How The Forge Is Checked

Every calculator says it's accurate. Here's ours, with the test attached.

I got tired of arrow tools that tell you a number and never tell you where the number came from. So this page does the unglamorous thing: it names the tests the Forge has to pass, publishes the score, and lists the places we're still guessing. If you want to check my work, everything below is reproducible — the validation harness ships in the repo and prints these numbers on demand.

Short version: the Forge is checked against the spine charts you already trust, and improved only where they're wrong or silent. Agreement is the baseline, not the ceiling.

What "98% Easton parity" actually means

Parity is a specific test, not a mood. Here is the whole thing.

We took Easton's published Hunting Arrow Size Selection chart 301055-A (2023) and transcribed the grid out of the PDF text layer: poundage bands down the side, cut length 23″–32″ across the top. That is 129 cells. Columns 33″ and 34″ were lost in extraction, so they are excluded rather than pattern-guessed. One cell (23″, 85–90 lb) was unreadable and is null.

For every cell, the harness feeds the engine the base case the chart itself assumes — 100 gr point, insert under 25 gr, release aid, 301–340 fps bow, 85% let-off — and compares our recommended spine window against the chart's own cell interval.

• A match is an overlap: our window and Easton's cell share at least one spine value. If Easton says 400–350 and we say 340–400, that's a match, because a shooter following either one lands on a shaft the other accepts.
• A miss is no overlap at all — we demand something stiffer than anything the chart allows, or weaker than anything it allows. There is no partial credit.
• A red verdict is the harsher second test: take every stamped spine the chart accepts for that cell, run it through our verdict logic, and ask whether all of them would light up red in the app. Red means we'd be telling a customer that the shaft the manufacturer's own chart sent him to is wrong. Our verdict tolerance is ±25 spine units outside the recommended window before anything turns red.

Scores, current engine, run 2026-09-10 (scripts/spine-chart-validation.mjs):

The harness runs on every push. If a change to the spine engine drops those numbers, the build fails and the change doesn't ship. That is the entire point of publishing them: a number you can't lose isn't a check.

The two charts we validate against — and the third we treat as weaker evidence

Easton is the primary, because it's a first-party PDF with a readable text layer and the widest poundage/length grid published by anyone. Victory is the second, and it's the better source in one specific way: their charts are copied verbatim out of the JavaScript behind their own selector tool, so we're testing against Victory's live logic rather than someone's photograph of a poster. Victory publishes two grids — chart 1 for the light-insert shafts, chart 2 for the 50/60 gr TKO family — and we score them separately, because merging them would hide exactly the disagreement that matters.

Gold Tip is in the harness at 85%, and I'd ask you to read that number with a hand on it. That grid came off an image by OCR with four cells independently confirmed against published worked examples. Medium confidence. It is evidence, not proof, and we label it that way rather than rounding it up into the headline.

Where we disagree with the charts, on purpose

The two Easton misses are both long-arrow, mid-poundage cells (52–56 lb at 31″, 47–51 lb at 32″) where Easton's chart calls for 300–250 and we recommend 340–400 — we are the weaker of the two there. The Victory misses run the other way: at long draws and light poundage their chart stays on 500 where our model has already moved to 340–400.

And one honest oddity: Victory's chart 2 reads 300 in the 47–51 lb row at 27″, which breaks monotonicity in both directions and is almost certainly a typo for 400. We test both the as-published grid and the corrected one, and we score ourselves against the as-published number. Grading yourself against your own correction of somebody's chart is not parity.

Why rear weight stiffens dynamic spine

Nearly every free calculator in this space either ignores nock and vane weight or treats it as more mass out front. Both are wrong, and the second one is wrong in a way that will send you a shaft group too weak.

Think about what the arrow actually does off the string. The string shoves the back of the shaft forward; the mass out front resists, and the shaft buckles sideways around its middle — that's archer's paradox, and it's plain beam physics. Now the two ends do opposite jobs. Weight at the point is the far end of the lever — it's what the shaft is bending against, so adding grains there increases the bend and the arrow behaves weaker. Weight at the nock is at the pushed end, riding with the force rather than resisting it. Heavier back end means more of the arrow's mass is being accelerated directly by the string instead of levered against, the column loads less, and the shaft flexes less. It behaves stiffer.

In the Forge that's a coefficient, not a hand-wave: front mass runs at 1.25 spine units per grain off a 100 gr reference, rear mass at 0.60 per grain off a 12 gr reference, applied with the opposite sign — roughly half the magnitude, opposite direction. Both scale with the square of the flexing length, because a longer beam amplifies everything you hang on either end.

Two more behaviours most tools miss:

• Front mass saturates. Going 100 → 125 gr up front is a real move. Going 200 → 225 gr barely registers. Our model is linear to +100 gr of front delta — which keeps the validated 75–150 gr band exact — and halves the slope beyond it. A purely linear model blows up at extreme FOC and reports numbers no shaft has ever produced.
• Internal footing stiffens by length, not just by weight. A footed or CTI'd shaft has a shorter flexing section, which is very nearly the same thing as trimming the arrow without trimming it. We subtract footing length from the working length before any of the weight math runs — front footing, back footing, or both. Calculators that treat a footing as "an insert that weighs more" get this backwards at both ends.

I'll say the uncomfortable part too: we had the rear sign backwards ourselves until 2026-06-25, which is the same error Archer's Advantage ships today. It was found by grilling the model against Stu Miller, OnTarget2 and the tuner's rule of thumb until it broke. It's fixed, it's commented in the source, and the comment names the mistake.

