prism0x2A · scan excerpt

prism0x2A · scan excerpt · frozen, not a live run

Three lenses, one file, every line cited.

A real excerpt from a real scan — not a mock-up and not the whole report. BLUE reads the shape, AMBER reads what the code does, GREEN reads whether it still adds up. Each answer names the file and the line it came from, and the sentences that could not be backed are shown struck out rather than quietly dropped.

Repository · pixel-peloton (our own)
Scanned · 5 Nov 2026 · tier skeleton
Files read · 1,429
Statements · 2,446 across 51 capabilities

Why our own repository.The critical view runs on code we own, so nothing here is a verdict on anyone else’s work.

Blue · the shape of it

What is this file, and what leans on it?

BLUE never opens the prose. It reads the import graph and the syntax tree, and every figure keeps the file it was counted from.

lib/supabase/useEntitlement.ts · category: hook · 168 LOC

24depend on it
7it depends on
22complexity
3.4heat
0duplication
Ask BLUEWhat could break here?

Twenty-four modules resolve through this hook at runtime and it resolves seven itself, so a change to what it returns is felt in twenty-four places, not one. It is a hook, not a component: nothing renders from it, which is why the blast radius is import-shaped rather than visual.

from import graph · 1,429 files · lib/supabase/useEntitlement.ts

Every figure in this answer is in the panel above and can be recounted from the graph.

Across the repository · 1,429 files

12nothing imports
2,093lines in those
93clone groups
1,226repeated lines
1import cycle
The 12 files nothing imports
lib/storyMechanics.ts575 LOC
lib/hoehenTeile.ts488 LOC
lib/outrunGenerator.ts234 LOC
lib/zwoToStory.ts159 LOC
lib/trainingBlocks.ts154 LOC
lib/storyMood.ts117 LOC
…and 6 more366 LOC

None of these is a grade. prism says nothing imports those twelve files. It does not say they are waste — a module loaded by name at runtime, a plugin entry, a script a manifest calls, all look exactly like an orphan from the import graph, and prism cannot tell which is which. You can. So it hands over the list and stops. Likewise a repeated block may be the right amount of repetition, and one cycle between lib/leistungsStufe.ts and lib/vfx.ts is a measurement, not a verdict.

Amber · what it does

The same file, read for behaviour

AMBER makes one narrow statement at a time and attaches the lines it read. Here is one from that file, opened all the way up.

Reads from pp_entitlements through select.

lib/supabase/useEntitlement.ts · line 67 · capability peloton.cloud.sign-in-and-access

  64   if (!user) { setEntitlement(FREE); setCreatedAtMs(null); setAnker({ … }); … }
  65   setCreatedAtMs(user.created_at ? new Date(user.created_at).getTime() : null);
  66   const { data } = await supabase
  67 >     .from("pp_entitlements")  68     .select("status, premium_until, tier, trial_ends_at, trial_started_at, …")
  69     .eq("user_id", user.id)
  70     .maybeSingle();
1 · staticpassed
2 · semanticskipped
3 · contradictionnone
4 · counter-evidencenone
verified→ printed in the report

Check 1 does the work. It does not ask whether the sentence sounds right — it re-opens the file and requires the quoted text at the cited line, character for character. A citation that drifted by one line fails it. Check 2 says skipped because this excerpt ran with no language model; that is why anything needing judgement caps at inferredhere, and why nothing on this page rests on a model’s confidence.

Ask AMBERWhat does this file do?

It reads the rider’s entitlement row: from pp_entitlements, through a select — lib/supabase/useEntitlement.ts:67–68, where the selected columns are status, premium_until, tier, trial_ends_at, trial_started_at, strava_enabled and the three premium-minutes fields.

That is the only statement in front of me about what the file you asked about does. The capability it belongs to also reads pp_admins through a select (lib/supabase/useIsAdmin.ts:49), but I have been shown 2 of its 33 statements, so I cannot tell you what else this file does. Nothing in front of me says it writes.

2 statements· useEntitlement.ts:67–68 · useIsAdmin.ts:49, both re-checked against their lines

Written from those two statements and their lines, without an API key, the way the Claude Code path works. Every citation above was then put back through runStaticProofCheck, the check the verifier itself runs. 2 of 2 hold.

Green · does it still add up

Where the account does not close

GREEN reads across the layers rather than down a file: which capabilities carry statements that did not survive, and what that says about the gap between claim and code.

CapabilityVerifiedWithheldWhy
marketing.export.gif-worker11values past the quote in a minified file
The other 50 capabilities: nothing withheld.
Ask GREENWhich capabilities have drift?

One of fifty-one, and it is not drift in the code. The GIF worker holds one statement that verified and one that did not. The second names constants of a minified file, whose code sits on one line of 16,535 characters. The quote stops before three of the values, so prism withholds the sentence rather than print what it did not show.

from 2,446 statements · 51 capabilities · 1 with anything withheld

A gap here is not a defect report. It is the difference between what the repository says about itself and what its code can show — which is the one thing a reader cannot check by reading either alone.

Defines the constants 1 (“xend”), 12 (“BITS”), 5003 (“HSIZE”), 254(“a_count”) and 1 more constants.

public/gif.worker.js · line 2 · static check

rejected→ withheld12, 5003 and 254 appear in none of the quotes

This one is a false alarm we know about. Those values are in the file. The line is too long for the quote to reach them. We would rather withhold one true sentence than print one false one, but the number is one, not zero, and it belongs here rather than in a footnote.

The ladder · printed or not

The verdict decides whether the sentence exists

It is not a confidence percentage. It is the gate between a statement and your report.

VerdictCountIn the reportMeans
verified2,319yesSurvived all four checks.
inferred126yes, markedHeld, but one check could not run without a model.
unknown0noEvidence exists, but sits in code marked parked.
rejected1noThe evidence did not support the statement.
contradicted0noThe code says the opposite.

Where it stops

What this scan cannot tell you

  • No model ran in the scan. The semantic check is skipped throughout, which caps anything needing judgement at inferred. The one model-written passage on this page is the Ask answer above, and it is marked as such — nothing it says was allowed past the same check.
  • One known false alarm in a minified file, shown above rather than quietly dropped.
  • TypeScript and JavaScript are complete; other languages are partial. A repository prism cannot read properly is refused, not half-measured.
  • This is an excerpt, frozen at the date above — not a live run, and not the whole report.
$ npx -p prism0x2a prism scan .

Runs on your machine, against your files. Nothing is uploaded in order to score it — which is also why we can only show you our repository here, not yours.

Interactive

Ask your own question.

The same three lenses, but you pick the repository and the question, and forge turns the answer into something shippable. Five questions and three generations per address — a magic link, no password.

Open the sandbox →