for a while. The point is a vision that answers "what is | |
| 8 | +near me" rather than "what is near Davis Square". | |
| 9 | + | |
| 10 | +This is a scope, not a description of something that exists. What is built so | |
| 11 | +far is the plumbing underneath it, noted below. | |
| 12 | + | |
| 13 | +## Where the position lives | |
| 14 | + | |
| 15 | +**In a GenServer, and nowhere else.** A Plani's row in the database holds its | |
| 16 | +configuration — a name, a home foci, the sources attached to it — and never a | |
| 17 | +coordinate. The current position exists in the worker's memory, expires on a | |
| 18 | +timer, and goes when the process does. | |
| 19 | + | |
| 20 | +That is a deliberate constraint rather than an implementation detail. A table | |
| 21 | +of where somebody has been is a different kind of thing from a board of when | |
| 22 | +the next bus is, with different obligations attached to it, and this avoids | |
| 23 | +having one. It also makes the fallback obvious: when there is no position in | |
| 24 | +memory there is nothing to fall back *from*, so the home foci answers. | |
| 25 | + | |
| 26 | +``` | |
| 27 | + fresh fix? ──yes──▶ the client's position | |
| 28 | + Plani.anchor() ──┤ | |
| 29 | + └──no───▶ home foci's place (from the database) | |
| 30 | +``` | |
| 31 | + | |
| 32 | +## What already exists | |
| 33 | + | |
| 34 | +More than it looks, because the pieces were built for other reasons. | |
| 35 | + | |
| 36 | +| Piece | Where | State | | |
| 37 | +| --- | --- | --- | | |
| 38 | +| A client may publish to its Ankyra | `RoomHermesWeb.TopicController`, `<topic>.up.` | done | | |
| 39 | +| A client reports its position, opt-in | `clients/android`, `LocationReporter` | done | | |
| 40 | +| Positions arrive and are held for five minutes | `RoomSanctum.Worker.Ankyra` | done | | |
| 41 | +| The trail is visible | `cfg/ankyra/:id` | done | | |
| 42 | +| A client can ask for a fresh board | `<topic>.publish.`, `Pythiae.query_current_now/1` | done | | |
| 43 | + | |
| 44 | +And one piece of shape that matters more than any of it: **the storage layer | |
| 45 | +already separates resolving an anchor from searching near a point.** | |
| 46 | + | |
| 47 | +```elixir | |
| 48 | +def nearest_aqi_stations(source_id, foci_id, limit) # resolves a foci, then: | |
| 49 | +def nearby_aqi_stations(source_id, %Geo.Point{}, limit) | |
| 50 | +``` | |
| 51 | + | |
| 52 | +A Plani is largely a matter of handing a different point to functions that | |
| 53 | +already take one. | |
| 54 | + | |
| 55 | +## The part that is not uniform | |
| 56 | + | |
| 57 | +"Most of our datapoints have a latlon attached someplace" is true, but they do | |
| 58 | +not have it in the same place, and — more importantly — **"the closest N" only | |
| 59 | +means something for some of them.** Attaching a source to a Plani means one of | |
| 60 | +three different things: | |
| 61 | + | |
| 62 | +**Sources that store many located things.** Nearest-N is the natural question, | |
| 63 | +and the work is a spatial query. | |
| 64 | + | |
| 65 | +| Source | Where the location is | Ready? | | |
| 66 | +| --- | --- | --- | | |
| 67 | +| `gbfs` stations | `gbfs_station_information.place`, PostGIS | yes — `ST_DWithin` + `<->` ordering already used | | |
| 68 | +| `gbfs` free bikes | `gbfs_free_bike_status`, PostGIS | yes | | |
| 69 | +| `aqi` monitors | `airnow_hourly_observations.point`, PostGIS | yes — `nearby_aqi_stations/3` | | |
| 70 | +| `gtfs` stops | `stops.stop_lat` / `stop_lon`, **plain floats** | **no** — no geometry column, no index, no nearest-stop query anywhere | | |
| 71 | + | |
| 72 | +GTFS is the one anybody would want first and the only one needing real work: a | |
| 73 | +geometry column on `stops`, a GiST index, and backfilling it on import. It is | |
| 74 | +also the largest table in the system, so the index is not optional. | |
| 75 | + | |
| 76 | +**Sources answered *at* a point rather than *near* one.** Weather, sunrise, | |
| 77 | +pollen, the tide at the nearest station: there is one answer and it is | |
| 78 | +computed for wherever you are. These already take a `foci_id` in their query | |
| 79 | +config (`weather`, `ephem`, `pollen`, `calendar`, `cronos`, `icarus`), so the | |
| 80 | +change is to accept a point instead — the smallest useful stage, and it needs | |
| 81 | +no spatial indexing at all. | |
| 82 | + | |
| 83 | +**Sources with no location.** `github`, `gitlab`, `packages`, `mailbox`, | |
| 84 | +`treasury`, `bourse`. A Plani should either refuse these or pass them through | |
| 85 | +untouched; "the nearest CI job" is not a thing. Worth deciding explicitly | |
| 86 | +rather than discovering at runtime. | |
| 87 | + | |
| 88 | +## Suggested order | |
| 89 | + | |
| 90 | +1. **The anchor.** A Plani row (name, home foci, sources) and a worker that | |
| 91 | + resolves a position: freshest fix from the Ankyra worker's trail, else the | |
| 92 | + home foci. Publishes nothing yet — but the resolution, the expiry and the | |
| 93 | + fallback are the part with the design risk in it, and it can be watched on | |
| 94 | + a page before anything depends on it. | |
| 95 | +2. **Relocatable point sources.** Give the query workers that take a `foci_id` | |
| 96 | + a way to take a `%Geo.Point{}`. Weather and ephem where you are. No schema | |
| 97 | + changes. | |
| 98 | +3. **Nearest-N for what is already indexed.** GBFS stations and bikes, AQI | |
| 99 | + monitors. The queries exist; they need a limit and a caller. | |
| 100 | +4. **GTFS stops.** Geometry column, GiST index, backfill on import, then | |
| 101 | + nearest-stop. The biggest piece and the one worth having. | |
| 102 | + | |
| 103 | +## Open questions | |
| 104 | + | |
| 105 | +- **N or a radius?** "Five nearest" and "everything within 500m" answer | |
| 106 | + differently when you are somewhere empty. GBFS queries already take a | |
| 107 | + radius; nearest-N would be a second mode rather than a replacement. | |
| 108 | +- **Is a Plani its own thing, or a mode on a Pythiae?** A Pythiae already has | |
| 109 | + `curr_foci`. A Plani could be a foci whose place is resolved at read time, | |
| 110 | + which would mean nothing else has to know about it — the smaller change, and | |
| 111 | + probably the right one. | |
| 112 | +- **What does a client see while it has never reported?** The home foci, which | |
| 113 | + is correct but silent. A board that says which anchor it used would save | |
| 114 | + somebody wondering why the times are for the wrong town. | |
| 115 | +- **Which Ankyra?** A Plani follows *a* client. An Ankyra with two clients | |
| 116 | + reporting has two positions and no rule for choosing between them. |
Parents: 721d2fc