Why the spine check ignores your chronograph

People ask about this more than anything else on the tool: I gave you my real speed, why isn't the spine answer using it?

Because arrow manufacturers publish their charts against the IBO frame. Easton's grid assumes a 301–340 fps bow. Victory's assumes theirs. Those charts are the thing we're being checked against and the thing your shaft was stamped for — so the spine recommendation is computed in the same frame the chart was written in. Feed your real, slower, accessory-loaded chrono number into a chart-frame calculation and you don't get a more accurate answer, you get an answer that no longer means what the chart means. It's a unit mismatch dressed up as precision.

So the spine check runs on IBO. On purpose. And it says so on the card.

What your chronograph is used for

• Sight tapes. A tape is only as honest as the speed it was drawn from — an estimated number is out at 60 yards no matter what printed it. Tapes use your measured speed, always.
• Kinetic energy and momentum. Real speed, real mass, at the bow and downrange.
• Trajectory, drop and wind drift in the Shot Solver.
• The rig's service record — chrono readings logged over time, so a bow losing 6 fps across a season shows up as a line instead of a feeling.

Per-bow chrono offsets, with the sample size showing

An IBO number is a factory claim on a bare bow. Real bows come in under it, and by different amounts. We carry a table of 39 bow models with a measured offset — the mean of (measured fps − IBO-formula prediction) — compiled from Podium Archery's published chrono tests.

Now the part a marketing page would leave out. 21 of those 39 rows are a single chrono test — n=1, standard deviation zero because there is nothing to deviate from. The app prints the count beside the number ("1 chrono test") so you can weigh it yourself. Four rows sit at exactly 0.0 offset from one test, which most likely means a bench-condition test rather than a bow that truly delivers its full rated IBO. Treat a one-test offset as a hint and a thirteen-test offset as data. That is what the n column is for, and it's why we show it.

The sight database

178 sight entries — Ultraview, Spot Hogg, Black Gold, HHA, Axcel, IQ, Sword, Trophy Ridge and more — each carrying its pin diameters and its sight-radius range, taken from current manufacturer pages. That's the geometry a tape generator actually needs: get the radius wrong and the tape is the wrong length; get the housing wrong and it doesn't fit the sight you own. The list is frozen by a test that fails if an entry moves, so a tape you printed last season still means what it meant.

Energy floors, and the tag on every one

We publish our own energy floors instead of quoting a number nobody can source, and every floor carries a tag saying how strong its basis is:

• charted — the number is Easton's, from their game-class KE table, and the species is named in that row. Deer/antelope 25 ft-lb. Elk/black bear/boar 42. Big game 65.
• derived — our number, placed between charted anchors on body mass and anatomy. Heavy game (moose class) at 50 ft-lb is the example: above elk on mass, nowhere near cape buffalo on armour.
• judged — our number where the evidence is genuinely thin, published with our name on it and no one else's. Small game at 15 ft-lb is judged.

Across 61 species the split is 10 charted, 39 derived, 12 judged. Two things we say every time we show that chart: it is retired — Easton no longer publishes it, and their current KE page gives no thresholds at all — and it is an at-the-bow chart, taken feet from the riser, so comparing a downrange, derated energy figure against it invents a standard a third stricter than anything ever published. The code enforces that with separate types for at-bow and impact energy; the compiler rejects the mistake.

And momentum is displayed as information only. It can never turn a verdict red, because no per-species momentum minimum is published anywhere — the tables circulating online are Easton's KE bands converted with an unstated arrow weight, which is circular and adds nothing.

How that compares

Only cells I can cite are filled. "Not found" means I looked at the published product and could not find it — not that it doesn't exist.

Where we're still guessing

If this page only had good news it would be an ad. Current open items, straight off our own audits:

• Broadside vital widths are not sourced. Only two of 61 species have an independent width (wild turkey, Nile crocodile). The other 59 are vital height multiplied by a class ratio of 1.15–1.3. Those ratios are ours; nobody publishes a per-species broadside heart-lung width, and we couldn't find a primary source for any ratio. It moves the first-shot hit-probability number, so it matters, and swapping one unsourced ratio for another would be motion rather than progress. The fix we intend is rebuilding it off per-species chest through-width, which is derived from published shoulder heights.
• 16 shaft rows have an unresolved diameter. Six are Altra CENTRUM entries whose stated inside diameter is larger than their outside diameter, which is physically impossible and means one of the two figures is mislabelled at the source. They are listed by name in the validation script, and the list is only allowed to shrink. We will not invent an OD to make a check pass — that is exactly the mistake that produced customer-visible wrong diameters once already.
• Some makers publish no straightness tolerance, so we print none. K2 Arrow Co is the current example: their published pages give spine tolerance and per-spine OD but no straightness figure, so that field stays empty rather than getting a plausible ±.003″ that nobody stands behind.
• Altra publishes no outside diameter for its entire line. Those entries render "OD not published". Archers shop by inside diameter anyway — that's how shafts are named — so we lead with ID and label both.
• Gold Tip parity is OCR-based, medium confidence, and stays labelled that way until we can source the grid first-party.
• Recurve, traditional and youth (500+) spines are extrapolated. The dynamic-spine model is validated for compound, 250–500 spine, 26–31″ arrows, 75–150 gr points. Outside that band we're outside our anchors and the tool says so.

If you find something wrong in here, tell me and I'll re-run the check: sal@sparrowexpeditions.com. A human reads it, and a correction with a source attached will get a version number.

Open the Forge — free, no install, phone or desktop.